One wiki. Two readers.
You browse pages in the dashboard. Your agent reads and writes them through the API. Same data, same links, same graph.
For the human reader
A clean, readable wiki dashboard. Browse pages, follow links, check history — all the things you'd expect from a good knowledge base.
Readable pages
Rich text, code blocks, tables, links. Clean and readable — no markup soup. Revision history on every page.
Full-text search
Cmd+K to search every page. Find what your agent wrote, what you wrote last week, what someone linked to.
Visual graph explorer
See how pages connect. Spot orphaned pages, find gaps, discover what your agent has been writing about.
Revision history
Every edit tracked. See what your agent changed, when, and why. Roll back if it wrote something wrong.
Auth uses Auth0 OIDC. Resolution path: JWT → User → Workspace
Three surfaces share the same models: HTTP API, MCP server, and this dashboard.
→ Linked from: docs:architecture, journal:2026-08-12
For the agent reader
A REST API and MCP server your agents use to read and write knowledge. No vector DBs to tune — just pages, links, and search.
REST API
CRUD pages via bearer JWT. Optimistic concurrency with ETags. It's just HTTP — no SDK required.
MCP server, built in
Six tools out of the box: getPage, savePage, appendPage, listPages, searchPages. Point any MCP client at your workspace.
Semantic search
pgvector embeddings on every page. Ask for "how does auth work" and get the right page back — even if those exact words aren't in it.
Graph traversal
Your agent doesn't just read pages — it navigates. Wanted pages, orphans, backlinks. The same navigational intelligence a human reader uses.
Same data. Same API.
Whether you're in the dashboard or your agent is hitting the API, it's the same pages, the same links, the same search index.
One backend. Three surfaces. No sync.