Release Notes

Your Conversations Show Only What You Typed

When an AI coding tool loads a skill or runs a command, it adds its own text to the conversation. Grasppy no longer shows that text as if you had written it.

• Your turns are yours again. Skill instructions and command notes that Claude Code adds behind the scenes are no longer imported as User turns. Your real request stays, and the added text is left out.
• Everywhere conversations arrive. The fix applies to file imports, Auto Capture, chat sync, appending to an existing analysis and Create Analysis.
• Newest first in the Chat Thread. A new Sort menu shows the latest turn on top by default, so a turn that arrives while you watch appears where you are looking. Switch to oldest first at any time.

Recall: What Your AI Coding Sessions Decided, Fixed and Learned

Recall replaces Sources & Memory. It turns your analysed AI coding sessions into short notes about what was decided, fixed and learned, each linked to the exact turn it came from, and an analysis now builds its reports when you ask for them.

• One place for what happened. Recall shows your notes by day, a checkpoint for every session (what you asked, what was completed, what is next) and the original conversations. Open the exact turn behind any note, or show it in the Semantic Map. Approve the notes you trust, reject the ones you don’t, and mark notes that later work replaced.
• Summaries of whole stretches of work. Every 12 turns of a session are summarised together into a few notes that record where the work ended up, so a plan that was proposed and then built reads as built, not as pending.
• Analysis stays your choice. In Recall → Memory settings you turn automatic analysis on or off for a workspace and choose the date it starts from; older conversations are never analysed by accident. With it off, every captured conversation stays readable and searchable for free.
• Reports when you ask. Press Analyze to build the summaries, diagrams and relationships for the whole conversation at once, so they always agree with each other.
• Search that finds the right passage. Searching your captured conversations and project documents now finds the right passage far more often and returns in a fraction of a second. It covers conversations from every integration, including browser chats, and needs no setup.
• Your secrets stay hidden. API keys, tokens and passwords are replaced with a label on your computer before a conversation is uploaded, and checked again on our servers.

Custom GPT Chats Export and Save to GRASPPY Capture

Conversations inside a ChatGPT custom GPT or Project now export in full and save to GRASPPY Capture, and long captured conversations continue cleanly into their next part.

• Custom GPTs and Projects. Chats opened from a custom GPT or a ChatGPT Project were not recognised as conversations, so Save to GRASPPY Capture stopped with “Could not read this conversation” and the Markdown export held only the messages visible on the page. Both now read the whole conversation, the same way regular ChatGPT chats do.
• Long captures continue cleanly. When a captured conversation grows past the size of one part, the new messages now go into the next part. Before, an update could leave an empty extra part behind.
• Update the extension. The fix ships in the Grasppy browser extension version 2.9.2, which reaches Chrome through the Web Store update. No new permissions are requested.

Scout: Create Scene Animations With Local AI

Turn your finished narration into editable motion graphics with HyperFrames and your local AI coding agent. Create one scene at a time or prepare a sequence, then choose the finished clips for your video.

• Your script sets the context. Animation prompts include the full story and script, identify the scene to create, and carry its approved voiceover timing. Narration stays unchanged.
• Bring images or let AI create the graphics. Select the visuals you want to include, add creative directions, and copy the prompt into your local agent. Selected visuals are optional; the agent can build original typography, diagrams and animated objects.
• Keep working files on your computer. Connect a verified project folder to browse local visuals, preview files and save scene selections. Existing local resources can be reused without copying their folders into every new animation.
• Compare finished versions. Completed animations appear in Scout for preview. Create another version when you want a different approach, and add your chosen video to Assets for use in Collage.
• Clear working files when you choose. Preview generated files for one scene or the whole project before deleting them. Finished videos, audio, manually added files and saved prompts remain; deleted editable sources must be regenerated for future edits.

Scout Scenarios: Generate Video, And An Assets Page That Shows What Everything Costs

Scenes can now be built from generated video as well as generated stills, every asset for a scene lives on its own page with its cost and file size visible, and any picture you bring in can be turned into a whiteboard drawing.

• An AI video tab. Generate video the same way you generate images, with the model list filtered by what you are giving it — a prompt on its own, a still to animate, or reference footage. Duration, resolution, aspect ratio and audio are set per generation, and a video that is still rendering keeps working if you close the page and come back.
• A page for a scene’s assets. Choosing and arranging are now separate jobs. Everything a scene uses sits on one page you can search, showing where each asset is used, what it cost to generate and how large the file is — so you can see what you are spending before you spend more.
• Turn any picture into a whiteboard drawing. Prep to Whiteboard is now available on imported images, not only generated ones. The conversion runs on the server and costs nothing extra.
• Transparent copies, made for you. Every image you upload gets a transparent version created automatically, with the original kept untouched. A toggle on each placement switches between them, so a picture can sit over a background in one scene and keep its own backdrop in another.
• Preview matches the finished video. Effect timing now comes from one shared source, so what plays on the canvas is what renders — rather than two implementations drifting apart.
• Shorter route to a storyline. The story generator takes a couple of sentences instead of a filled-in brief, and pasting in your documentation is optional. Scenes can be reworked one at a time, with the new version shown next to the old one before anything is replaced.

Scout Scenarios: Point At Things, And Fix Timing Without Re-recording

You can now annotate a scene the way you would on a whiteboard — bubbles, arrows, notes, highlights and a moving cursor that clicks — and the timing edits that used to mean recording again now leave your original footage and voiceover untouched.

• A Callouts panel. Four speech bubbles, four arrows, five notes, five highlights, four cursors and plain text, each in its own gallery. Drag one onto the stage, type into it, and set text and outline colour, fill or no fill, line thickness and opacity — or reset to the preset. Callouts sit above your pictures and video, and you can restyle one in place without starting again.
• A cursor that moves and clicks. Give a pointer a start and a destination and it holds, travels, then holds again, with the timeline showing when it leaves and arrives. It can click once on arrival, or at a moment you choose if it is standing still, with a ripple, double ripple, burst or soft pulse in a colour you pick. Arrows can be straight, curved, elbowed or double-headed, with both ends editable.
• Effects on every callout. Right-click one to choose how it arrives — drawn, whiteboard reveal, fade or pop — what it does while it is up, including float, wobble, circled, underlined or arrowed, and how it leaves, by fading, being erased or shrinking away. Effects show in the finished video rather than the editing canvas, so the stage stays readable while you work.
• Pauses that can be taken back. Hold the scene at any point for as long as you need. The pause lives in the output timeline only — your media files are not altered, so removing it puts everything back exactly where it was.
• Split a clip where the playhead is. One clip becomes two, each still playing the right part of the original file, so you can move a segment or drop something between the halves without the footage jumping.
• Re-align the words in a voiceover. If the word timings drift out of step with the recording, realignment fixes them against the audio you already have, keeping your sentence boundaries and speeds. If too little of the recording matches, it stops and leaves your existing timing alone rather than guessing.
• Characters can speak from video. Assign an uploaded video to either character track and approve it for that scene. If you later swap the video, approval is asked for again, so a scene never renders with a character you did not sign off.
• Callout suggestions, when you want them. Ask for up to four suggested annotations for a scene and they arrive as editable elements rather than finished work — nothing is saved and no video starts until you say so. Suggestions use your provider credits, and the screen says so before you ask.

Scout Scenarios: Save A Layout Once, Reuse It Everywhere

A long video is the same arrangement repeated dozens of times. You can now build that arrangement once and drop it onto any scene, and several things that used to need a second pass by hand now happen on their own.

• Collage templates. Arrange a scene the way you want it, save it under a name, and apply it to any other scene of any video. Positions, sizes, rotations, layers, timings and effects all come across, and every picture or clip the template needs is added to that scene's assets for you. Applying replaces what is on the stage and asks first.
• Green backdrops remove themselves. Record a speaker on a green screen, upload the clip, and the green is measured at upload and dropped when the scene is generated. Only green qualifies: a blue, purple or teal backdrop is left exactly as you shot it, and a tick on the asset lets you switch the removal off.
• One backdrop for the whole video. A new Backdrop setting sets the paper every scene is drawn on, from white through cream, butter, sand, sky, mint, blush, lilac and stone. Choosing one also stops the background drifting from sentence to sentence, so the colour you pick is the colour you get.
• A Composition tab. Choosing assets and arranging them are now two visible steps on their own tab, one row per scene, each with a tick when it is done. At a glance you can see which scenes are ready and where the work stopped, instead of opening every clip to find out.
• Videos have thumbnails. Uploaded clips used to show a grey box in the asset library while the whole file downloaded, and often never appeared at all. Each one now carries a still, so a page of clips loads immediately and you can see what you are picking.
• Smaller, clearer settings. The background music track now lives with its own volume and fade controls, with the file's name and a listen button. Imported voiceover files show their name rather than a code. Controls that only steered automatic illustration have been put away while the hand-built approach is the one in use.

Repo Radar: Written For Every Reader, And Readable On A Phone

Every edition in the archive has been rewritten so it reads for any developer rather than for the person who commissioned it, and the daily email has been rebuilt for the screen most people open it on. No fact changed anywhere: same repositories, same versions, same licences, same star counts, same dates.

• Every edition now reads for you. The archive had grown up addressing one reader, naming the products and the setup it was written for. All twenty-six editions and every repository page have been rewritten in a voice that assumes nothing about who is reading. Only the wording moved — the research, the numbers and the verdicts are untouched, and the three hundred and fifty pages were checked line by line afterwards.
• The email is readable on a phone. The message was being laid out at desktop width and then shrunk by a third to fit a phone screen, so sixteen-point text arrived at about ten. It now lays out at the width of the device you are holding. Nothing below fourteen points appears anywhere, including the small print under each repository — the stars, the licence and the version.
• The full summary, not the first line of it. Each repository in the email carried a single sentence cut off at a fixed length, often mid-word. It now carries the whole summary the edition wrote — what the project is, and who it is good for — in full, exactly as written.
• Corrections can be published. An edition that is already out can now be corrected and republished, with every previous version kept. Until now a published edition was final, which meant a mistake stayed up.
• The edition goes out even when we are working on it. Maintenance and publication used to share one daily allowance, so a morning spent correcting the archive could stop that day's edition being delivered. They are now separate, and an automatic check every hour confirms that every edition the archive lists can actually be opened — rebuilding any that cannot, without anyone asking.

Scout Scenarios: Approve The Voice, Arrange The Collage, Add Effects

Every scene now goes through three steps you control before a clip is made: approve the voiceover, arrange each picture and video on the Collage stage with its own seconds, and give any element its own effects. The AI keeps only what is tedious to do by hand.

