Symbol Docs
Connect to AI

What Your AI Can Do

A plain-English guide to every tool Symbol gives your AI assistant.

When you connect Symbol to an AI like Claude, your AI gains a set of tools it can use on your behalf. You never call these tools yourself. Instead, you just talk to your AI naturally: "save this," "search for Docker issues," "load my Solutions." The AI picks the right tool and runs it for you.

Here is what each tool does.

Finding your knowledge

List capsules: Browse all your capsules, or narrow the list by type, tag, category, project, whether they are Skills or still unfinished, and by when they were created, last updated, or archived (a rolling window like the last eight hours, a named period like this week resolved in your timezone, or an exact date range), and choose the order. Within one type you can also filter, order, and total by the fields on your capsules (for example deals in the "won" stage, sorted by amount, with the pipeline summed), including capsules whose field is any of several values (for example deals in either the "won" or the "lead" stage) in one request. Filtering on a reference field matches by the capsule the value points to, so "grants whose action is @menu/change" finds every grant pointing at that capsule however the reference was written, and still works after its alias is renamed. Your AI can also ask for only the titles, summaries or a few field values of each capsule instead of the whole capsule, so a quick look across a large type stays small and fits in one reply. Ask for a choice that none of your capsules use and is not one of the field's options, and you are told the valid choices instead of getting an empty list. When a field holds a scope written as a pattern, you can ask which records apply to one thing: a rule stored as GravityKit/* is found by asking about GravityKit/GravityView, with the most specific rule first. It can also list what other people shared with you by name instead of your own library. Each row carries only what your AI can use: a flag that is off or a detail that is empty is left out, so a row without is_encrypted is not encrypted, and a listing of one type names the type once instead of on every row. Attached files are listed by their details (name, kind, size), never by where they are stored. Try asking your AI: "Show me everything in my shopping list from the last eight hours," "What is my open pipeline?" or "What has been shared with me?"

Search: Search across your entire knowledge base by keyword and type, and narrow by when capsules were created or last updated using the same words the rest of the app uses (a rolling window like the last eight hours, a named period like this week resolved in your timezone, or an exact range). Each result names its capsule once, by reference like @docker/3, and is marked encrypted only when it is. Try: "Search for anything about authentication errors from this week."

Open a capsule: Open a specific capsule by its reference, like @docker/3. Your AI does this automatically when you mention a capsule by name or reference. An archived capsule opens too, marked as archived, with a note that it can be restored from Archive in the web app. The reply carries the capsule's text once, which keeps reads small: the stored body is in content_md, the text you would read is in content.text only when it differs from the stored body (a template filled in with field values), and field values are in content.fields.values. Replies no longer include content_rendered, or a flat field_values that repeats content.fields.values, so a tool that read those reads content.text and content.fields.values instead. A capsule whose values sit outside its type's fields still sends them in field_values, as their only copy. A capsule whose type computes some fields from others (derived fields) also sends field_values, holding only the values that can be written back. Loading a type, loading a project and resolving several references send each capsule's text once in the same way. The reply also carries the capsule's own summary, so when your AI changes the text it can see what the summary says and send it back as it is, or rewrite it, without asking you for it. When a capsule brings in others it links to, each linked capsule's summary also comes once, however many of them link to it. Each attached file comes back with only what your AI can use: its id (to rotate or delete it), name, kind, size, an image's width and height, whether it is gone and why, when it expires, when it was added, embed_markdown (the form to put it in a capsule's text, safe to keep) and download_url (a link to fetch the file now, which expires and must never be saved). Reading a shared capsule returns its files the same way. Warnings and notes about a reply (that a capsule is archived, out of date or not referenced anywhere, say) lead its notices, inside the reply itself, so every AI reads them, including one that reads only a tool's structured result. A PDF's text comes in its attachment's file_text and each page with no text as an image named in notices; the PDF file itself comes too only when your AI asks for it with include_original_pdf, or when its text and pages could not be read. A log that belongs to a capsule names that capsule in log.owner_ref, and the capsule names its log in log.owned_log_ref and carries its recent entries. Each is shown only to someone who may see it: a teammate who cannot see the capsule is not told which capsule the log belongs to, and one who cannot see the log gets no entries and no log when opening the capsule.

