Changelog.

Edit tags now opens the tag editor directly from a file on your canvas. It shows the file’s current tags, ready for you to change, without asking you to run the automatic organizer first.

Search shows matches as they arrive, so you can choose a file while the rest are still loading. Narrow by type, move through results with your keyboard, and keep your selection as more details appear. If part of a search fails, the results already found stay available.

Publish a Brain as a page people can open in their browser. Add a title and introduction, choose anyone with the link or a list of people, and decide whether to offer original file downloads. Readers get the content you selected, with file previews built in.

Start a published workflow, follow its progress, and inspect the result from the command line. If your connection drops, return to the same run to see where it stands. Retrying the same start request recovers that run instead of starting the work twice.

Before importing a website from the command line, preview which pages will be kept, which will be left out, and the estimated cost. Duplicate addresses and links that aren’t web pages are left out. The preview adds nothing to your project.

Open a type or relationship with one click anywhere on its row, or with Enter or Space. Both lists stay open together, and relationship details use the names in your Shape, so you can follow the model without translating it as you go.

The Penumbra command line now runs on Linux, on both Intel and ARM machines. It installs as a single executable, with no Node.js setup required, so you can use Penumbra from your workstation or a remote machine.

Load more records into the graph and keep exploring. Counts now reflect what you can actually see, connections stay with their records as the graph grows, and labels make better use of the space as you zoom.

Install the Penumbra command line directly from PowerShell on Windows x64. Sign in through your browser, work with your projects and files, and update the installed command in place. There’s no Node.js or Linux environment to set up.

Give Shape generation reference files or a collection from your project, alongside your brief. The command line accepts references by name, so a report, interview set, or working document can help define the knowledge you want to capture.

Search across your records with a question or phrase and get a more relevant ordering of results. Penumbra now considers how well each candidate answers the query, while exact keyword search keeps its familiar ordering.

The starter library is better at capturing useful findings even when a source leaves something unsaid. Research, market intelligence, AI planning, and other Shapes now distinguish missing optional detail from a reason to discard the whole finding.

A generated or refined Shape that needs corrections now stays available as a draft, with the problems called out. You can work from what was produced instead of losing the draft and starting again.

Run penumbra on its own to see where you are and what to do next. The welcome screen shows your selected project, or a short list of projects with the commands to choose one. Help is still there when you need the full reference.

The Decision Log Shape now captures a decision question whether it is still open or has already been answered. Link it to the decision that resolved it, so the reasoning includes the question as well as the conclusion.

Command-line and agent search can return whole files as well as individual passages. Short summaries help you decide which file to open, while passage matches point you toward the relevant page or section. Narrow the search to a collection when you already know where to look.

Use the microphone in Add Files to dictate a note for an upload. Capture the context while it’s fresh, then use the resulting text as the note attached to that batch.

Add a note to a group of files as you upload them. The organizer can use that shared context when proposing tags, so you can explain what the files belong to once instead of repeating it for each one.

Long documents no longer lose their page images partway through because of the previous preview-size limit. Newly processed files can keep those later pages available for visual search, and existing pages without images no longer get pushed down the results for that reason.

Open a citation in a PDF or slide deck and see the quoted passage highlighted on the page. When Penumbra can’t confidently locate the quote, it leaves the page unmarked. Older files may need to be reprocessed before their citations can highlight.

Search can now find relevant rendered pages in PDFs, presentations, and Word documents using what’s visible on the page as well as its text. That gives diagrams, charts, and screenshots another way to surface when their meaning isn’t captured by a caption.

Scanned PDF pages get a more careful reading of their text and numbers. Table rows that wrap onto another line are kept together more reliably, so the extracted text is easier to work with. These improvements apply when a file is added, replaced, or reprocessed.

An interrupted submission can resume from the work you already approved. You don’t have to repeat the review, and retrying the submission won’t save the same findings twice.

The Files list now uses the available height of the page. You can see more of your library at once instead of scrolling through a short panel with empty space underneath.

Wordcraft is now available through Penumbra Labs. The open-source editor lets you and an agent contribute to the same Markdown document, save it to your project, and reopen it later. If a connection drops during a save, your draft stays available to retry.

Text inside pictures in a Word document now joins the surrounding paragraphs in the extracted content. Screenshots and embedded diagrams can contribute to search and analysis, instead of disappearing between the text. Small decorative images are left out.