• Voiceover approval. Each scene has a voiceover page: pick a voice or record your own, split the lines between two characters, set speed and delivery, listen, and approve. A script edit marks the voice out of date until you approve again. With a Speechify account, the voice list now holds 125 English voices grouped by accent and gender, with favourites.
• Voice directions in the script. Pauses, emotions and delivery cues such as slow, fast, soft or loud go straight into the script as tags, with a helper panel and presets in the scene editor.
• The Collage. A stage between Assets and Generate. Drag your drawings, pictures, text and videos where you want them, size and rotate them, and set when each appears and leaves on a timeline that shows the spoken words. Playback uses the approved voice as the clock. The AI adds backdrops, camera moves and the presenter's reactions around your layout, plus callouts and labels when you allow them.
• Effects per element. Right-click any element for Effects: how it enters (drawn by hand, fade, pop, cut), its look, what it does on stage (float, wobble, circled, underlined, arrow) and how it exits (fade, erased by a rubber, shrink). Notes to the AI per element are there too.
• Videos in the clip. GIFs you select are saved as videos in your library, and they, along with videos you upload, play inside the finished clip in their boxes and windows, underneath the presenter and captions.
• Cleaner drawings and faster clips. Library drawings render with the artist's own lines again, without the thick double edges seen recently. A Collage clip now generates in about two minutes, and your clip settings are remembered per scene.

Claude Exports Capture the Full Conversation, and Auto Capture Keeps Itself Current

Long Claude chats now export from the first message to the last, and the Auto Capture list refreshes on its own instead of waiting for a click.

• Full Claude conversations. Claude recently began showing only the most recent part of a long chat, and exports started partway through. The extension now brings the whole conversation into view before reading it, so the export begins at your first message.
• Honest completeness check for Claude. If an export still cannot start at the first message, the file carries a Warning line at the top instead of passing as complete.
• Auto Capture syncs every hour. Conversations saved by GRASPPY Capture now reach the Auto Capture list on a schedule as well as on Refresh, and a large backlog is no longer cut off partway through. While the GRASPPY Capture app is open it keeps ownership of the archive, and the scheduled sync waits until it is closed.
• Update the extension. Reload the Grasppy browser extension (version 2.9.1) to get the Claude fix. No new permissions are requested.

Scout Scenarios: Write The Script, Direct The Story, Choose Every Picture

A scenario now goes from a one-line brief to a directed film inside Scout: the script is written for you, the storyline is a shot-by-shot treatment, and an Assets step lets you pick exactly what appears on screen, beat by beat, before any clip is made.

• The script is written in Scout. New scenario asks who the video is for, its purpose, one voice or two people talking, and a brief: paste text, add web addresses or pick items from your Library. One press writes the scenes. The same Script button then reworks the whole script or one scene from a sentence of direction, keeps every previous version, and can export the scenario as Markdown or import one back.
• Scene 00, About this video. Every scenario opens with a notes card that is never a clip: goals, references, things to remember, editable any time and collapsible.
• A director’s treatment instead of a summary. The storyline now sets one visual language for the film, the cast with their looks and gestures, the world and its props, and for each scene the picture, two to four shots with timing and camera, on-screen text, sound and the transition. Write it in Scout, or copy the full prompt to any AI and import the result. A Remove button clears it when you want a clean start.
• References step. Look, Characters, Product, Motion and Sound slots hold the images and clips that define the style; the Sound slot supplies the background music.
• Assets, per clip and for the whole film. Each clip slot and a new 00 Film props slot open an Assets pop-up. Generate groups splits the narration into beats; each group gathers options from the drawing library, Giphy and Klipy with one press, as many times as you like. What you switch on is what the clip uses: one picture per beat, replaced by the next as the words move on, or up to three side by side. Groups can be renamed, re-worded, folded and rearranged, and every picture opens in a large preview.
• Looks and placement. Every picture can be drawn in ink, in one of your three brand colours, in a three-colour mix, or in its own colours, and placed left, right, in a corner or full frame. No file is changed: the look is applied when the clip is drawn, from the palette in Settings.
• AI images in the pop-up. Generate a picture for a clip with your choice of image model and style, or from the images in your Look references. Backgrounds are cut out automatically, with a second pass when a model paints a white backdrop, and a Cut out background button for older pictures.
• GIFs from Giphy and Klipy. Save a key for either in Scout settings under GIF search. Searches bring fifty results per call with Show more; nothing is downloaded or stored beyond a link.
• Your own voice, as recorded. Clip narration recordings now take the raw microphone signal without phone-call processing, and the recorder shows which microphone it is using.
• Models with their prices. The planner and script writer pickers list OpenAI, Google, xAI and Anthropic models with each one’s list price per million tokens, and settings show a cutaway rhythm of one picture every one and a half, two or three seconds.
• Outline housekeeping. Click scenes in the outline to select several, then move them together or delete them in one confirmed step.

Any GitHub Repository In Scout, With The Repo Radar Card

Repo Radar shows the twelve repositories we choose each day. Scout now lets you add any repository you choose and get the same record: checked facts, pictures, and a plain-English review, on the same card. What you add stays in your own workspace and never appears on the public radar.

• Alternative tools on every website Tool report. A new section names three to five products that do the same job, opens each website to check it is real, and shows them as cards with an “Add to Scout” link. The cost of the search is printed under the cards; on a typical report it is a fraction of a cent. Section 10, Technical and delivery signals, now always shows the same six rows: delivery model, platforms, API, data ownership, reliability claims and release activity.
• Two models of their own. Scout’s AI settings gain “GitHub repositories” and “Tools — alternatives search”, separate from website Tool reports, so each can run on a fast model.
• Paste a GitHub address into Scout. Within about a minute the tool page shows the repository’s stars, licence, latest release, last code change and language, read from GitHub that moment, plus the owner’s avatar, GitHub’s preview card and screenshots from the README. Pictures open in a lightbox.
• The review is written as part of the import. Every repository gets the same plain-English review Repo Radar publishes, without a second click: In simple English, what it is, what it is good for, good things, bad things, how to install it, and two effort meters. It follows the same writing rules and uses your own configured model.
• The README, read for you. The opening paragraph, the Features list, use cases, requirements, install, usage, supported formats and limitations come straight from the README in the author’s own words, tables included, each folded to a few rows with “View more”. Video links in the README become tiles that play inside Scout.
• Similar repositories. The analysis names up to three repositories that do the same job, each checked against GitHub before it is shown, with an “Add to Scout” link. Repo Radar’s own picks in the same area are listed underneath, and Radar’s daily editions carry the same cards.
• Filed and tagged on arrival. The analysis places every new tool, website or GitHub repository, straight into your group, category and subcategory, creating a missing subcategory under existing parents when needed, and gives it up to five tags. There is no approval step; move or retag anything from Edit tool or the tags window.
• One tags pop-up everywhere. Tools, GitHub repositories, YouTube channels and videos, website reports and documents all show their tags in the same place on the page, and “Add tag” opens one small window that edits only the tags. Scout proposes up to five from the analysis; you add or remove any. Tags are shared across every item type, so one tag groups a tool, a repository and a channel together.
• The Tools menu is now Library. Scout’s main menu reads Opportunities, YouTube, Library, Studio, Motion Lab. Inside Library the tabs are Tools, SEO, Docs and Categories, in that order, and the page opens on Tools.
• Edit any section of any report. Every section of a website Tool report, a YouTube channel or video study, and a GitHub repository record now has a pencil that opens the same Markdown editor the notes use, with the section’s current text ready to correct, complete or rewrite. Your version is what the report shows, reads aloud and answers from. The transcript and the GitHub facts stay as collected.
• Categorize with AI. A tool that arrived without a category has one button: the analysis picks the place in your taxonomy, creates the subcategory when none fits, and files the tool. Change it any time in Edit tool.
• Sites behind a bot wall. When a website refuses plain requests with a security checkpoint, Scout reads it with the browser instead and says so in the report; a site that refuses even that is reported plainly rather than as a generic failure. Failed web searches are no longer kept for thirty days.
• SEO cards show the trend. Each website card in the SEO library draws its six-month traffic line instead of repeating the summary in text.
• Phone menu on grasppy.com. The Why Grasppy group in the phone menu now collapses and expands, and a long menu scrolls.
• Observation and Ask Scout on every repository. Repository pages carry the same right panel as website tools. Ask Scout answers from the stored facts, README and review, and when a question calls for it, reads issues, releases, a named file, recent commits, contributors, or a page the README links to, live from the repository. Each answer names what it looked up. Nothing outside the repository and the sites it links to is ever read.
• Shorter Edit tool form. Tags moved out to their own window, and the remaining fields sit in a wider, denser layout that fits on one screen.
• Long report sections fold. On website Tool reports, any section taller than about eight rows shows its first rows and a “View more” button, the same fold the repository report uses. The Sources table no longer runs past a whole page.
• GitHub repositories stand out in the library. Repository cards have a graphite header, the GitHub mark, a corner badge, and a facts line with stars, licence and last code change, so a mixed page tells the two kinds apart at a glance.
• Facts stay fresh, dormant projects stay visible. Facts are re-read after a week and pictures after a month when you open the page. A repository that is archived or has had no code change in eighteen months is marked “needs a look” with the reason, never hidden.
• Track this in Scout. Every repository card and page on Repo Radar now carries one link that adds that repository to your Scout workspace, or takes you through sign-in first and lands you on it afterwards.
• Repo Radar has an area index and a page for every repository. The stats line on the front page opens sixteen areas, each listing its repositories with descriptions and avatars; every repository has its own page with the full card and its screenshots; search results open that page directly.
• Repo Radar is listed under Tools in every footer across grasppy.com, including the Russian pages.
• A Motion Lab composition example. A reference clip brings Milano illustrations and Scout Man together with a title banner, an explanation note and a supporting icon, showing how the elements can share a coordinated colour palette. This is a curated example, rather than a new automatic-generation mode.
• Generate an alternative illustration. Create an image from your clip’s narration and optional description, using examples from the selected illustration style and your scenario colours. Choose the image model in Settings; background removal is automatic.
• A reusable personal asset library. Generated images are saved automatically in Studio’s Assets tab. Import images, videos and animations, edit asset names and tags, and find saved work with search and tag filters.
• Keep alternatives with each clip. Save several image choices, preview them at a larger size, and explicitly choose which to use. Removing an image from a clip keeps it in Assets; permanent deletion is a separate, confirmed action.
• More room to choose and create. Larger asset windows provide space for descriptions and previews. The asset picker opens with the six latest saved items, with search and more results available as your library grows.
• Choose the pace of movement. Each clip offers Subtle, Balanced or Lively animation, with Balanced as the default. Illustrations can enter, hold and leave with the narration instead of staying fixed throughout the clip.
• Clearer library and recovery controls. Get Illustrations doodles have their own library grouping. Interrupted image requests can be removed once they are no longer running.
• Read and edit the whole storyline. Expand The story into a large window with separate Preview and Edit tabs. Preview shows the full storyline in a clean reading layout; Edit keeps the summary, story arc and scene details together, with highlighted scene cards and numbered badges. The right panel uses the same scene styling.
• Choose illustrations for the next clip. Find SVGs shows relevant choices from the scenario’s illustration style. Select up to three required illustrations, request another selection with Regenerate, or collapse the results while keeping your choices.
• Inspect a clip’s asset pool. Video clips and Asset pool now have separate tabs. In the private visual-library workflow, review the candidates gathered for a clip and select a pool for regeneration.
• A quieter clip workspace. Previous attempts fold behind “Show previous clips”, and the wider right panel gives story details and settings more room. Illustration style is set once in the scenario’s main settings and shown in each clip.
• More readable animated clips. Caption highlights have better contrast, narration has shorter unnecessary pauses, and composition fitting keeps artwork clear of frame edges and the presenter.

