01
A runbook changes in the wiki while the agent still cites an older ingestion snapshot.
Doco use case
A team wiki humans and AI agents maintain together
Use Doco as an AI team wiki when people need a polished collaborative editor and agents need scoped search, citations, and protected maintenance operations against the same documents. The design goal is one source of truth, not a wiki plus a separate stale RAG copy.
The real failure mode
Two knowledge systems drift apart
Teams write in one application while agents answer from another index or copied file set. The first missed update creates two conflicting truths.
02
Users can see a page, but an integration token may have broader or different access than intended.
03
An automated maintenance job reports success after replacing a teammate's concurrent edit.
Operating loop
Operate one shared wiki
- 01
Organize knowledge bases, nested folders, documents, members, and permissions around real team ownership.
- 02
Let agents browse structure, search with evidence, and read only the relevant blocks from the current document.
- 03
Require version-protected block or batch writes; broadcast accepted changes through the same Yjs document humans have open.
- 04
Review comments, changes, versions, relationships, summaries, and concepts without confusing derived material for authority.
Optimistic concurrency prevents silent stale overwrites; it does not automatically resolve every semantic disagreement. Humans still review high-impact policy changes.
Capability map
What the system must provide
| Dimension | The need | How Doco handles it |
|---|---|---|
| Source of truth | A shared source visible to the whole team | Knowledge bases, folders, documents, memberships and document sharing |
| Retrieval | Fast discovery across folders and long documents | Tree/outline traversal and structured search with context |
| Evidence | Answers that point to exact current text | Stable block citations, versions, completeness and freshness |
| Writes | Controlled automation for routine maintenance | Scoped tokens, read/write tools, atomic batches and idempotency |
| Concurrency | Live co-editing without hidden last-write-wins | Yjs for live text plus If-Match for snapshot API writes |
| Portability | A credible exit and self-hosting path | Markdown/ZIP/native exports and MIT deployment stack |
Good fit
Engineering runbooks, product specifications, support knowledge, onboarding, policies, architecture decisions, release notes, and maintained internal documentation.
Not the right fit
A full project-management suite, CRM, data warehouse, ticketing system, or enterprise governance platform requiring mature SSO/audit features that Doco has not implemented.
Bottom line
Use the right memory
An AI team wiki works when people and agents look at the same text, cite the same block, and encounter the same visible conflict instead of silently maintaining parallel copies.
01Does Doco replace Confluence or Notion for every team?
No. Mature suites have broader databases, administration, integrations, and enterprise controls. Doco is narrower and focused on human-agent documents.
02Can an agent edit while a person is typing?
Yes. Agent writes enter the same Yjs document, while the API version gate rejects a stale snapshot instead of silently overwriting newer content.
03Does search use RAG embeddings?
Doco currently uses structured full-text search with block context, versions, and completeness. It does not claim semantic retrieval it has not implemented.