Read a skill's files: A skill can carry files next to its instructions, such as scripts, data, templates and images. Opening the skill lists them, each with its path, size and whether it is executable, but never their contents. Your AI then reads a text file a page at a time, every page fetched fresh with your access checked again, so it never pieces a file together from two different versions: if the file changes while it is reading, it starts again from the new version. A binary file, such as an image, is described by its size and digest rather than shown. A file that was deleted, or that you cannot see, is not found. Someone who can edit the skill can also ask for a file's history, the list of its earlier versions, and read any of them; anyone else is told that history needs edit access. Symbol Bridge reads skill files from the server too, never from its local copy, so it needs a connection to read them. Try: "Open @skills/deploy and read its scripts/run.sh."

Load a full type: Pull a type's capsules into the conversation in bulk, so your AI has the picture before you start. You can narrow to a creation window (a rolling span, a named period in your timezone, or an exact range). Large types arrive a page at a time. When the type has an output template, your AI sees it too, so it knows that a capsule with no body of its own shows the template, and each capsule arrives with its body as you would read it, template and field values filled in. For a quick overview your AI can load just the titles and summaries, without each capsule's full text, or, cheapest of all, just the titles, references and dates, with no summaries. The type arrives with the limits your AI writes within (the longest body, the most field content per write, the longest guidance, each in characters) and one sentence saying which of its settings your AI may change, without display details such as its icon or color. Try: "Load all my Docker fixes before we begin."

A type's guidance, once: The first capsule your AI opens from a type brings that type's guidance with it. While your AI still has the guidance, the next capsules it opens from the same type come without it, and each one says the guidance was left out and how to read it in full. To show it still has the guidance, your AI quotes a few words of it that Symbol asks for; when it cannot, because its conversation was shortened or ten minutes have passed since the guidance last arrived, the guidance simply comes again with the next capsule. When the guidance is edited, the next capsule brings the new version. Saving to the type asks the same of your AI: it must show it still holds the guidance, not only that it received it once. Short guidance always comes in full.

Load a project: Load a project's capsules across every type in one go, and narrow by when they were created or last updated the same way the main listing can. Again a page at a time if there are many, and again with the choice of titles and summaries only, or just titles, references and dates. Useful when a project spans multiple categories.

Search public content: Search what creators have published on Discover, not your own library. Narrow by category or tag, and sort by newest or most popular, exactly as the Discover page does: the same search returns the same entries in the same order, each with its public page link, its creator and its like, save and view counts. Only content its owner made Public appears, never link-only, team or private content. Try: "Find popular public prompts about cover letters."

Resolve several references at once: When one message mentions two or more capsules or types, your AI fetches them all in a single step instead of one at a time. A mentioned type brings up to 50 of its capsules, fitted to one reply; your AI can ask for fewer, and asking for more than 50 still returns 50. Try: "Compare @clients/1 and @clients/2."

Saving and editing

Save a capsule: Save a new capsule to your knowledge base at any point in a conversation. Just say "save this" or "add this to my Solutions." When your AI writes a prompt with blanks like {{platform}}, it can describe each blank and mark the ones the prompt cannot run without, so anyone whose assistant uses the prompt later is asked the right question. When the type has an output template and your AI sends no body, the capsule shows the template filled in with its field values. An empty body on such a type is refused rather than saved, so a capsule never loses its template by accident; to keep a capsule deliberately blank, use "Start blank" in the web app, or send authored_blank: true through the API. To show an attached picture or file in the body, your AI writes its file name, like ![chart](chart.png), in the same request, or pastes the embed_markdown from the upload reply; Symbol stores a link that keeps working for everyone who can read the capsule, including on share pages. The image_url and download_url in a reply expire within an hour, so a body never stores them: Symbol rewrites your own capsule's links and refuses another capsule's. On an encrypted capsule the body is never rewritten, so your AI pastes the embed_markdown itself. An alias that belonged to a deleted capsule stays reserved: a capsule saved with it is saved without the alias, and your AI says so. Ask it to reclaim the alias to reuse it on purpose. Mentions written before the deletion stay unresolved, so they never point to the new capsule. Before it writes to a type that has guidance, your AI reads that guidance in the same conversation and sends the proof it got back with every write. A read in another conversation does not count, so a new conversation reads the guidance again before its first write.