Repo Radar Publishes Itself, Mails Itself, And Has An Outline

Each edition of Repo Radar now goes from written to published to delivered in your inbox with nobody involved, heals itself when something arrives incomplete, and every edition page has an outline so you can jump straight to one repository.

• An outline for every edition. A new Outline button opens a panel listing every repository in the edition with a one-line description under each. Choose one and the page jumps to it. On a phone the Outline and Share buttons sit together under the header and stay there while you scroll.
• The daily email is automatic. When an edition is published it is sent to every confirmed subscriber within the same minute. Edition 18 was the first to arrive this way, at 7:35 in the morning Pacific time.
• Partial data no longer stops an edition. If a delivery arrives with a missing summary or missing bookkeeping, the site fills what it can from the edition’s own text, publishes it, and records what was derived. Only an edition with no usable text is held back.
• An hourly check that a reader can actually open the page. Yesterday an edition appeared as a card that led nowhere while every internal check said it was fine. A watchdog now opens the newest editions on the live site every hour and rebuilds any that does not answer, with no one involved.
• Missed days catch up on their own. Editions may now arrive out of order, and the daily run asks the site which editions are missing before delivering. A missed morning is filled the next morning instead of blocking every day after it.
• Subscriptions tidy themselves. An address that never confirms gets one reminder after two days and is removed after thirty. Confirmed subscribers are never touched.

Fifty-One Corrections On Repo Radar

The radar proved a project was alive by reading its release page. Projects that ship code without cutting releases were recorded as dormant or unverifiable — and fifty-one of them were neither. Every one is corrected, and the check now measures the last commit instead.

• Twelve projects recorded as dormant were alive. Decap CMS was published as “no longer maintained” on the strength of a release from June 2024. Its last commit landed the same morning we checked. On six of them the gap between the newest release and the newest code was fifteen to twenty-seven months.
• Twenty-one filed as “could not verify” were healthy. Among them Motion, Radix Primitives and SearXNG — none of which publish tagged releases, which is a habit and not a health problem.
• “Not a fit” is retired. Nineteen projects had been set aside for suiting the reviewer’s own products rather than for anything about the software. That is no longer a reason to leave anything out, and all nineteen are back.
• Seven projects had been renamed without anyone noticing. Hoarder is now Karakeep; Presidio has left Microsoft. Their histories are joined rather than split in two.
• A licence is a warning now, never a disqualification. Directus, NocoDB and tldraw were excluded for their licences while seven projects with the same kind of licence were published with a note. The note is the right answer; the exclusion was not.

Repo Radar Is Now Public

A dozen open-source repositories, opened and checked by hand, every morning — including the ones that did not make it, named with the reason. Free to read, no account needed.

• Fourteen editions, free to read. Every repository has been opened, not just listed: the last release date checked, the licence read, and a project that has stopped moving said to have stopped moving. Live now at grasppy.com/radar.
• Search every repository reviewed. Type a project name or describe what you need, and search looks across all 163 published reviews by name and description at once, instead of opening editions one at a time.
• Copy a whole entry in one click. Each repository has a copy button that puts the complete review on your clipboard as Markdown — the facts, the good and the bad, and the exact install commands — ready to paste into whatever tool you are working with.
• The ones left out are published too. Sixty-seven repositories were opened and not published across these editions, each with its reason: could not be verified, did not qualify, or simply not a fit. A list that only ever shows wins is less trustworthy, not more.
• A new About page. How one edition is made, step by step, the rules every check follows, and the mistakes found later and corrected in public.
• Read it your way. By email, by RSS, or on the site. The whole archive stays open whether you subscribe or not, and one click unsubscribes.

Journal Subscriptions Now Work

Anyone who signed up to follow a Grasppy journal never received the confirmation email, so no subscription ever completed. That is fixed, and the privacy policy now sets out plainly what happens to an address you give us.

• Confirmation emails are sent. Subscribing to a journal now sends a real message with a working confirmation link. Until today nothing was sent at all, so every subscription since launch stayed incomplete through no fault of the subscriber.
• Nobody is added without confirming. Entering an address does not subscribe it. The click on the emailed link does, and only that.
• Emails are readable in any mail client. Rebuilt on a light background with real text throughout, so nothing disappears in dark mode and the message still reads in full when images are blocked.
• A clearer privacy policy. A new section explains what a subscription stores, why the signup IP is kept, that the address is used for that one purpose, and that unsubscribing takes one click with no login. A postal address now appears on the policy and on every email we send.
• Leaving is one click. Unsubscribe links work without signing in, and take effect immediately rather than eventually.

Topics Group Themselves in Scout

Scout now reads your saved Overview videos and groups them into topics on its own, and those topics become filters you can browse with.

• Topics for everyone. Topic Opportunities is now open to all Scout users. It reads only the videos already saved in Overview, and never runs new YouTube searches or reads transcripts.
• No button to press. Grouping refreshes on its own whenever Overview brings in a new set of videos. An unchanged set reuses the saved result instantly.
• Filter by topic. Topics appear as their own group at the bottom of the Overview sidebar. Select one or several to narrow the video list, and clear them all in one click.
• See the shape of your research. A new chart ranks your topics by how many saved videos sit in each, so the balance of your sample is visible at a glance.
• Results survive imperfect analyses. A video listed under two topics is now counted once, in the first, instead of discarding the whole grouping, and an incomplete run is repaired from what it already produced rather than starting over.
• Easier to read. Larger video thumbnails, one-line result details, and a clearer place for the Overview search action.

Organize Your YouTube Research and Collections

Scout adds more control over video discovery, easier collection assignments, and clearer navigation for documents and assets.

• Balance your research. Overview divides a target of 50 videos across your active research interests and reuses compatible results collected within the last 24 hours. See estimated searches remaining and the countdown to the midnight Pacific reset.
• Choose your sources. Search across all channels or your saved channels, exclude blocked channels, and choose videos, Shorts, or both formats. Browse saved results with separate display filters.
• Review and export results. Switch between videos and a channel table with handles and saved-channel highlights. Export filtered research as Markdown with channel summaries and video links.
• Build video collections. YouTube now includes a Collections tab. Create an empty collection or assign multiple saved videos in one popup, edit assignments later, and use a video’s menu for quick collection assignment.
• Find documents and assets. Docs, including its add-document action and reader, now sits under Tools after SEO. The main Scenarios section keeps Scenarios and Assets together.
• Scan Scout more easily. Consistent bold sidebar headings, clearer group spacing, compact channel tables, and a simpler export action make research easier to browse.

Explore YouTube Topics with Fewer Searches

Scout’s Overview brings popular videos into one place, with saved results and more control over when searches run.

• Choose your exploration topics. Save topics and choose up to five for Overview. Select one or several to filter the results.
• Browse popular videos. Choose a 1, 3, 7 or 30-day publication window, see ten results first, and expand to fifty. Search saved video titles and channel names, or filter by video format.
• Update when you need to. Results stay in the same browser for up to 30 days. Opening Overview no longer starts a search, and manual updates ask for confirmation before using Google’s search allowance.
• Fewer searches, clearer limits. Popularity-only discovery uses one search per topic per language. If Google’s allowance is reached, Scout explains the limit and keeps existing saved results visible.
• Choose your listening languages. Set a primary language and an optional secondary language. Discovery excludes videos with missing or unmatched reported audio languages.
• Organized settings. User and super admin tabs group related preferences and connections into consistent sections.

Scenario Builder Gets Animated Presenters

Whiteboard clips can now carry a character who stands beside the drawing, talks with the narration, gestures on the sentences and points at what is being drawn.

• Two characters, one rig. Him and her share every gesture, reaction, emotion and movement, so a pose tuned once works on both. The mouth opens on the narration’s own word timings and loudness, and the head stays calm apart from a small dip on stressed syllables.
• Switch it on per clip or per scenario. The clip dialog’s Drawing group and the scenario’s new Brand kit group choose the presenter, its side and its size. Off by default, so existing clips render exactly as before.
• Brand kit. Pick a character, or build your own from approved heads, bodies and colours with a live preview, and make it the default for new scenarios. Every gesture works on every kit.
• Dialogue scenes. Start a line with a character’s name and a colon and the two talk to each other, each in its own voice; the speaker gestures while the other listens and nods.
• Pointing that reads. A pointing act names the drawn object and the character picks the arm on that side and aims at it, never over the head.
• Movements built from reference clips. In the Studio, every pose can carry the GIF or MP4 it was copied from, shown beside the rendered loop with an eight-frame compare sheet, and a list of the emotions it plays. The planner reads those emotions and picks the movement that fits the feeling in the sentence. New poses start as request cards: a name, the emotions and the clip, built into keyframes afterwards. A Copy prompt button on each card hands the whole job to any coding agent: the pose, its clip, the rig’s conventions and the method.

Scout Finds Opportunities Across YouTube Channels

Scout now turns saved channel research into a focused view of breakout videos, comparable channel signals, and promising topic patterns.