Search is more willing to say it didn’t find a useful match. Loosely related results are filtered out, while exact-word matches still have a way through. If nothing fits, you can try a more specific phrase instead of sorting through a page of near misses.

A maintained document can build a running log of observations from source material, then use that log to update a focused working document. Citations connect the written account back to its sources, so you can follow how a conclusion developed.

Connected agents can propose changes to a native document and wait for your approval. Review the replacement text, approve it, and keep working from the saved version. Reopening the approval returns the same result rather than applying the edit again.

Connect a Granola folder to bring meeting notes and transcripts into Penumbra as searchable files. Scheduled sync picks up the source again without duplicating unchanged notes. Disconnecting the source leaves the files already imported in your project.

Browse the Shape Marketplace, preview what a Shape contains, and add it to your project. Start with a published model or customize an editable copy. Linked Shapes have an update preference, so you can choose whether to follow changes from their publisher.

Ask the project assistant about the records and relationships in your knowledge graph. It can also propose changes for you to review and approve before they are saved, keeping the conversation connected to the work in your project.

Your Shape can become a typed interface for working with its records and relationships in TypeScript. Build queries using the concepts you defined, with generated types that help you see which properties and connections are available.

Bring files and records onto a canvas and arrange them around the question you’re working on. Keep useful material together in a spatial view you can return to as your thinking develops.

Give a file its original source date, even when you added it to Penumbra much later. The file editor keeps that date separate from its arrival in your project, and record details make source and due dates easier to inspect.

An extracted relationship can carry its own source passage, just like a record. See where a connection came from and read the supporting quote alongside the records it joins.

Extraction estimates the cost of a run before the work begins. You can weigh the files and effort you’ve selected against the estimate, rather than starting without a sense of the spend.

Trace a file version to the records and relationships that cite it, including findings still waiting in a review. Connected tools can follow that trail from the source to the knowledge built from it.

Send feedback from the App and attach a screenshot when it helps explain the problem. Preview, replace, or remove the image before sending. You can also submit feedback from the command line or through a connected assistant.

Choose Fast or Standard extraction for the files and knowledge you’re working with. Stop a run when you need to, then resume completed portions without starting everything again. Findings and proposed connections still arrive for review before they enter your graph.

A Brain gives a selected part of your project a home of its own. Choose the content, decide which tools an agent can use with it, and manage access without handing over the whole project. Its activity view helps you see how that knowledge is being used.

Open a PDF in Penumbra with page thumbnails, zoom controls, and page navigation. Move through the original document while keeping it close to the rest of your work.

Work through extracted findings in a focused review queue. Edit a finding where you see it, remove one with an option to undo, and add or connect records as you go. After confirming a review, move on to the next item.

Browse published Shapes and drafts in one library, then look across their types and relationships. Search, filter, and sort the lists to find a concept or inspect its properties without opening every Shape in turn.

Start a command-line login, finish signing in in your browser, and return to your terminal ready to work. Your browser session completes the connection without a manual key-copying step.

Create a View that selects files, collections, Shapes, or types of records. Limit a client, connected agent, or read-only key to that View, and inspect what it has read from the Access page. The rest of your project stays outside that connection’s view.

Give an assistant a focused Shape-authoring Kit to help you start, edit, validate, or fork a Shape. Publishing remains a step you approve, so you can develop the model in conversation without every draft becoming the version your project uses.

Describe the knowledge you want to organize and generate a first draft of a Shape. Review its types, properties, and relationships, then refine the draft before publishing it for use in your project.

Word documents and PowerPoint presentations now get rendered page previews and recognizable covers. Browse the document’s layout alongside its extracted text, including files whose useful content lives in a header, footer, or image.

Describe how your team names and groups files, then turn that guidance into an organizer setup. Review the draft, refine one part, or edit it directly before publishing. Writing the rules doesn’t rename your files; proposed changes still come separately.

The notification bell brings completed work and problems back to your attention. See when findings are ready for review, when a run found nothing new, or when something failed, with a clear distinction between each outcome.

Choose a set of Penumbra tools and save it as a Kit for a connected agent. Keep each connection focused on the work it should do, and review changes that require your approval before they take effect.

Use your Google account to sign in to Penumbra and reach your organizations. One less password between you and your work.

Use a Shape to draw records and relationships from your files. Findings keep the source passages that support them and arrive in a review workspace, so you can inspect what was found before adding it to your knowledge graph.