Rewrite a capsule: Replace the title or the body of an existing capsule. Try: "Update @docker/3 with this new approach." When the type has an output template, your AI can also hand the capsule back to it: the capsule drops its own body and shows the template again, filled in with its field values. Try: "Put @recipes/4 back on its template." An empty body on such a type is refused rather than saved, with a message pointing at the template instead. It also changes who can see a capsule: Private (every link stops working), Anyone with link (you get its link), or Public (listed on your profile, and you get its address). If publishing needs a step first, your AI says which: turning your public profile on, which it can do in the same step once you agree, or claiming a username, a one-time step in Settings. Try: "Make @docker/3 public." A capsule saved before its required fields were known completes itself once an update fills in every required field, and your AI tells you so; from then on, an update that leaves a required field empty is refused instead of quietly dropping the value. Giving a capsule an alias that belonged to a deleted capsule is refused until you ask your AI to reclaim it. An archived capsule can be read but not changed; your AI says so and that you can restore it from Archive in the web app. The text of an end-to-end encrypted capsule can be changed only by someone who holds its key: anyone else who asks gets a refusal and nothing is saved, so your AI sends the other changes without the text.

It can also publish a capsule as a skill, or stop publishing it, the same as the skill control on the capsule page. Your AI can give the skill a name you choose, and a chosen name stays the same when you rename the capsule's alias, move it to another type or rename its type, so the skill your assistants load is never renamed under you. Without a chosen name the skill is named after its type and alias, as in the web app. Publishing needs the right to share the capsule and skills turned on for the workspace; if something is missing, your AI says what. A published skill whose name or content has fallen behind the capsule is pointed out when your AI reads the capsule, with the name it would change to. When your AI edits or restores a published skill, or edits many at once, its reply says whether the skill your assistants load has started updating, which finishes on its own within seconds, or why it has not (for example you asked it not to, or the publish limit was reached), so it never assumes the update went out. If an automatic update was refused afterwards, reading the capsule says why, and that it will not retry by itself. Try: "Publish @guides/onboarding as a skill called onboarding-guide."

Fix one part of a capsule: Change a single line or section without rewriting the whole thing, which keeps the rest of a long capsule exactly as it was. Try: "In @docker/3, change the port from 8080 to 3000."

Saving, rewriting and fixing a capsule each answer with a short confirmation rather than the whole capsule again: when it was saved (updated_at), any warnings, and whether a skill or agent still fits. That keeps a small edit to a long capsule small. Your AI can ask for the text back once with include_content, and the reply names what it left out in content_omitted. Each mention in the text still comes back with the capsule it points to, without that capsule's summary.

Delete a capsule: Permanently remove a capsule you no longer need. If other capsules mention it, the first attempt stops and tells you how many and which ones, because their mentions would stop pointing anywhere; nothing is deleted until you confirm. If the capsule owns a log, your AI first tells you which log and how many entries it holds, and deletes only after you confirm; a log that belongs to someone else is kept as its own capsule. The reply also says whether the capsule was deleted for good or moved to the trash.

Duplicate a capsule: Make a copy of a capsule, exactly like Duplicate in the web app. The copy keeps the original's attached files and pictures, and its image and file fields point at the copies, so a template full of attachments copies in one step. The copy counts toward your plan's capsule limit and storage like any new capsule. Try: "Duplicate my weekly-report template."

Create a type: Set up a new type (category) for organizing capsules. Your AI writes the guidance automatically based on the name and what you are saving. Guidance can mention up to five capsules, like @rules/1, and each one's content comes along whenever your AI loads the type. Each must be a capsule in the type's own workspace that the type's readers can see, and not encrypted; otherwise the type is not saved and the reply names each mention to fix. A mention marked as a pointer, like ~@rules/1, is checked the same way but brings only its title, not its content. To write an example mention without bringing it in, put it in inline code. It can also choose how the type's capsules look on their share page (article, prompt, or showcase); a type made for prompts shares as a prompt unless you say otherwise. When the type gets an output template, every {{field}} and section in it must name one of the type's fields, and every section must be closed; otherwise the type is not created and the reply names what to fix.