• Opportunities is now the starting point. The empty Home section has been replaced by a YouTube intelligence workspace organized around Rising Now, Compare Channels, and Topic Opportunities.
• Breakouts are ranked against each channel’s own baseline. Filter observed videos by format and publication window, inspect public views and publish-age pace, and open the source video directly.
• Channels can be compared on equal terms. Select two to five channels and review recent reach, engagement, publishing cadence, strongest breakout, relative reach, format, and momentum in a visual comparison and detailed table.
• Topic ideas stay connected to evidence. Repeated patterns show their confidence, median breakout strength, contributing channels, and the exact videos that support the opportunity.
• SEO-style visuals make movement easier to read. A breakout timeline, normalized comparison bars, and topic-share donut use saved public evidence and clearly distinguish catalog pace from live views-per-hour.

Scout Scenarios Becomes a Complete Video Studio

Scout can now plan a consistent story, build richer whiteboard scenes, and export the complete video with its narration, captions, music, and branding.

• One story across every scene. Start with a planned length, import a complete script, and use the editable Story tab to keep the premise, cast, setting, and scene progression consistent from the opening to the close.
• Richer whiteboard direction. Scout researches concrete visual subjects, chooses suitable library objects, keeps printed labels relevant, composes the scene on one canvas, and moves the camera through the action instead of presenting a row of unrelated icons.
• Real product demonstrations. Screenshots remain intact inside traced or clean frames, while animated cursors, click effects, circles, arrows, labels, colour, and camera movement explain what is happening on screen.
• Per-clip creative control. Adjust duration, voice or uploaded narration, speaking speed, captions, handwriting, text weight, colour treatment, framing, and other presentation choices for each clip or for the whole scenario.
• A finished video from Scout. Join selected clips with transitions and pauses, add background music with automatic voice ducking, place a configurable logo watermark, preview the estimated duration and size, and download the completed landscape or vertical export.

Scout Gives Customer Review Search Its Own Model

Tool reports can now search independently for customer reviews, complaints, community discussions, testimonials, and case studies without changing the model used for the rest of the report.

• Web search has its own model choice. Scout Settings keeps customer-review discovery separate from Tool report synthesis, so each job can use the model that gives the best balance of quality and cost.
• Model prices are visible before selection. The searchable model menu shows the provider, model description, and current input and output prices per million tokens.
• One focused live search per refresh. Customer evidence uses a single bounded web-search request and reports when no citation-grounded review or customer proof was found.
• Search leads are screened before publication. Exact-product matches can appear as independent reviews, customer proof, complaints, or community mentions, while rejected and unverified leads remain available in Review evidence.
• The model picker is now reusable. Scout and GRASPPY administration share the same searchable selector, provider badges, descriptions, pricing, and keyboard behavior.

Scout Scenarios: Whiteboard Clips From Your Own Scenes

Write a scenario one short scene at a time and turn each scene into a narrated whiteboard clip drawn from your own screenshots.

• A Scenarios tab in the Docs library. Scenarios are their own kind of document: create one with a name, add scenes as short sections, and Listen to each scene as you already can with docs. They never mix into the Docs shelf.
• One clip per scene, drawn from your screenshots. Open a scene's clip slot, explain what the clip should show, paste or upload screenshots, and click Generate. The scene text is the narration, read in the same voice Listen uses, and the drawing appears stroke by stroke in time with the words, with circles, arrows, labels and word-by-word captions.
• Every attempt is kept. Comment on any attempt and press Send to get a new one that applies your notes; select the attempt you like and it appears beside the scene. Earlier attempts stay available.
• Exact clip length, with or without a voice. Type any length from 3 to 60 seconds for every clip and it comes out exactly that long, the drawing holding once the narration ends. Switch on No voiceover for a silent clip that still follows the scene text.
• Settings per scenario. Orientation, resolution, style, accent colour, captions, pacing, branding and the AI models used for planning, with an Auto choice that explains what it picked on each attempt.
• A library of 4,500+ hand-drawn objects. The Library tab beside Docs and Scenarios holds people, devices, arrows, emoji and everyday objects, each with a one-sentence description, a category and tags you can correct in place. Search by what you mean, such as "someone stressed at a laptop", not only by name, browse by category, and page through 50, 100 or 200 at a time.
• Private by design. Uploaded screenshots, recordings and rendered clips are stored under your own account and served only to you.

Read Scout Reports in Your Own Language

Scout reports can now be read in the language you choose, arrive in a clearer layout, and keep research separate from the tools that produce it.

• Choose a reading language once. Set your language in Scout Settings and every report page offers two views: the original and yours. A translation is prepared once, saved, and switches instantly afterward, with a check mark when a saved version is ready.
• Reports read as findings, not walls of text. YouTube channel reports now present each finding on its own line with a clear label, and the strategy section speaks to you directly about what you could adapt.
• Independent reviews are easier to reach. Tool reports search for review profiles, customer stories, and independent write-ups about that exact product, then link them with a clear note that the source was discovered rather than verified.
• Company, security, and trust is answerable at a glance. Tool reports lay out ownership, contact, legal, security, and reliability signals in one table and say plainly when something is not published.
• SEO lives with the software it belongs to. SEO moved into the Tools section alongside Categories, All tools, and Collections, and channel cards show when analysis is still running.
• Reports stay focused on the research. Report text describes the findings rather than the internal services used to gather them.

Scout Honors the AI Model Chosen for Each Workflow

Scout now keeps private report processing on the model selected for that workflow without starting unrelated analysis afterward.

• Each report preserves its selected model. Tool, SEO, YouTube video, and YouTube channel runs keep the model chosen when processing begins.
• Private Scout reports avoid duplicate processing. Completed Scout imports no longer start unrelated background analysis that adds time and provider charges.
• Report imports finish more cleanly. YouTube and SEO reports connect to their Scout records before parsing begins, preventing an extra processing path from starting during import.

Scout Keeps Each Add Dialog Focused

Scout’s YouTube import dialogs now show only the source type that belongs on the current page.

• Add Video is dedicated to videos. The YouTube video page opens a clean video URL form without Website, GitHub, Research Topic, or channel choices.
• Add Channel is dedicated to channels. The channel page accepts channel URLs without asking users to choose a research type first.
• Wrong YouTube links are caught before import. A channel URL on the video page, or a video URL on the channel page, now produces a clear correction instead of creating the wrong record type.
• YouTube pages remain available while jobs are active. Scout’s persistent activity indicator now resolves workspace-aware links correctly instead of interrupting the page render.

Scout Lets Every Research Workflow Choose Its AI Model

Scout settings now keep separate AI model choices for Tool, SEO, YouTube video, and YouTube channel reports, making quality and processing cost easier to control.

• Each Scout workflow has its own model setting. Tool reports, SEO reports, YouTube video reports, and YouTube channel reports can use different models without changing the others.
• Model choices can be changed at any time. New imports and refreshes save the current selection for that run, so later settings changes do not alter work already in progress.
• Lower-cost model testing is available. DeepSeek V4 Flash joins the Scout model list for users comparing output quality, speed, and cost against existing options.
• Existing Scout setups keep working. Current model preferences carry forward automatically, while the previous video-model setting remains compatible with older records.
• Usage reporting reflects provider charges more accurately. Scout records the cost reported by OpenRouter when it is available and retains estimated pricing as a fallback.

Scout Strengthens Tool Research and Speeds Up Analysis

Scout now produces clearer software research, finds stronger public evidence when requested, and shows what is happening while a Tool report is being refreshed.

• Tool reports answer the practical product questions. Identity, features, pricing, audiences, integrations, customer proof, and documentation are organized into concise evidence-linked sections instead of loose page excerpts.
• Independent evidence can be discovered on demand. An optional search finds review profiles, customer stories, documentation, guides, support resources, roadmaps, changelogs, and status pages while clearly identifying anything that was not found.
• External proof stays fresh without unnecessary searches. Results are cached for 30 days, and a separate refresh action lets users deliberately search again without coupling it to the independent SEO workflow.
• Tool analysis completes more efficiently. Tool and SEO jobs use separate processing lanes, while sitemap discovery and supported page rendering run in parallel within bounded limits.
• Long-running refreshes explain their progress. The Tool page displays the current collection or writing stage with elapsed time, then reloads the completed report through Scout’s existing update flow.
• Scout has clearer account navigation. Eligible users can open Scout from the Grasppy account menu, and Scout’s own menu provides Settings, a return to Grasppy, and Sign out.

Scout Adds a Docs Library with Voice-Ready Books

Scout can now keep a private document library — files, pasted text, Google Docs, and full EPUB books — where every document becomes structured, editable chapters and sections that can be read aloud.

• Documents import from the formats people actually have. Add PDF, Word, EPUB, Markdown, or text files, paste text directly, or link a shared Google Doc; each becomes a readable document with sections, formatting, and voice playback.
• EPUB books keep their real structure. Imports follow the book’s own table of contents, so chapters and sections match the published book, the title comes from the book itself, and long chapters gain navigable sections.
• Reading happens in a dedicated workspace. A clickable outline, numbered section cards with per-section Listen, a private note, and Ask Scout questions grounded in the document’s own text sit on one page.
• Documents stay editable after import. Sections can be added between cards, edited in place, or deleted; chapters can be created; and both chapters and sections reorder by drag & drop in the outline.
• The library stays organized as it grows. One Docs view aggregates every format with search, sorting, import-state filters, file-type filters, tags, and automatic short summaries on each card.

Scout Opens Anywhere and Finds Better Tool Evidence

Scout is easier to open on phones and additional browsers, while software Tool research can collect more complete product evidence without changing the separate SEO workflow.

• Scout can be opened directly after sign-in. The dedicated Scout sign-in uses the existing Grasppy account and returns users directly to their private research workspace.
• Local folder setup no longer blocks cloud access. Mobile and unsupported browsers bypass the desktop-only folder request, while desktop users can dismiss it and continue reading documents or using Scout.
• Tool reports can find stronger product evidence. When enabled, Firecrawl discovers real site pages and targets missing pricing, features, integrations, customer proof, reviews, and product updates within strict request limits.
• Tools and SEO remain separate workflows. The improved collection applies to software Tool research; SEO keeps its existing collection behavior and report workflow.
• The startup taxonomy is cleaner. Development tools now live under “Development & Coding,” and overlapping CRM and customer-data email categories are consolidated without losing existing Tool assignments.

Doc Reader Voice Adds Read-Aloud Playback

Grasppy can now read supported document and Scout report sections aloud while keeping the original text available for reference and copying.

