DreamLake

Related resources and activity

Session and resource activity connects work across Memos, artifacts, annotations, infrastructure stacks, repositories and agent conversations. A memo can be linked to several sessions, and each session can work on several resources. Explicit relationships are shown separately from observed access. Opening a memo does not automatically declare it the session's documentation.

Related sidebar

The Memo sidebar orders Related → Table of contents → Comments → …. Attachments, variables, version history and extension views remain in …. First is a position, not a forced default: each Memo restores its explicit sidebar selection. Without a saved selection, the existing Contents/Comments automatic choice remains. On a narrow or collapsed pane, Related opens the same information in a dialog; closing it returns focus without replacing the editor.

Related groups context into one surface:

  1. Links to this memo: incoming references from persisted Memo content and directed explicit resource relationships, deduplicated by source identity.
  2. Linked resources / outgoing mentions: explicit affiliations and references from this Memo. Direction and reason remain visible.
  3. Contributing sessions: sessions explicitly linked to, or recorded as creating or editing, this Memo. Contribution is distinct from a content backlink.
  4. Recent activity: recorded reads and edits, loaded when expanded. Access alone does not establish a link; this is not a complete edit history.

Incoming content references support :memo[ID], :memo{id="ID"}, leaf ::memo[ID], their legacy :note equivalents, #note:ID, and Markdown links to DreamLake Memo routes (/namespace/memos/ID, legacy /namespace/notes/ID, and project ?note=ID navigation). Fragments do not duplicate a source. Code, escaped references, reference text inside unrelated link labels, invalid directives, external lookalike URLs and bare IDs are excluded. ID links survive title and filing changes. Public article links (/namespace/public/slug) follow the current reserved permalink and its aliases. Editor routes also support public permalink aliases and current catalog slugs, following the same permalink-first resolution and viewer access as navigation. A catalog slug is an existing resource identifier, not an arbitrary title: changing a Memo title does not change that slug. ID links remain the stable choice across namespace moves. Conversation transcripts, arbitrary artifact code/HTML, and program or tool output are not automatically scanned.

Content backlinks follow the persisted Memo projection. Active RTC edits can arrive after the paced archive sync (normally up to about 30 seconds, longer under backlog or transport failure). A removed reference disappears after that projection updates. Existing Memos are indexed in bounded background batches. During rolling deployments, references known to belong to an older saved body are omitted until a bounded background audit repairs them; repair latency depends on the indexed catalog size and archive-loop backlog. The index stores at most 4,096 distinct reference targets per Memo; exceptionally large reference catalogs beyond that bound are not exhaustively represented. Explicit links persist until unlinked; unlinking does not edit source content.

Current permissions are checked on every backlink page. Deleted, archived and inaccessible source Memos are omitted, without exposing their titles or totals. Source rows navigate using current names and stable IDs. Browser Back restores the Memo and selected sidebar view. Session drill-in has a return action.

Each section has independent loading, empty and retry states. Refresh related resources refreshes authorized metadata; returning to the browser also refreshes it. Expensive reads run on demand, with bounded pagination and cancellation of obsolete requests. Metadata failures do not disable Memo editing.

GET /resources/note/:id/backlinks?cursor=... returns {resources: Resource[], nextCursor: string | null}. Each resource has kind, id, name, url, and writable, matching the relationship API's resource descriptor. This endpoint returns incoming Memo content references; explicit typed links remain in the relationships endpoint below.

Pages inspect at most 50 indexed source candidates. Permission filtering can produce an empty page with a continuation; continue until nextCursor is null. The opaque cursor expires after ten minutes and is bound to the viewer and target. It grants no access. Invalid, expired or mismatched cursors return 400 invalid_backlink_cursor; restart from the first page. No workspace scan or source-body download occurs during this lookup. This is a REST endpoint; no new CLI, SDK or agent-tool command is implied.

Artifact and annotation context

Artifact and annotation viewers have a collapsed side panel with Related and History tabs. Open it to inspect linked resources, accessible saved sessions, and recent recorded activity without replacing the content. The Memo editor can also record explicit links to artifacts and annotations. Links do not edit content or change access; unavailable context leaves the editor or viewer usable.

Optional context

Memos continue to read and save when activity is unavailable or session metadata is missing or invalid. Activity never grants access, changes content, or retries a saved edit. An unavailable activity panel does not disable the editor.

Authorized API requests can attach X-DreamLake-Session-Provider (codex, claude, browser, or agent), X-DreamLake-Session-Id (the external session ID), and optionally X-DreamLake-Session-Name. IDs are scoped to the authenticated owner. Existing X-DreamLake-Agent-Id and X-DreamLake-Agent-Name are accepted as legacy agent context. This attribution is self-reported, not execution proof.

