Repository context for your coding agent

01

Start the session with context

A session brief combines repository identity, index freshness, planning context and available Focused changes. It summarizes repository state, not every conversation you have had about it.

02

Find the plan before proposing the change

Retrieve epics, briefs and architecture documents by name, role or documented lifecycle. Use repository conventions to organize new work, while keeping the plan in Git.

03

Investigate code relationships and evolution

Ask about available callers, implementations and nearby dependencies. Retrieve evolution events and commit references, or use maintenance signals to identify code worth investigating. Results depend on the indexed revision and supported analysis.

04

Search by meaning when the name is unknown

Describe a behavior to retrieve related indexed code and documents. Semantic search requires configured, ready embeddings, with a local ONNX option available. Exact-name and document-metadata queries use structured search instead.

05

Distinguish drafts from committed memory

With Watch Focused enabled, inspect scoped pending changes and temporary Markdown classifications. This overlay is separate from the committed index, not a complete working-tree index.

For explanations worth keeping, have the agent write a Git-backed Commit Note. Saving and committing it are explicit steps.

06

Check the source behind an answer

Retrieve source slices, patches and commit metadata behind a finding, then open the relevant Mnemon view. References make an answer inspectable; the agent's interpretation still needs review.

07

Questions for your next development session

• Which epic is active, and where is its implementation brief?
• Who calls this symbol, and when did its signature change?
• Are there notes explaining this commit?

MCP serves the agent. The desktop app and IDE companion give you direct access to the same repository memory.