• Listen controls now sit beside section copy controls. Start read-aloud playback directly from supported Plan and Scout report sections without interrupting the surrounding workflow.
• Playback follows the document one section at a time. The reader provides sentence-level progress, previous and next section navigation, voice selection, speed controls, and pause or resume behavior.
• English and Russian reading options are supported. Available voices are grouped by language and provider so the reader can match the source material.
• Personal Speechify access is supported. Users can connect their own Speechify key for premium voices, while access and cached audio remain restricted to the owning account.

Scout Adds Dedicated Software Tool Intelligence

Scout can now turn a software website into a structured product report, keep every tool in a practical startup taxonomy, and separate private software records from tools intended for a public curated library.

• Software websites now have their own Tools workflow. Add or reprocess a product website without sending it through SEO, then continue working while Scout collects public product evidence in the background.
• Reports cover the information needed to evaluate a product. Scout organizes identity, features, pricing, use cases, integrations, customer proof, updates, trust signals, technical delivery, alternatives, sources, and limitations into a consistent evidence-linked report.
• Collection can adapt to difficult websites. Standard page collection is supplemented by browser rendering when needed, with an optional Firecrawl fallback that can be enabled or disconnected in Scout settings.
• Every tool receives one clear taxonomy path. The startup-focused Group → Category → Subcategory structure supports search, automatic proposals with owner approval, reassignment, private groups, and deeper email-marketing coverage.
• The Tools workspace is easier to scan and manage. Compact metadata cards, app-name and logo fallbacks, clickable tool rows, usage and payment facets, status filters, edit/reprocess/delete menus, and stable loading keep the library useful as it grows.
• Notes and saved Scout answers remain usable. Notes can be copied as raw Markdown, saved questions and answers can be edited, deleted, and reordered, external links open at their exact destination in a new tab, and the right panel stays within its assigned space.

Scout Turns YouTube and Websites Into a Private Research Library

Scout gives focused research its own private workspace: send a YouTube video, channel, or website for analysis, continue working while it processes, and return to a structured report with your notes and saved questions beside it.

• One YouTube entry point recognizes videos and channels automatically. Imports continue in the background, pending cards remain visible, and completed work opens in the correct library without making you wait on the page.
• Channel research now reads like a strategy report. Performance signals, breakout and weaker videos, content patterns, opportunities, methodology, and sources share one consistent report structure. Existing channels can be reanalyzed when the evidence becomes outdated.
• Website analysis combines traffic and SEO evidence. Reports include traffic trends, visitor countries, organic keywords, search results, important pages, technical checks, and a captured view of the website, with collection sources and costs disclosed in the report.
• Your notes and Scout conversations stay with the research. Notes remain editable, while questions and grounded answers are saved automatically in newest-first history with clickable source links.
• Research is easier to organize and find. Automatic tags, editable labels, collections, search, channel cards, and tool taxonomy keep related videos, channels, and websites grouped without changing the generated analysis.
• Scout remains separate from ordinary Grasppy work. Its managed storage is protected from normal workspace cleanup, search, extension destinations, and sharing unless the Scout workflow explicitly allows the action.

Memory Snapshots Now Correct Outdated Facts Instead of Stacking Them

The Memory Snapshot skill used to judge only what was already stored. It now also decides what happens to each new fact before writing it, so a correction replaces the entry it disproves rather than sitting next to it.

• A correction rewrites the entry it replaces. When something you captured supersedes an older note, the older note is updated in place. Your memory stops holding two answers to the same question.
• Repeats from the same pass are collapsed before they land. New facts are compared against each other first, so one thing said twice in a session is stored once.
• Entries keep their full timeline. Each one now records when a fact was first seen alongside when it was last verified, so combining two notes no longer erases how long something has been true.
• Nothing is dropped silently. When a new fact is withheld because an existing entry already covers it, the snapshot report says so and explains why.
• New workspaces start with the updated skill. The standard skill seed already carries this behavior, and existing personal skill copies remain yours to update when you choose.

Plan Check-ins Now Finish Completely or Change Nothing

Plan Check-in now validates the complete file before applying any changes, so one malformed instruction cannot leave a project half-updated. Newly created workspaces also receive the current standard Grasppy skills.

• One check-in, one complete result. Every update and new node is prepared as a single operation. If any part fails validation, none of the file is applied.
• Bad files stop before they touch your plan. Missing headers, mismatched operation counts, broken marker pairs, duplicate IDs, and unsafe parent instructions are reported before any content changes.
• New chapters and sections keep their intended location. Parent references are resolved consistently, including sections created beneath a new chapter in the same check-in.
• Review actions cannot leave a half-finished version. Starting, approving, or rejecting a Plan Check-in now completes as one protected action or rolls back cleanly.
• New workspaces start with current skills. The standard customer skill seed is checked against the active skill definitions and refreshed whenever it drifts. This affects newly created workspaces; existing personal skill copies remain yours.

Better YouTube Imports, And Shared Plans AI Can Read

YouTube imports now open as navigable knowledge reports instead of stacks of disconnected blocks, while published plans can be read directly as clean Markdown by AI tools.

• The video is part of the report. Every new full YouTube analysis starts with an embedded player, so the source and the analysis stay together.
• An outline that matches the report. The Outline view lists all eleven generated sections in their real order, from the short summary and transcript through frameworks, evidence, and an action plan.
• The transcript is visible and reusable. It appears as the third section, starts collapsed so it does not overwhelm the page, remains searchable, and is still available as the transcript artifact.
• Reference facts and sources are where you need them. Video metadata, public statistics, chapters, tags, and meaningful links now appear as normal report sections instead of duplicate artifacts.
• Search by remembered fragments. Each summary includes 10–15 specific hashtag keywords, and partial text such as #hook can find a longer keyword such as #hook-tension-payoff. Grasppy also remembers the last analysis model you selected.
• Published plans now speak Markdown. A shared plan can expose the complete document, individual sections, and an llms.txt index, giving AI tools clean, structured access while keeping the same expiry, passcode, and privacy rules as the normal shared view.

Relay — Every Finding Gets A Number, And Nothing Runs Out Of Time

Two agents could disagree about the same problem for twenty rounds and nothing could tell whether it was one problem or three. Findings are now numbered, and a fix only counts when the agent that didn't write it says so.

• The Issue Register. Every finding gets a number and keeps it. It moves from open, to closed by whoever fixed it, to confirmed by the other side — and an agent can never confirm its own closure. A relay now ends when nothing is open and nothing is waiting on a confirmation, rather than when two counts happen to agree. You can close, reopen or merge any of them yourself, and issues left open carry forward into the next relay on the same plan.
• Nothing runs out of time. The round cap, the per-turn timeout and the time budget are all gone. A turn that needs nine minutes gets nine minutes, and a relay waiting on your answer is still waiting tomorrow.
• Approve in one word, and the agent does it. When an assistant needs your authority it now stops and shows you the exact command, why it wants it, and what changes. You reply once and it carries the work out itself. Committing, pushing, deleting data and database changes still always need you — deploying and restarting a service no longer do.
• Stop and Release, where you can see them. The relay list shows one action at a time: Stop while a relay is live, Release once it has ended, and nothing when it is finished with. Stop now also works on a relay that is paused or waiting for you.
• Clearer endings. "Done" is now "Agreed" — both agents reported nothing open — and "Released" is its own stage, so a finished relay that is still holding your two conversations is obvious at a glance.
• Better notes. Paste a screenshot straight into a note, or embed a document so the agents read it from disk instead of receiving a copy. The recipient you last chose is remembered per relay, and the send button now names who it is going to.
• The Ledger reads a long run for you. Ninety rounds is a lot of transcript. The Ledger shows what was decided and when, which issues are open against which phase, and how findings opened against findings settled over the whole run.

Relay — Three Ways To Pair, And No Limit On How Many

Relay could do one thing: two different AI tools, one at a time. It now matches the three ways people actually work — including just you and a single agent — and it no longer makes you wait for one relay to finish before starting the next.

• Solo — one agent and you. Pick a single conversation and it becomes a relay you can read and answer from anywhere. It replies, then waits — no counterpart, no rounds to close. It still tells you what it considers unresolved after every turn, which is the part worth keeping. It ends when you release it, like closing a chat.
• Same platform, with a Consultant. Two sessions of one tool, for the handover you already do by hand: a chat running low on room writes a summary, a new one takes the work, and the old one stays on to advise. The Consultant is told to hand over what it knows before the other needs it — the opposite of a reviewer's job, and the reason the pattern works.
• The first question is which kind. New Relay opens with three choices, and the rest of the form asks only what that kind needs. Cross-platform is unchanged — same fields, same behaviour, same defaults.
• Run as many as you like. The one-relay-at-a-time limit is gone. Start a second, a third, in the same folder or different ones. If two share a project folder you get a warning naming the other relay, and then it starts anyway — you know your own work better than the tool does.
• Pick a plan by name. Choosing the plan a relay serves used to mean typing an ID number. It is a searchable list now, newest first, showing which plans have phases.

Relay — See What Happened Without Reading All of It

A relay can run to seventy turns, and until now the only way to find out how it went was to read them. There is a second view now that keeps score: what was decided, which phase of your plan the work reached, what is still open, and one line for every turn.

• A Ledger tab, beside the journal. Switch between the full exchange and a derived summary of it: a line per turn you can scan, expandable when one is worth stopping on, plus the decisions, the files touched, and how the open points moved round by round. Nothing on that page is written by the agents — it is worked out from what they already said, so it costs no extra round.
• Your plan's phases, as a progress strip. If your plan has lettered phases, the Ledger shows A, B, C across the top with the current one marked. The names sit behind a legend you open once. Green means the work moved past a phase — not that anyone verified it.
• Open points are named now, not counted. Each agent used to end its turn with a bare number, and nobody could say what was in it. They now list what they consider unresolved, so "3 open" tells you which three.
• A private word with one agent. Tick Keep this private on a note and the reply comes back to you alone — the other agent sees neither the question nor the answer. Correct one side without the other reading the correction. Private exchanges are marked in amber wherever they appear.
• Pick a plan by name. Choosing the plan a relay serves meant typing its ID number. It is a searchable list now, showing your workspace's plans newest first, and telling you which ones have phases.
• Releasing a relay gives your chat its name back. A conversation held by a relay is renamed so you cannot type into it by mistake. That warning used to outlive the relay; now the real name returns shortly after you release it.

Relay — You Decide What Each Agent Can Do, and How Often It Stops to Ask

Two settings per agent, chosen by you: what it's allowed to do, and when it should stop and ask. Both can be changed while the relay is running. And the notes you write now actually reach the agents — that one was broken, and it's worth saying so plainly.