Update a type: Rename a type, change its address (slug), rewrite its guidance, or change its icon, colour, shared layout, and archiving rules. Only what you ask for changes; everything else stays as it was. Rewritten guidance follows the same rules for capsule mentions as a new type's. When the slug changes, links written with the old one keep working. Your AI can also rename a field: every capsule keeps its value under the new name, and the type's template and capsule bodies that show the field keep showing it. Changing how often a log starts a new dated capsule (daily, weekly, monthly or manual) stops first when the log already has dated capsules: your AI tells you how many there are and which capsule today's entries go to now and after the change, because the current period is split between the two. Existing capsules keep their names and entries, and you confirm before the change applies. When a new archiving rule would archive capsules right away, your AI tells you how many, and how many share links would pause, and applies it only after you confirm. A type can't be switched into a log here; your AI converts it instead (see Turn notes into a log under Logs). Try: "Change my Solutions guidance to always include the error message," or "Rename the company field on People to employer."

Delete a type: Remove a type you no longer want. If it still holds capsules, the first attempt stops and tells you how many, so nothing disappears by accident, and you decide what happens to the contents. It also stops if deleting the type would affect things elsewhere (capsules that mention the type, other types whose fields point at it, its collections, or its active share links) and lists them, so you can weigh that before confirming. If you choose to move the capsules into another type and that type has a public share link, Symbol pauses first and tells you the move would publish them, so nothing goes public by surprise; you confirm before it proceeds. Moving them also stops first when it changes the capsules themselves: capsules that would render blank or differently because they show this type's template, capsules that leave a required field of the new type empty (they are marked incomplete), and logs that detach. The capsules of a log type cannot be moved into another type. In an organization, a type that other members have capsules in can only be deleted by an owner or Admin: moving takes every member's capsules along, and deleting archives other members' capsules into General instead of deleting them, so they can be restored. A log type that other members have capsules in can't be deleted until they delete theirs.

Send a Decision (send_decision) (experimental): When your AI needs a decision from you, it can send a short structured message to your Inbox instead of a wall of text. Questions are capped at 20 words, and a choice question is answered with a tap. Requires the Experimental features toggle in Settings. Try: "Ask me in my Inbox."

Check for your answer (experimental): After sending you a Decision, your AI checks back to see whether you have replied yet, and reads your answers when you have. This is why an assistant can send a question, keep working, and pick your answer up later.

Withdraw a Decision (experimental): If a question goes unanswered and your AI decides to carry on without you, it retracts the Decision so it stops sitting in your Inbox waiting.

Sharing

Share a capsule, a type, a collection, or a project: Get a shareable link. Anyone with the link can view it without needing a Symbol account, and the link is the only way in: it is never listed or indexed. Asking again gives you the same link, so you never pile up links you have lost track of; ask for a new link when you want a separate one. Sharing a whole type or collection creates a browsable library page of the capsules you created in it that your workspace can already see (not encrypted, not archived); a Private or link-only capsule in an organization, or a capsule you keep out of library links, is never published. Sharing a project works the same way: its page lists the project's capsules that the project's owner created, grouped by type, and a project is shared by link only, never listed publicly. Name the project or give its id. Your AI tells you how many capsules the link publishes. You can set an expiry date or a maximum number of views, which always makes a new link. An expiry date must be in the future: one that has already passed is refused, and no link is created.

See who can see it: Ask who can see a capsule, type, collection or project, and which links reach it: whether it is Private, shared by link or Public, its public address, how many people it is shared with by name, and each live link with its views and expiry. For an archived capsule it also says how many of those people have no access until you restore it. Ask with no item to list every link you have made. Try: "Which of my links are still live?" or "Who can see @docker/3?"

Read a shared capsule: Pull a capsule someone shared with you into your AI, using the link or its token. Any signed-in holder of a working link can read it, and your AI gets the same content the link's page shows. When the link is yours and your AI is connected to the workspace the capsule lives in, your AI gets your full capsule instead: its type's guidance, every mention resolved with your own access, and who created it. Each read counts against the link's view limit if it has one, so your AI resolves it once and reuses the result. A link to a whole type, collection or project returns the library's name and page link: libraries can't be browsed through an AI yet. If you turned MCP access off for the organization that owns the content, the read is refused.

