Once you're connected

Talk to your agent in plain language — it picks the right brainattic tool for the job. A few things to try:

• Find something: "Search brainattic for our incident post-mortems on auth."
• Get the exact passage: "What does brainattic say our on-call escalation policy is?"
• Draft, with a safety net: "Draft a release note in brainattic for version 2.1 and let me review before saving."
• Check the rules first: "What rules apply before writing in our engineering space?"

Under the hood, those become MCP tool calls. A few examples of what your agent runs:

{ "tool": "kb_documents_search", "arguments": { "query": "incident postmortem auth", "mode": "hybrid", "limit": 10 } }
{ "tool": "kb_chunks_retrieve", "arguments": { "query": "what is our on-call escalation policy", "k": 8 } }
{ "tool": "kb_documents_edit_begin", "arguments": { "document_id": 1234 } }

What your agent can do, in brief:

• Search — hybrid keyword + semantic search, down to the passage.
• Read — open documents, browse projects.
• Write, transactionally — begin an edit, apply it, then commit or discard. Nothing is saved until it's committed, and edits can be rolled back.
• Author — create and update documents.
• Govern — your curated rules are surfaced to the agent before it writes.

brainattic exposes 50 tools in total; your agent discovers the full set automatically once connected.