• Access, per agent. Read only, write, or write and verify. This is real permission the tool itself honours, not a sentence in a note — an agent told it may edit files, but started without permission to, spent whole rounds failing at work it was allowed to do. Committing, pushing, deleting data and touching servers are never granted, at any setting.
• Autonomy, per agent, separately. Ask about everything, handle routine work on its own, or stop only when genuinely stuck. Access says what an agent can do; autonomy says when it should interrupt you. They failed separately — an agent with full write access still parked a relay to ask permission for something obviously fine — so you set them separately.
• Your notes reach both agents. A note is now attached the moment a turn is sent rather than when it was queued, so an answer you write while the relay is parked goes out with the very next turn. Addressed to both, it reaches both. Previously a note could be recorded, shown in the journal, and still never make it into what the agents read — which is why they sometimes re-asked things you had already answered.
• No round cap and no time limit. Both are gone. A relay ends when the agents agree it's finished, when one of them hands it to you, or on an error — never because a counter ran out mid-thought. If both sides go five rounds without anything changing, it stops and asks you instead of looping.
• An agent can no longer end your relay. When one needs you, it parks and waits — indefinitely, with nothing expiring and nothing lost. Reply with a note and it carries on from exactly that point, your answer in hand.
• Get your conversations back whenever you want. The Reserved marker is now the button, on every relay in every state — including one still running. Releasing hands both chats back and ends the relay cleanly; the journal keeps everything either way. The reading view also opens on the newest turn now, with buttons to jump between turns rather than scrolling for them.

Auto Capture Now Sees Every Platform You Use

Auto Capture used to recognise three kinds of conversation. It now recognises every platform Grasppy integrates — ChatGPT, Gemini, Claude, Grok, Perplexity, DeepSeek, Mistral, TypingMind, GitHub Copilot, Replit, Bolt and Lovable, alongside Claude Code, Codex and Cursor.

• One list for everything. Whatever you were talking to, the conversation shows up in the same place, with the same Import button and the same status. No separate flow to remember for browser chats.
• A platform filter that handles sixteen of them. Tick Claude Code, Codex and Perplexity and see only those. Filters stack with search, project and status, and clearing it always brings everything back — you can't accidentally end up on an empty list.
• Auto Capture looks like the rest of Grasppy now. Same page header, same margins, same table, same canvas as Analyses — so switching between the two tabs doesn't feel like switching applications.
• Relays can be paused and picked up again. Hold a relay to read something and let it carry on when you're ready. A turn that runs out of time no longer ends the run — retry it, and everything already done is still there.
• Time budgets count working time. A relay that's sitting idle, paused, or waiting on you costs nothing. Only the agents actually running draws it down — so a relay you left overnight is still there in the morning, with its budget intact.
• Stop, continue, and copy. Cancel a relay you no longer want, continue a finished one where it left off, and copy any single response — as markdown, with its heading — straight out of the journal.

Relay Now Continues the Conversations You Already Have

A relay no longer starts two strangers on your plan. It picks up two of your existing chats — the ones that already know the problem — and lets them carry on working, between themselves, while you do something else.

• Nothing has to be imported first. If Auto Capture can see a conversation, it can be relayed. That's the only qualification. Importing into Grasppy stays optional, stays separate, and is never in the way.
• They remember everything. Both sides continue their own conversation, so the agent that argued the design with you three weeks ago still knows why the shape is the shape. No re-explaining, no starting cold.
• Lead and Partner, or two Peers. The Lead owns the work; the Partner argues the other angle and carries real work of its own — research, documentation, building the alternative it's arguing for. Two Peers, with nobody in charge, is there for when you're still deciding what to build.
• Disagreement is the point, so you choose how it's settled. Work it to agreement means nothing gets left behind: no point closes until both of them accept it, and the relay ends only when both say there's nothing open. Or let either side call it. Or let neither, and read the argument yourself.
• Your chats come back when you say so. While a relay runs it holds both conversations, and it tells you plainly — in the list, in the sidebar, and inside the chats themselves. Release them with one click and they're ordinary conversations again, still remembering everything the relay worked out.
• Notes are a proper editor now. Markdown, a preview, and room for twenty thousand characters — because the useful thing to send is usually a corrected table or a specification, not a sentence. One button turns a finished relay into a full analysis too: key points, entities, searchable, in the semantic map, with every turn naming who said it — including you.

Relay — Two AI Agents Work a Plan While You Do Something Else

Hand a plan to two coding agents and walk away. They review, revise, and check each other's work in turns — and the whole exchange lands on one page you can read, share, and step into whenever you want.

• Two agents, working it out between them. One revises, the other reviews independently, and Grasppy carries every message between them. What used to be an afternoon of copy-pasting runs unattended while you're elsewhere.
• They can be two different tools, or two sessions of the same one. Claude Code and Codex give you two genuinely different sets of priors. But two isolated sessions of one platform work too — and often better, because each gets its own workspace and neither can see the other's history.
• An older session as consultant, a fresh one doing the work. Choose how the two relate: a seasoned session that carries the whole design history reviewing a fresh one that starts each round with no inherited assumptions. That asymmetry is what makes the pairing sharp rather than an echo.
• The whole run reads like a document. Every exchange lands in the reading view — phase list on the left, the conversation set in book typography on the right — and updates itself while the run is live. Open it on your phone mid-run if you want.
• Step in whenever you like. Add a note addressed to either agent or both. It never interrupts the turn in flight; it rides the next one. Send it to both and you get two independent answers rather than one shaped by the other.
• It stops on its own, three ways. A clean verdict, a round cap, or a time budget — and never by hanging. The time limit is worked out from your own settings, so what the screen promises is what actually happens.
• You can see exactly what ran on your machine. A dedicated Agents & Safety tab shows which two commands are permitted, every invocation with its exit code and duration, and a kill switch that works even if the app is unreachable. Refused commands are shown, not hidden.

Focus Reader — Read a Conversation Like a Book, and Share It as One

A new distraction-free reading page for any analysis: your outline on the left, the conversation set in book typography on the right — and a link you can send to anyone, no account required.

• A reading page, not a workspace. Open any analysis and click the book icon to enter Focus Reader — no tabs, no panels, no chrome. Just a contained reading column in proper book typography, with your topic outline pinned alongside it.
• Your outline, three ways. Switch the outline between Sequential (topics and subtopics), Thematic (grouped by meaning), and Turns. Selecting a topic focuses the page on that topic alone — so even a 500-turn conversation opens instantly — and Next / Previous buttons walk you through the rest.
• Key points in one click. Toggle from Full text to Key points for a scannable digest of what was decided, hit, and resolved across the conversation, tagged by Decision, Problem, and Solution. It's built from analysis you've already run — instant, and free.
• Find anything, fast. Search filters the reading view as you type and highlights every match, working with whatever topic or theme you're focused on. Jump-to-top and jump-to-bottom controls sit within reach at all times.
• Share a conversation as a link. Create an unlisted reading link and send it in an email, paste it into notes, or keep it as a permanent reference. Recipients see exactly what you see — outline, full text, key points, search — with no account and nothing they can change. Links are never indexed by search engines.
• You stay in control. Add an optional passcode for extra protection, see how many times a link has been opened, and revoke it instantly. Shared conversations are marked so you always know which ones are live.

Auto Capture — Every Workspace Gets Its Own Destination

Auto Capture is now workspace-aware: each workspace connects the archive to its own document, opens filtered to its own projects, and imports always land exactly where you see they will.

• Per-workspace destinations. Connect Auto Capture separately in each workspace, each with its own target document. Imports from one workspace can never land in another workspace's document again.
• Project filter, set automatically. A new Project dropdown on the filter row opens pre-set to the workspace's own projects. One click switches to All projects — or a single project — when you want to review or import related work.
• Claim your projects. The Connect dialog lists every project found in your archive with conversation counts — tick the ones that belong to this workspace and the list opens scoped to them. Status counts follow the filter.
• Destination always visible. The toolbar shows which document imports land in (click it to change), and bulk-import confirmations repeat the destination before anything runs.
• Links open in the right workspace. Opening an imported analysis from the Auto Capture list now always lands in the workspace where that analysis actually lives — no more mixed workspace views.
• Clear errors on connect. If connecting the archive fails, the screen now tells you why instead of doing nothing.

Auto Capture — Same-Day Reliability Round

Fixes from the first hours of Auto Capture in real use: huge conversations import smoothly, the sync daemon heals itself, and archive filenames get a cleaner, ID-first structure.

• Very large conversations import reliably. A million-character conversation could stall analysis partway through diagram detection. Detection is fixed and now time-limited, so processing always completes — worst case, a huge chat simply gets fewer auto-detected diagrams.
• The sync daemon heals itself. A hung server call could previously leave the daemon looking disconnected until restarted. All its calls now carry timeouts — if anything stalls, the daemon recovers on its own within minutes, no reinstall, no re-pairing.
• Imports carry your current chat name. An imported analysis is now titled with the conversation's live name from Claude Code or Codex at the moment you import — not an older label.
• Claude Code imports parse correctly. Fixed a mismatch that could leave a freshly imported Claude Code conversation empty until reparsed.
• Cleaner archive filenames. Paired with today's Grasppy Capture app update: files are now named provider–ID–title with the conversation ID in a fixed position, existing archives are renamed automatically on first launch, and Grasppy reads both old and new names during the transition.
• Always know which workspace you're in. A small badge at the top of every workspace page — Plans, Documents, Codebase, Semantic Map, all of them — shows the current workspace name. Click it to jump to that workspace's home.

Auto Capture — One-Touch Conversation Sync Into Grasppy

A new Auto Capture tab lists every conversation Grasppy Capture has saved on your Mac, syncs the archive for you with one click, and imports exactly what you choose — full analysis, no duplicates, no second app. Plus: pick the AI model your whole pipeline runs on. Part of Grasppy Pro.