Revoke a share link: Turn off links you shared so they stop working. Paste the link, or name the item (a capsule, type, collection or project) and revoke every link to it, or one of them, with no URL needed. A public item's listing is not a link you revoke: to unpublish it, make it Anyone with link or Private. Try: "Revoke every link to @docker/3."

Share with specific people: Share a capsule, a type or a collection with named people by email, at view or edit, and see, change or remove who has access. Only those people can open it, from Shared with me. Content in an organization can be shared only with its members, and a viewer member can only view. Try: "Share @recipes/3 with sam@example.com so they can edit it." or "Who has access to my Prompts type?"

Collections

Collections group capsules inside a single type, so one broad type can hold several narrower shelves.

List collections: See the collections inside a type. Try: "What collections do I have in Prompts?"

Create, rename, or delete a collection: Set up a new grouping inside a type, rename one, or remove it. Removing a collection does not remove the capsules in it.

Add to a collection: Put one or more capsules into a collection. They must already be in the same type as the collection. If one capsule cannot be added, the rest still go in and the assistant tells you which one and why; a capsule already in the collection is not counted as added. An archived capsule, or a capsule you may not publish into a collection that has a share link, stops the whole add, and the reply names it. Name the collection by its name or slug together with its type, by its alias on its own, or by its id. Try: "Add my three newest books to Sci-Fi Classics."

Remove from a collection: Take a capsule out of a collection. The capsule stays in its type. If one capsule cannot be removed, the rest still are and the assistant names the one that was not; a capsule that was not in the collection is not counted as removed. You name the collection the same way as when adding.

Projects

List your projects: See every project you can access, with names, descriptions, and visibility. Try: "List my projects."

Create a project: Spin up a new project to group capsules across multiple types. You can keep it private (the default) or share it with an organization workspace so teammates can read or edit. Try: "Create a project called Onboarding 2026."

Update a project: Rename a project, change its description or color, or change who can see it. Try: "Rename my Onboarding project to Onboarding 2026 and set the color to teal."

Delete a project: Permanently remove a project. The capsules that were in it stay in your knowledge base. Only the project grouping is removed.

Add to a project: Add one or more existing capsules to a project to group them together. If one capsule cannot be added, the rest still go in and the assistant tells you which one and why; a capsule already in the project is not counted as added. An archived capsule, or a capsule you may not publish into a project that has a share link, stops the whole add, and the reply names it.

Remove from a project: Remove a capsule from a project. The capsule itself stays in your knowledge base. If one capsule cannot be removed, the rest still are and the assistant names the one that was not; a capsule that was not in the project is not counted as removed.

Move to another Type: Move one or more capsules into a different Type when they no longer belong where they are. The capsule keeps its identity, share links and attachments, and any links that pointed at it by its old reference keep working, so you do not have to hunt down and edit the capsules that mention it. If the Type you are moving into has a public share link, Symbol pauses first and tells you the move would publish those capsules on its public page, so private notes never go public by accident; you confirm before it proceeds. A move also tells you what else it changed: collections it dropped the capsule from, required fields of the new type the capsule leaves empty (it is marked incomplete until you fill them in), a log that detaches because the new type keeps no log (the log keeps its entries and reconnects if you move the capsule back), and other types whose fields can only point at the old type, with how many of their capsules cite it. It refuses a move that would mix up encryption, land in a log type, move a log itself, or attach the capsule to a different log type than the one its log lives in. To put capsules into a log type, convert them into entries instead (see Turn notes into a log under Logs). If one capsule in the batch cannot be found, the others still move and the assistant names the one it could not find.

It can also take your personal capsules into a team: name the organization and a Type in it, and your capsules move there, or copy there if you ask for a copy and want to keep the originals. Symbol first tells you who in the team will be able to read or edit them and which share links keep working, and waits for you to confirm. Old mentions of a moved capsule keep working for anyone who can read it in the team. Try: "Move my onboarding notes into the Engineering Type of Acme."

