Branche la mémoire à ton agent.

# Memexa — persistent project memory (MCP)

This project uses Memexa: shared memory that persists across AI coding sessions.
Every new session starts amnesic — Memexa restores what previous sessions knew.
Use the tools yourself, proactively; do not wait to be asked.

## Session protocol
1. At the START of every session, before anything else:
   - get_context — load the project's memory (decisions, bugs, conventions, progress).
   - Empty result = new project: start fresh and begin recording.
2. WHILE working, record as you go (never batch to the end):
   - add_decision — on every architectural/technical choice, with reasoning + rejected alternatives.
   - log_bug — on every bug found or fixed, with root cause + fix.
3. Before ENDING a task:
   - update_context — persist this session's progress, decisions and bugs.
   - create_commit — snapshot the session so it shows in the project timeline.

## Reach for these when relevant
- search_memory — before building something, check if it was already decided/tried here.
- search_cross_project — how you solved a similar problem in your other projects.
- get_decision_history — the chronological trail of architectural decisions.
- get_contradictions — surface entries that now conflict; resolve them.
- get_project_status / validate_context / get_insights — health of the memory (coherence, patterns, alerts).
- compare_sessions — diff the current state against N sessions ago.
- get_entry_versions / restore_version — inspect or roll back a specific entry.
- propose_change / list_proposals / merge_proposal / reject_proposal — with multiple agents, suggest edits as reviewable changes instead of overwriting.
- list_commits — list the project's state snapshots.

## Rules
- Load memory FIRST; never re-ask for what is already in memory.
- Record decisions and bugs the moment they happen, concisely, in your own words.
- Update or search existing entries instead of creating duplicates.