• Your archive, listed automatically. With Grasppy Capture and the Grasppy daemon on your Mac, the new tab (Analyses → Auto Capture) finds your archive on its own and lists every saved conversation — provider, ID, project, turns, size, status, and the conversation's live name: rename a chat in Claude Code or Codex and the table follows.
• One-touch Refresh. The Refresh button doesn't just re-read the list — it runs Grasppy Capture's own sync for you, so pending conversations land in the archive, names update, and statuses recompute. One button, one app.
• Import what you choose — updates append cleanly. Tick conversations and Import: up to 10 per run, processed one at a time, into the document you pick. Session-ID identity means re-imports never duplicate, and when a conversation grows its row shows Update available — syncing adds only the new messages.
• Full analysis on every import. Each import runs Grasppy's complete pipeline — summaries, outline, diagrams, entity graph — exactly like a manual upload, on your own AI key. Costs show up front (about $0.30 per 100 turns on Gemini 2.5 Flash, scaled to your model), bulk imports confirm before spending, and 500+-turn conversations split into linked Part analyses automatically.
• Work a big archive fast. Search across names, projects, and IDs; sort any column; filter by status; Exclude conversations you don't want listed (restore anytime); long catalogs page smoothly with Show more. Deleted an analysis? Its conversation returns to "Not imported" so you can bring it back.
• Choose your Analysis Pipeline Model. Settings → AI Provider now has a real pipeline selector: one model powers your entire analysis, on your own API key, with per-model cost estimates and recommendations. Leave it unset and analysis pauses until you choose — nothing ever runs on a key that isn't yours.
• Tables that stay current. The Analyses list adds Updated and Created columns, a Refresh button, and newest-activity-first sorting — new imports appear the moment they land — and the dashboard's activity calendar now always reflects today, including automated syncs.

The Capture User Guide, and Publishing Without Moving the Date

Grasppy Capture gets a complete user guide on grasppy.com, and blog authors can now push updates live without rewriting their post's publish date.

• Grasppy Capture user guide. A full guide now lives at grasppy.com/capture/guide — installation, your first sync, what every status means, the dashboard calendars, privacy, and troubleshooting. It's searchable, works in dark mode, and every section has a shareable link. The Capture app links straight to it from the new question-mark button in its header.
• Keep your publish date when re-publishing. The blog's Publish window now has an Update published date checkbox. Leave it checked and publishing works exactly as before. Uncheck it and your content, pin, and metadata changes go live while the post keeps its original date — your timeline stays honest, no matter how many times you re-publish.
• Pinned posts, your way. Together with the checkbox, unpinning a post from the Journal landing page no longer costs you a date change: toggle the pin, re-publish with the date box unchecked, done.

Grasppy Capture — a Free Mac App for Your AI Conversations

Your Claude Code, Codex, and Cursor conversations already live on your Mac — in formats you can't read. Grasppy Capture turns them into plain Markdown files in a folder you own, and it's free and open source.

• Meet Grasppy Capture. A small Mac app that finds the AI coding conversations already stored on your machine and archives them as clean, readable Markdown — with a stable name, message markers, and metadata on every file. One click to sync; unchanged conversations are skipped, so re-syncs take seconds.
• Entirely local. Capture makes no network requests while cataloging or syncing. No account, no sign-up, no background process — your conversations never leave your computer.
• Free and open source. The full source code is public under GPL-3.0, so you can read exactly what it does — or build it yourself.
• A new home at grasppy.com/capture. Screenshots, straight answers, and a direct download. You'll also find it under Why Grasppy → Free Capture App in the menu, and in the site footer.
• Mac today, Windows in progress. This release supports Apple Silicon Macs; a Windows build is underway.

Quick Share, Embedded HTML, and a Faster Way Around Your Workspace

Share a few sections with someone through a link that expires on its own, drop a self-contained HTML page straight into your documentation, and jump to any project in your workspace from one menu.

• Quick share. Select any sections, chapters, or parts, choose Quick share, and get a link to exactly those items — nothing else in the project is reachable through it. Pick how long it lives (15 minutes by default, up to 24 hours) and it stops working on its own. Private items are never included, and you can create as many separate links as you need, each with its own address and its own clock.
• Embed a full HTML page in a section. Paste a self-contained HTML file into an html-embed code block and it renders inside your document — diagrams, styled layouts, and mockups included. It runs isolated from the rest of the page, so its styling can't affect anything else, and there's an Open full size link for a closer look.
• Workspace menu with quick access. The chevron beside WORKSPACE opens a shortcut list to Skills, Prompts, Rules, Memory, Research, Plans, Documentation, and Journal. Clicking WORKSPACE itself still opens the workspace as before, and whether you leave the list open or closed is remembered — including across sign-outs.
• "Plans" now reads "Projects." The word better describes what these actually hold: user guides, documentation, research, and journals as well as plans. Only the wording changed — your links, bookmarks, and content are untouched, and My Plans keeps its name.
• New workspaces connect their Data Vault on their own. Once you've connected a project folder, Grasppy now finds the grasppy-exports/ folder inside it without asking you to pick a second time.
• Two fixes. Keywords chosen in SEO research now appear in Target Keywords the moment you click Use in post, instead of after a reload. And blog check-ins can now be reviewed before they replace a post, with a one-click restore of the previous version for 30 days.

New Workspace Exports Appear Automatically

Claude Code and Codex exports from a new workspace now connect that project folder to Grasppy's local sync helper automatically, so the conversation is ready to select in Automation without another setup step.

• No more manual watched-folder updates. The first chat export from a workspace automatically adds its project folder to the local sync helper.
• Works for both Claude Code and Codex. The same automatic registration runs before either exporter creates its JSONL conversation file.
• Built into every new workspace. Grasppy includes the registration support whenever it prepares a new local project folder.
• Safe to run repeatedly. Folders already connected remain unchanged, with no duplicate entries or extra configuration.

Automation Now Sees Every Project Folder — And Tells You When It Can't

The Add-chat dialog now knows exactly which local folders your sync helper watches. When a workspace's folder isn't one of them, Grasppy says so plainly — and one click fixes it.

• No more mystery empty dropdowns. If your exports live in a folder the local sync helper isn't watching, the export list used to come up empty with no explanation. The dialog now names the exact folder that isn't connected instead of leaving you guessing.
• Connect a folder with one click. A new Select folder… button lets you pick the project folder right in the dialog. The sync helper starts watching it automatically within seconds — no config files, no restarts.
• Manage watched folders from the terminal too. New list-roots, add-root, and remove-root commands show and change which project folders are scanned for exports.
• Changes apply instantly. The sync helper now picks up folder changes while running — add a folder and its exports appear in the dropdown moments later.
• Safe by design. Adding or removing a watched folder never touches your files — it only changes where Grasppy looks for chat exports.

Skills Scans Are Fast Again — Even With Big Third-Party Skills Installed

Scanning your skills for drift is back to a couple of seconds, no matter how large your installed skills grow — and repeat scans are now near-instant.

• Fast scans with code-heavy skills. Third-party skills that ship large code bundles were slowing the skills scan from seconds to about a minute. The scan now checks only the files Grasppy manages, so extra skill content no longer drags it down.
• Near-instant repeat scans. Grasppy now remembers what it saw on the last scan and skips files that haven't changed, so running a second scan finishes almost immediately.
• Your skills stay untouched. Scanning remains strictly read-only — Grasppy never opens, modifies, or removes files inside skills it doesn't manage, including anything a third-party skill installs.
• Nothing changes in your skill list. All your skills still appear with the same names, descriptions, and sync behavior as before.

Automation Gets Its Own Tab — Sync Every Chat From One Place

Automation moves out from inside a single document into its own tab beside Analyses, so you can auto-sync your Claude Code and Codex chats across every document from one place — with built-in guidance and a more dependable connection.

• Automation now lives next to Analyses. Instead of being tucked inside one document, Automation is a top-level tab that shows every synced chat across your whole workspace, with a Document column so you can see where each one lands.
• Pick the document when you add a chat. The Add-chat window now lets you choose which document an automation syncs into right as you set it up — no need to open the document first.
• Built-in help for exporting a chat. A new help icon in the Add-chat window walks you through creating the export step by step, and a full Automation page guide is available from the on-page help tab.
• A safer, more reliable daemon connection. "Copy daemon token" now simply copies your current token — clicking it can no longer disconnect a running sync by accident. A separate, clearly marked Regenerate option handles deliberate token changes, with correct setup instructions.
• Cleaner help panel. The tutorial guide now opens crisply on top of dialogs instead of appearing dimmed behind them.

Meet Ask: Searchable Conversations, a Clearer Menu, and Unified Artifacts

Chat is now Ask — with conversations you can search, pin, and rename. Analyses gets its own spot in the menu, and every artifact you save now lives in one place.

• Chat is now Ask. The workspace Q&A is renamed Ask everywhere — the main menu and the tab inside each document — because that's what it does: ask questions across everything you've captured, with answers that cite their sources.
• Find any past conversation. The Ask conversations panel now searches across both titles and message text, lets you pin favorites to the top, rename conversations inline, and opens from a clearer labeled button so your history is easy to reach.
• Analyses is back on the main menu. Your analyses — the heart of Grasppy — now have their own top-level spot instead of being tucked inside the Library. Library keeps Documents and Artifacts together as two tabs.
• Everything you save shows up in Artifacts. Artifacts you save from Ask now appear in the unified Artifacts view alongside the rest, always newest-first, so nothing stays hidden. Click any one to preview it in place.
• Cleaner artifact reading. The artifact preview drops the grey code background and moves its actions (copy, export, delete) neatly into the top bar. Conversations and Artifacts panels now share one consistent, roomier design.
• Faster marketing pages. The live-chat widget now loads only after the page is ready, so grasppy.com pages become interactive noticeably sooner.

Semantic Map Sync, Bulk Plan Moves, and More Reliable Local Workflows

The Semantic Map is easier to keep current and search, plan content moves in bulk, and local sync, chat history, and agent exports are more dependable.

• A map that stays current. Sync Now re-syncs the Context Vault mirror with live progress, refreshes the graph automatically, and shows when the map is becoming stale.
• AI Librarian and clearer search. Copy the vault navigation prompt directly from the Semantic Map, then use more visible graph, topic, and connector searches to find what you need.
• Move plan content in bulk. Select multiple sections or chapters in a plan and move them to a new home in one operation instead of one at a time.
• Cleaner local-agent workflows. Generated coding plans now require a Lean Build Check, while Claude Code and Codex exports avoid turning tool-written project files into duplicate chat artifacts.
• Local mirrors that repair themselves. Every Sync button now uses the same verified path, detects missing folders and incomplete writes, keeps each workspace isolated, and rebuilds affected content instead of silently accepting a partial mirror.
• Conversation history that stays visible. Workspace Chat now restores every saved conversation after navigation or reload, including chats created before this fix.

Live Support Across Grasppy

Help is now available without leaving the page you are working on.

• Chat from the public website. Open the Grasppy support messenger from product, pricing, extension, comparison, and release-note pages.
• Chat inside the app. The same support channel is available after sign-in, so you can ask for help while working in your workspace.