Archive capsules: Archive one capsule or up to 50 at once, the same archive as in the web app, when they no longer belong in your day-to-day lists. Ask for a dry run first: your AI tells you which capsules it would archive, which are already archived, which you can read but not change, and which it could not find, plus how many share links, public pages, published skills and embeds would pause, and how many people the capsules are shared with would lose access. An archived capsule stays readable by the same reference, so your AI can read it back to confirm, but it drops out of lists unless you ask for archived capsules, and it cannot be changed until it is restored. Its links, public page, published skill and embeds pause until then, the people it is shared with lose access (unless your whole organization can see it, since they are members), and a log it owns is archived with it. Anything that reads the capsule on a schedule will find it archived. Restoring is done from Archive in the web app; your AI cannot restore a capsule. Try: "Archive these three old drafts, but show me what would happen first."

Edit many capsules at once: Set or clear one field, or add, remove or replace tags, across up to 50 capsules in one go. Ask for a dry run first to see how many would change and which would be skipped. Try: "Set status to done on all my open tasks from last week."

Rename or merge a tag: Rename a tag across every capsule you can edit, or fold it into another tag you already use. Capsules you cannot edit keep their tags.

Import notes: Bring in a CSV file, a folder of Markdown notes (such as an Obsidian vault or a Notion Markdown export), or a Symbol export, as typed capsules. Your AI can preview the plan first: the columns it found, the fields they map to, and any rows that would fail. A reference field that points at a note imported later, even in another Type of the same export, is filled in once that note arrives; one whose note never arrives is listed at the end of the import, and the note it sits in is still imported. A note whose body carries an attachment access link copied from another capsule is not imported and is listed with the reason, while the rest of the notes import. Restoring a Symbol export keeps the values a note held for fields its Type no longer has: they come back hidden, as they were in your account. A reference that points at a note of another Type than its field accepts is left out with a note, rather than costing the whole note. The preview lists both before anything is imported. References to a Type you renamed or merged keep working, because an export writes each reference, and the Type each reference field accepts, at the address it has now. An export that could not tell which Type a reference field means is refused before anything is imported, naming the Type and the field (see Import and export). A Symbol export with several Types is always previewed first: the preview settles where each Type lands, so when one of its Types shares a name with one you already have, the imported copy gets its own address and every reference in the export points at the copy, never at yours. On the Free plan the preview also says how many notes fit in the capsules you have left; an import with no room left stops before it creates anything.

Import selected Symbol Types for approval: When any selected destination protects its content, your AI uploads the complete selection and saves it as one change. Nothing arrives before approval, including items going into unprotected Types in the same selection. The review shows the imported items linked to each other, their Type context and placements, every invalid row that will not be imported, and any placement that was omitted. Review every item and report page in the browser or Symbol app, then confirm the whole import. An import can include up to 1,000 items including new Types, with up to 64 MiB of uploaded input. For larger selections, import unprotected Types separately or select fewer Types. Your AI can resume or cancel an unfinished upload and report its status. It cannot approve an import. If no selected destination is protected, it directs your AI to continue with the ordinary import. See Protected Types for the review and size limits.

Load a project: See "Load a project" under Finding your knowledge.

When one capsule mentions another with a reference like @docker/3, that link is real and your AI can walk it.

What this capsule points to: List the capsules a given capsule references. Try: "What does @runbooks/2 link out to?"

What points at this capsule: List the capsules that reference a given one, so you can see what would be affected before you change it. Try: "What links to @standards/1?"

Map the neighbourhood: Follow the links outward and inward together, up to three hops, to see how a cluster of capsules connects. Try: "Show me everything connected to @onboarding/1."

See what a read costs: Measure what reading a capsule costs in characters, or a set of up to 20 read together: the same figures as Read cost on the capsule page, including each capsule it inlines, text a read returns twice, and every capsule a set delivers more than once. The figures are for your AI's own access, and a reference is looked up across all your workspaces unless your AI names one. Try: "What does reading @agents/penny cost together with the capsules it names?"

Find orphans: Find capsules in a type or project that nothing else references. A good way to spot notes that have drifted out of reach, and you can ignore ones you only just made so recent work does not clutter the list. Logs are never listed, because a log is reached through the capsule it belongs to, and neither is a type you set to hide orphan warnings. Try: "Which capsules in Prompts are orphaned, ignoring the last three days?"

