athena-diary

SanctumOS · Athena

Off-context diary for Letta agents: SQLite source of truth + MCP tools so journaling does not stuff core memory. Pattern matches origin_conversation (DB + MCP wrapper).

Status: v0.1.0 on moya — Build slices shipped; diary MCP attached to Athena + sleeptime. Product requirements: DSC Tasks Doc #1039 (board Athena Diary).

What this is (v1 direction)

• Diary outside the context window — write freely; retrieve via MCP.

• Embed summaries (not full bodies) with sqlite-vec in the same SQLite file; FTS fallback.

• Default embed path: Sanctum / Letta vectorization lanes; optional external provider for standalone.

• Sleeptime clerk tags / lesson-family / re-embeds (turn-based).

• Sensitive material stays in Broca human blocks (instructional), not the diary.

Sequencing: ship diary first; core compaction later so brain surgery has somewhere to put what matters.

Related MCP server: Agent Progress Tracker MCP Server

Quick start (scaffold)

python3 -m venv .venv && source .venv/bin/activate
pip install -e '.[dev]'
# Later slices: pip install -e '.[dev,mcp,vec]'

athena-diary-mcp --version
athena-diary-mcp --describe
athena-diary-mcp health

# Coverage gate (Build slices require ≥90%)
pytest --cov=athena_diary_mcp --cov-report=term-missing --cov-fail-under=90

Copy .env.example → .env when configuring embed/DB paths (later slices).

Docs

License

• Source: AGPL-3.0

• Docs / non-code: CC-BY-SA 4.0

See NOTICE.