A Fairer Trial Start and Clearer Workspace Limits

New accounts receive their full trial when they actually enter Grasppy, with workspace limits aligned to the selected plan.

• Your trial starts when you arrive. The 14-day Pro trial now begins when your first workspace is created at sign-in, including when you return after an interrupted signup.
• Your first workspace is always available. Every account can create its initial workspace; additional workspaces are reserved for Team plans.
• Pricing is easier to understand. Outdated Founding Member references were removed from public and account surfaces.

Your Workspace Is Ready the Moment You Sign In

New accounts no longer go through a workspace setup step — Grasppy now creates your workspace for you, automatically.

• Automatic workspace creation. Sign up, sign in, and your workspace is already there — named after you and pre-loaded with your starter projects (Skills, Prompts, Rules, Memory, Plans, Docs, and more).
• Setup lock resolved. Some recent signups saw a "Workspace setup is locked" screen and couldn't get started. That's fixed — affected accounts are unblocked automatically the next time they sign in, with their full trial intact. We're sorry for the wait.
• One workspace per account. Every account gets exactly one workspace, keeping your projects, documents, and vault in a single predictable home.

A Simpler Menu: Library, Workspace Chat, and the Semantic Map

The main navigation is rebuilt around how you actually work — browse everything in Library, ask anything in Chat, and see it all connected on the Semantic Map.

• Library brings your content together. Analyses, Documents, and Artifacts now live on one page as three tabs instead of three menu items. The page remembers your last tab, and old bookmarks redirect to the right place.
• Workspace Chat — ask across everything. A new Chat page runs the same AI chat you know from documents, scoped to your entire workspace. Its conversations keep their own thread history, answers cite the exact source turns, and Deep Search, web fallback, and Save as artifact all work as they do in Document Chat — which is unchanged.
• Assign multiple projects to a chat. The Assigned Project picker now has checkmarks — attach sections from several projects at once, each keeping its own selection, with a combined size estimate before you send.
• Attach your codebase map. A pinned "My Code (codebase)" entry in the same picker lets you attach the generated code-map docs — architecture, dependencies, per-module summaries — straight from your local vault. Everything stays on your machine until you send.
• The Semantic Map is now its own page. The visual graph of turns, artifacts, entities, and topics moved out of Context Vault into the main menu. Context Vault keeps its configuration role and now sits next to Settings.
• Clearer names and truer numbers. Workspace "plans" are now called projects everywhere, and the dashboard's Artifacts count shows your real artifact total.

Multi-Workspace Vault Protection and Faster Document Search

A safety guard for accounts with multiple workspaces, a much faster search experience in large documentation projects, and truer activity stats on the dashboard.

• Each workspace stays in its own vault. In multi-workspace accounts, syncing one workspace can no longer write into another workspace's grasppy-exports folder. If a workspace's folder link goes stale, it now self-corrects on the next sync instead of touching the wrong vault.
• Search that keeps up with big projects. Typing in the Plan search box no longer lags on large projects. Type your query, press Enter, and results appear — the page stays responsive throughout.
• Content search across the whole project. Content mode now searches every section on the server instead of downloading the full project into the browser first, filtering both the table of contents and the reading pane to matching sections.
• Activity Calendar counts automated imports. Chats brought in by scheduled automation now show up as activity on the dashboard calendar and trend charts — days when your automation worked no longer look empty.
• Faster homepage. grasppy.com renders its first screen noticeably sooner on slow connections.

ChatGPT Exports Capture the Full Conversation Again

The browser extension now reliably exports your entire ChatGPT conversation — in the right order, including the final response — after a ChatGPT-side change had started truncating and reordering long chats.

• Full conversation, correct order. Long ChatGPT chats export completely, with messages in sequence and the final assistant response included — no more scrambled or cut-off exports.
• Root cause fixed. ChatGPT changed how its conversation data is accessed; the extension now uses the updated method to pull the complete thread instead of scraping the visible, lazy-loaded page.
• Honest completeness check. If only a partial thread can be read, the export is clearly marked as partial rather than silently shipping an incomplete file.
• Update the extension. Reload the Grasppy browser extension (version 2.8.15) to get the fix.

My Code: A More Reliable Handoff to Your AI Agents

A reliability update to the code map your AI agents read. The machine-readable map now stays internally consistent on every project layout, and the instructions it hands your agents match the data exactly — so an agent asking "what depends on this file?" gets a trustworthy answer instead of a dead end.

• Consistent lookups everywhere. Every file in the map now reliably resolves to its full dependency record — including deeply nested folders and non-standard project layouts that could previously leave an agent empty-handed.
• Accurate agent instructions. The copy-paste guidance that teaches an AI how to read the map now matches the real data format, with no references to steps that don't apply to your repo.
• Self-checking on every scan. The map now verifies its own internal consistency each time it's built, so this class of issue is caught immediately instead of slipping through silently.

Your Context Workspace, Color-Coded

The Context System diagram now gives every project its own color identity, so new and returning users can tell at a glance which cards are their projects and where a click will take them. The Context Vault page also caught up with what the vault can actually do.

• Color identity for every project. My Rules, My Memory, My Skills, My Prompts, My Research, My Plans, My Documentation, and My Journal each carry a unique color — tinted card, matching border, and solid icon — so your projects stand out clearly from the explanatory parts of the diagram.
• My Code looks like what it is. The code card is now styled as a dark editor window, making it obvious that your code lives in your editor, not inside Grasppy.
• Clickability is visible. Every navigable card shows a persistent "Open ›" hint instead of revealing it only on hover.
• AI Librarian, showcased. The Context Vault page now explains the AI Librarian navigation protocol — one copy-paste prompt teaches any AI to search your vault like a library, reading manifests and short summaries first and opening only the sources it needs.
• See it before you use it. New expandable screenshots of the AI Librarian console and the semantic map, plus a refreshed Context System illustration on the homepage.

My Code Understands Your Whole Stack — and Keeps Itself Current

The code map now sees the connections that imports can't: which frontend calls hit which backend routes, which files touch which database tables, and which files are high-stakes regardless of size. It also maintains itself — the map detects when your code has moved past it and refreshes in the background, no clicks needed.

Your Code Map Now Shows What Breaks Before You Change It

My Code grows from an architecture overview into a change-impact engine: see every file that depends on the one you're about to touch, hand your AI assistant a ready-made impact brief, and stop the "improved one thing, broke another" cycle.

Document Chat Now Searches Everything You Have — and the Web When It Has To

Ask a question and Grasppy searches across all your documents, finds the answer even when your wording differs from your notes, and — only when your own content comes up short — offers to check the web. Great answers can be saved as artifacts in one step.

The Files Your AI Creates Now Become Versioned Artifacts

Export a Codex or Claude Code session and the files your assistant actually wrote — HTML mockups, JSON, configs, and more — now arrive as artifacts with full version history. Plus clearer diagrams and more reliable exports.

Use Any AI Provider — One Key Runs Everything

Connect just one AI provider — Anthropic, OpenAI, or OpenRouter — and every Grasppy feature works on it. No more needing a specific provider for certain features.

Each Workspace Sees Its Own Chat Exports

When you set up automated sync, the export picker now shows only the chat exports from the workspace you’re in — not a mixed list from every project.

Local AI That Doesn’t Get Stuck

Analyses no longer stall when your local AI hiccups — they finish on the cloud automatically (when Cloud Assist is on), recover dropped connections on their own, and show a clear status instead of an endless spinner.

Blog FAQ Panel and Smarter AI Tool Routing

Blog posts gain a built-in FAQ tool, the AI behind blog writing now follows your selected AI provider, and post word counts are accurate again.

Clearer Release Notes Workflow

Grasppy release notes now follow a stricter changelog format, so product updates stay separated, readable, and easier to scan over time.

Onboarding and First-Use Guidance

Grasppy onboarding now walks new workspaces from folder setup into the first real capture workflow, with clearer help and friendlier workspace pages.

• Onboarding is split into clear phases. Workspace foundation and first-use setup are now shown separately, so new users can finish the required local setup and then learn how to start using Grasppy.
• First-use guidance now covers the Chrome extension and API key setup. The onboarding flow points users toward the Chrome Web Store, Settings, importing a first chat, and reviewing the first analysis.
• Tutorial videos can be attached to onboarding steps. When a tutorial chip has a video URL, Grasppy shows the video with that step; steps without videos stay clean and uncluttered.
• Documents, analyses, artifacts, and workspace pages feel more consistent. Page headers, document cards, and list layouts were polished so the main workspace areas line up more predictably.
• The Chrome extension link now points to the live store listing. Public extension pages and onboarding links send users directly to the approved Grasppy Chrome extension.

Context Vault and Review Polish

This update makes Context Vault evidence navigation and check-in reviews easier to trust when working with local documentation.

• Context Vault source opening is more reliable. Semantic Map source buttons now open the right local evidence instead of landing on a blank review page.
• New check-in sections are easier to review. Brand-new sections clearly show that there is no previous content on the left while proposed content appears on the right.
• Semantic Map technical documentation was expanded. The system now has a clearer technical reference for how graph nodes, source files, and evidence routing fit together.

Guided Workspace Setup

This release gives every new workspace a guided setup path, so you can connect your local folders, confirm Grasppy is ready, and start working without guessing what to do next.

• Guided onboarding is now built into Grasppy. New workspaces show a clear setup checklist until the workspace is ready.
• Project Root and Data Vault setup are easier to verify. Grasppy now shows whether the selected root folder, export vault, and local mirror are connected.
• Workspace setup can repair common starter issues. If starter projects or local paths need attention, Grasppy can fix them from the onboarding screen.
• Secondary workspaces are properly isolated. Each workspace gets its own starter structure under the folder you selected.
• Claude and Codex local sync is part of onboarding. The setup flow prepares project-local folders for AI tools without asking you to type long system paths.
• Step-by-step help is available inside the flow. Each onboarding step includes detailed instructions that can be expanded when you need them.

Chrome Extension and Release Notes Page

Grasppy now has an approved Chrome extension and a public place where customers can follow product updates.

• The Grasppy Chrome extension is available on the Chrome Web Store. It helps capture supported AI conversations from the browser into Grasppy.
• Release notes now have a public page. Customers can review recent Grasppy updates without needing access to internal project notes.
• The release-notes page is connected to the public site. It has clean routing, sitemap support, and a footer link under Account.
• Updates are shown newest first. Each release stays as its own card, with older updates preserved below newer ones.