Keeping linked capsules current

When a capsule changes, every capsule that links to it is flagged for review, so a page that relies on an old answer does not quietly go out of date. Your AI sees the flag on reads and lists, with each capsule that changed named, and can list just the capsules waiting for review.

Confirm still valid: After reading a flagged capsule in full and checking it against what changed, your AI can confirm it still holds, with a short note saying what it checked. The note is kept with the review. If the capsule needs changing instead, your AI updates it, which clears the flag too. Confirming needs a fresh full read, made after the change, so a capsule is never marked current from a summary or a listing, and it needs edit access to the capsule. Try: "Check the capsules flagged since I changed @pricing/plans and confirm the ones that still hold."

Turn review flags on or off: Switch dependency review off for your own account, or back on. Off hides the flags from you in every app and AI connection; your teammates keep seeing theirs, and turning it back on shows what is still waiting. Try: "Stop flagging capsules for review on my account."

Logs

Some types are logs: instead of one long document you keep appending dated entries.

Add a log entry: Append one timestamped entry to a log, in a single step that is safe even when something else is writing at the same time. Each call adds a new entry rather than replacing the last one. Logs that keep dated buckets get today's bucket created on the first entry of the period; logs attached to a specific capsule append straight to that capsule's log. Like a capsule write, an entry in a log type with guidance needs that guidance read in the same conversation.

Correct a log entry: Change an entry your AI appended earlier, using the reference it got back when it wrote the entry. The entry keeps its number. Try: "I logged 30 minutes on that, it was 20."

Delete a log entry: Remove one entry. The other entries keep their numbers and references, and the next entry takes the next number.

Read log entries: Read a log's entries newest first, a page at a time, and narrow to when they happened (a rolling span, a named period in your timezone, or an exact range). For a log that files into dated buckets you can read a whole period across its buckets at once, so asking a monthly log for "this week" returns just this week. Each entry carries its number, when it happened, its values and who wrote it; details that are empty, such as when it was last edited on an entry nobody edited, are left out. Try: "What did I log last week?"

Turn notes into a log: Turn a whole type of notes into a log, or move chosen notes into a log you already have, each note becoming a dated entry. You choose which date each entry is filed under: when the note was created, when it was last edited, or a date field it has. Your AI first shows you a preview, covering how the fields map, any values with nowhere to go, how many entries land in each period, which share links and access are withdrawn, and anything that stops the conversion with the reason, and converts only after you agree. Mentions of the old notes keep working and open their entries, and each entry keeps a link to its note's earlier versions. See Turning notes into a log. Try: "Turn my Journal type into a weekly log."

History and snapshots

Edits to a capsule, and to a type's guidance and summary, are kept as versions, and your personal workspace can be snapshotted. Your AI can read that history and put things back. Reading or restoring versions needs a plan that includes revision history.

List versions: See the saved versions of a capsule or a type, newest first, each with when it was made, whether it was an edit or a restore, and who made it. Versions saved before Symbol recorded who made each edit show no name. Versions you kept show their label, and your AI can list just those. When the history is full, your AI can tell you which version your next change will remove. Long histories arrive a page at a time. An end-to-end encrypted capsule keeps no history Symbol can read, and your AI says so. Try: "Show me the history of @docker/3."

Read or compare versions: Open one earlier version in full, or compare two to see exactly what changed: the title, the summary, a prompt's details and an agent's settings, which lines of the body were added or removed, and each field's value before and after. For a type it compares the guidance, the summary, the output template, the fields, and its name, address, icon and color. When you ask what your last few edits changed, your AI picks the versions from the history rather than guessing their numbers, so it still works after old versions have been removed. A body rewritten so heavily that a line-by-line comparison would be unreadable is reported with its line counts instead, and the other changes are still listed. Try: "What did my last three edits to @docker/3 change?"

Restore a version: Put a capsule or a type back the way it was at an earlier version. The restore is saved as a new version, so it can be undone too, and your AI tells you which version it restored from and the number of the new one. Restoring a type also brings back its name, address, icon and color when that version recorded them, and your AI tells you when its address changed. If restoring a type's fields would change capsules that hold data, your AI lists what is affected and asks you first. Try: "Undo the bad edit Sam made to @runbooks/2."