Relationships

Register a conversation with POST /resource-sessions:

json
{"provider":"codex","externalId":"your-conversation-id","name":"Launch API work"}

Use its returned session.id as a resource reference. Supported kinds are note, artifact, annotation, session, repo, and infra_stack. Create an idempotent relationship with POST /resource-relationships:

json
{
  "source": {"kind": "session", "id": "RETURNED_SESSION_ID"},
  "relation": "documents_in",
  "target": {"kind": "note", "id": "NOTE_ID"}
}
RelationSourceTargetOptional properties
defined_byInfraStackRepodirectory, ref, description
works_onSessionInfraStack or Repodescription
documents_inSessionMemo, Artifact or Annotationdescription
documented_byInfraStackMemodescription
related_toAny supported resourceAnother supported resourcedescription

GET /resources/:kind/:id/relationships returns authorized links in both directions and a nextCursor. Pass it as cursor to continue. Each page examines at most 100 candidate links; hidden resources can yield an empty page with a continuation. Other owners' private session links are excluded before paging. DELETE /resource-relationships/:id removes the association only. Creation requires write access to the source and read access to the target; removal requires write access to the source. A link does not share either resource. When an artifact or annotation is the source, relationship management requires namespace membership or inherited project write access. Read access suffices when it is the target of a relationship from a source the caller can manage. A shared annotation's content edit permission alone does not grant link management. Artifact and annotation detail responses expose this separate capability as canManageLinks. An incoming Memo link remains removable only by a writer of that source Memo.

Links use stable IDs and survive namespace moves. Inaccessible or deleted targets are omitted. Session details are private to their authenticated owner. Restoring a soft-deleted annotation revives its existing catalog identity and therefore restores still-authorized relationships and retained activity; it is not a new resource with the same name.

Catalog identities

Memo, artifact and annotation references use stable 24-character catalog IDs, not titles, slugs or storage names. Get an artifact's resourceId from GET /namespaces/:slug/artifacts/:artifactId (also present in catalog list entries), or an annotation's id from GET /namespaces/:slug/annotations/:name. Both endpoints apply current read permissions. For example, link a Memo to an annotation:

json
{
  "source": {"kind": "note", "id": "MEMO_CATALOG_ID"},
  "relation": "related_to",
  "target": {"kind": "annotation", "id": "ANNOTATION_CATALOG_ID"}
}

Saved terminal sessions

Saved terminal sessions are separate from the owner-private conversation context registered by /resource-sessions. Their saved IDs are 24-character IDs; conversation context IDs are 64-character hashes. Neither can substitute for the other, and neither is a credential.

GET /namespaces/:slug/resources/:resourceType/:resourceId/sessions lists currently readable saved sessions and returns canManageLinks, hasMore, and nextCursor. Types are note, artifact, and annotation; use the Memo catalog ID, artifact logical ID/current slug, or annotation immutable name/current slug, respectively. Optional pageSize is 1–50 (default 25); send cursor to continue. Permission filtering can produce an empty page with a continuation.

PUT or DELETE at the same path plus /:sessionId links or unlinks a saved session. Both require a JSON body with sessionNamespaceSlug. The caller must be able to manage resource links and read the saved session. Repeating the same operation is idempotent. GET /namespaces/:slug/terminal-sessions/:id/resources returns authorized links in the reverse direction. These affiliations grant no new resource permissions or artifact/annotation execution authority.

Activity

GET /resources/:kind/:id/activity returns up to 50 events and an opaque nextCursor; pass it as before for the next page. A session feed contains activity on that session's resources, filtered by the viewer's current access. GET /resource-sessions lists up to 100 of the caller's recent sessions.

Successful authenticated Memo metadata/body, InfraStack metadata/state and Repo ID reads/edits generate metadata after the response. Artifact catalog detail, preview metadata reads, catalog registration/apply, slug and preview metadata writes, plus annotation catalog creation/detail/title/visibility/slug changes, also record activity against their stable catalog IDs. Upload credentials and read-token grants do not prove content was read or changed and generate no such event. Opening an existing Memo editor records a view, not an edit. Repeated reads and views are coalesced per minute. Failed or anonymous requests and presence/feed polling do not appear. Direct RTC edits are not yet observed by this API mechanism; this feed is not a complete edit audit. The standalone GET /notes/:noteId/content endpoint is not currently recorded. Relationship changes are represented by the current explicit links, not additional activity events. Edits retain request identity, so a repeated HTTP request can produce another event even when its relationship write is idempotent. Activity is retained for 90 days. Explicit relationships and session identities persist independently of activity retention.

A repository relationship identifies relevant source; it does not prove which commit was applied. Use the existing operation and infrastructure state receipts for deployment provenance.