Context
Context stores revisioned Markdown documents attributed to the member and client that wrote them.
Context is Dreamstate's workspace knowledge store. It keeps ordinary Markdown documents for facts about the product, buyers, competitors, and voice, so people and agents work from visible shared material rather than an invisible private memory.
Open Context at /dashboard/context when a decision or source should guide later work. Use a workbook for working rows and use tags, comments, and saved views for workspace organization, not as a replacement for a documented business fact.
What Context holds
Each Context item is a document in a workspace wiki tree. A document has a stable identity, Markdown content, revisions, and a logical path. The application exposes Files for creating, organizing, reading, and editing visible documents, and Graph for traversing the same accessible documents and their authored Markdown links.
Write a document when the statement is meant to survive a single task: a current positioning decision, an ideal customer profile, a competitor observation, a research-backed claim, or a voice rule. Do not treat a run receipt, provider response, schedule, or analytics result as Context by itself. Those are operational evidence; Context is where the team records the conclusion it wants future work to use.
Architect uses the knowledge skill to find and use this material. The skill teaches the practice, but it does not grant authority or bind a tool. The actual work is performed through the knowledge capabilities, and every listed Context capability is free: its credit class is free and its cost policy has zero credits.
| Surface | Use it for | What you can inspect |
|---|---|---|
Files | Writing and organizing ordinary Markdown documents. | Documents and folders that are visible to the workspace. |
Graph | Following authored document links. | Bounded nodes and edges that use the same references as Files. |
Search | Finding a business fact before creating another page or starting web research. | Context documents you are permitted to read, by title and logical path. |
Find the current document before you write
Search Context before creating a page about the business. The Context search capability exists specifically for titles and logical paths, and its catalog description directs agents to search before duplicating a document or researching product, buyers, competitors, or voice. In the app, first check the document title and the revision shown on the page. That check keeps a new claim connected to the team's existing language and evidence.
Ask Architect to locate the relevant page, read its current revision, and identify what is absent. Architect uses the knowledge skill and read-only knowledge controls for this step. A read is free, has no business side effect, and is successful when it returns the permitted canonical revision with typed claims and exact source spans.
The CLI commands above intentionally begin with tools get. The canonical surface requires discovery before execution, so the returned schema, not a guessed payload, defines the parameters for the search or read. A successful search does not mean that a document has been edited; it only confirms that the current material was found.
Create a document and its first revision
knowledge.brain.knowledge.create_document creates a Context document and its first revision for the signed-in editor. Pass null for parent_folder_id to create the document at the workspace root. If you provide a folder id instead, it must identify an existing folder. Context documents belong in this wiki tree, not in a growth-profile Brain section.
The create call is an internal, reversible draft write. It has no provider call, no spend, and no approval system step. Use mode: "live" because a mutating call must name a mode and the server never defaults a mutation to live. The idempotency key makes a retry identifiable as the same requested creation, rather than a reason to manufacture another Positioning page.
Check the Files view after the call. You should see the document at the intended root or folder path, then use its stable document id rather than its title for later reads and revisions. The operation is free, and the useful completion signal is the saved document and revision receipt, not merely a generated paragraph.
Save against the revision you read
knowledge.brain.knowledge.save_revision saves a new revision only when its expected_document_revision matches the current document. This is how you save attributed revisions safely: read the exact revision, make a focused update, then save against that revision instead of silently replacing an intervening editor's work.
The history capability reads bounded revision metadata for one document. The revision comparison capability compares two revision coordinates, including revision number, content object, and content digest, but it does not compute a text diff. Use that distinction when investigating a change: history tells you which revisions exist, while a comparison checks their coordinates.
Architect should use the knowledge skill for the writing practice and the context read and save capabilities for the change. In the app, verify the revised Markdown and revision number before relying on it in a workbook, sequence, or visibility analysis. Saving a revision is free and internally reversible, but it changes the shared document, so accuracy matters more than speed.
Organize documents and recover a removed page
Folders are editable workspace objects. You can create a folder at the root by omitting its parent, passing null, or passing an empty string. Move operations use exact folders and current revisions. Renaming a document preserves its stable identity and revision bindings; renaming a folder rewrites its logical path and every descendant path.
Removing a document is recoverable. knowledge.brain.context.remove_document moves a document into its removed lifecycle state, and list_removed lists documents still inside retention for restoration. restore_document restores an archived document against its current revision. This gives a team a recovery path when a page was moved aside by mistake.
These operations remain free, but the permanent-delete effect is not reversible. When a document should continue to guide work, prefer a new revision or a recoverable removal. When it should move, use a move operation with the document's current folder and revision rather than recreating it elsewhere.
Read links, revisions, and permitted exports
Context has separate read operations because each answers a different question. get reads one permitted canonical revision, including typed claims and exact source spans. history reads bounded revision metadata for one document. diff_revisions compares revision coordinates, while backlinks reads committed incoming wiki links to one destination.
Use graph when the question is about a bounded relationship between Context nodes and edges. It uses the same node references as Files. Use browse when the question is what persisted nodes sit under one exact parent. Neither read operation fabricates children or links that are absent from the workspace.
The export operation produces the Context documents you are permitted to read as canonical Markdown with source attribution. It is a read-only, side-effect-free operation. An export never grants a caller access to a document they are not permitted to read, and it should not be treated as a substitute for reading the current revision before editing.
| Question | Context operation | Result |
|---|---|---|
| What does this document currently say? | knowledge.brain.context.get | One permitted canonical revision with typed claims and source spans. |
| Which revisions exist? | knowledge.brain.context.history | Bounded metadata for one document node. |
| Which pages link here? | knowledge.brain.context.backlinks | Permission-filtered incoming links with stable keyset pagination. |
| What can leave as Markdown? | knowledge.brain.context.export | Permitted documents with source attribution. |
All four reads are free. In the app, start with Files when you know the page and use Graph when the relationship matters. In Architect, ask for the question you need answered and name the document or folder; it should choose the read capability whose result matches that question, then report the returned document identity or revision.
When it goes wrong
If the needed page appears to be missing, search by title and logical path before creating a replacement. If it exists in a different folder, move it against its current revision. If another editor saved a newer revision, read the current document again, preserve the newer content, and save a merged revision against the revision you just read. Do not use a stale expected revision as if it were current.
If a linked page cannot be understood, inspect the graph or read the document directly. Backlinks answer which committed incoming wiki links point to one Context document, while knowledge.vault.link_resolve is the distinct operation for changing one unresolved wiki link to a chosen record. If an export omits a page, it exports only the Context documents the caller is permitted to read with source attribution, so check the document's access and publication state rather than assuming the Markdown was lost.
If removal was accidental, list removed documents and restore the correct archived document against its current revision. If someone needs permanent deletion, confirm that the document is already archived and that the selected revision is the intended one. Context has no credit charge for these actions, but the irreversible delete behavior makes verification essential.
A document that was renamed retains its stable identity and revision bindings, so use its document id when troubleshooting a link or a stale title. A moved folder can change logical paths for every descendant. Check the Files location after either change, then read the document again before using it as evidence in another workflow.
Do not repair a broken reference by inventing a replacement page. The documented recovery path is to search, browse, read the current revision, and restore an archived document when appropriate. These reads are free and side-effect-free, so they are the safe first response to uncertainty.
If Context is empty, report the fact that is missing and its needed source. Do not substitute a generated statement for verified workspace knowledge. A later editor can create the document once there is a source worth recording.