Keep a version: Keep a version you want to hold on to, such as the prompt that scored best, and give it a short label. A kept version is never removed by the history limit or the 30-day clean-up until you stop keeping it, and it does not use up a place in the 50 most recent versions. You can see up to 10 kept versions of one capsule or type at a time. Restoring a workspace snapshot replaces all history, kept versions included. See Keeping a version. Try: "Keep version 3 of @prompts/7 and label it best scoring prompt."

List snapshots: See the snapshots of your personal workspace, when each was taken, and whether it was scheduled, taken by hand, or the safety snapshot made before a restore. See Snapshots.

Take a snapshot: Snapshot your personal workspace right now, for example before a big reorganization. There is a daily limit on snapshots taken by hand. Try: "Take a snapshot before we start."

Restore a snapshot: Return your whole personal workspace to a snapshot. This replaces everything in it and starts every capsule and type on a fresh history, so Symbol always runs a dry run first: your AI shows you what would come back, what would be removed and what cannot be restored, and goes ahead only with the confirmation the dry run hands back. Whether your assistant app asks you before the restore itself runs depends on the app, so your AI should show you the summary and wait for your answer. A safety snapshot is taken first, so a restore can be undone, and you can restore once an hour.

Your account

List types: See all your types with their descriptions, how many capsules each holds, and when each type and its AI guidance last changed. Your AI can also list only the types that changed since a time you choose, in one workspace or across all of them, so checking which guidance changed this week is one small step. To read a type's AI guidance, your AI loads that one type instead. Try: "Which of my types had their guidance changed in the last week?"

List workspaces: See the workspaces you can reach, meaning your personal one plus every organization you belong to. Your AI uses this when you name a workspace, so it saves to the right place. Try: "Which workspaces do I have?"

Stats: Get a quick summary of one workspace: how many capsules it holds, broken down by type. The reply names the workspace it describes. Ask about an organization by name to see its numbers rather than your personal ones. Try: "How many capsules are in my Acme workspace?"

Storage: See how much storage you have used and which files are attached to your capsules. Your storage is counted in three parts, each with its own limit on your plan: attachments (the limit that stops new uploads), capsule content, and version history. Files kept for a limited time show the date they will be deleted, written as year, month and day.

Delete an attachment: Remove a file attached to a capsule to free up storage space.

Rotate an image: Turn a stored image attachment a quarter turn left or right until it sits the right way up. The correction is saved to the image itself, so it shows up everywhere the picture appears. The reply includes the image's new width and height, so your AI can confirm the turn without downloading the picture.

Everything you liked: List the capsules you liked on share links, public pages and Discover, newest first. Each item says whether you can still open it right now (public, in Symbol, or through the link you liked it on) and links to the creator's current version. Something since made private or revoked shows as no longer available, with no title.

Account info: Check which Symbol account is connected to your AI, which plan it is on, how many capsules your plan allows and how many you have left, and how long uploaded files are kept. Archived capsules do not count toward the limit. Try: "How many more capsules can I save on my plan?" On a connection that uses an organization API key it names the organization instead of a person: principal is organization_api_key, organization holds the organization's id and name, and email and member_since are empty. A connection made as a person reports principal as user, along with your username, your public profile address, and whether your public profile is on.

Parameter names your AI uses

Tools use one name for each idea: type_name for a type, capsule_refs for a set of capsules, target_type_name for where capsules move to, collection_name for a collection, and cursor to fetch the next page of any list. Assistants that learned the older names (type, capsule_ids, target_type, collection, offset and before) keep working until 31 December 2026, and each reply tells them the name to use instead.

Attaching files from your computer

Installing Symbol Bridge adds three more tools: two that let an assistant read files off your own disk and attach them, and one that reports what the Bridge resolved when something is misconfigured. Turning on end-to-end encryption adds eight more for working with encrypted capsules. All of them are covered on that page.

With the Bridge installed alongside this connector, an assistant reads through the Bridge's local copy on its own while writes keep going here. See When both servers are connected.

On this page