Aichronicles

aichronicles

**Capture your AI coding sessions, locally, with secrets scrubbed before they hit disk.**

**Status: work in progress.**

aichronicles runs as a tiny user-mode daemon that ingests hook events from Claude Code, Gemini CLI and Codex CLI into a SQLite store, exposes the corpus as an [MCP](https://modelcontextprotocol.io) server your agent can query for context from past work, and generates structured summaries on demand.

flowchart LR
    cc[Claude Code]
    llm[(Anthropic / OpenAI
HTTPS, on-demand)]

    subgraph local["Your machine — single user, local disk"]
        direction LR
        hook["aichronicles hook
hook subprocess"]
        api["aichronicles-api
systemd --user · UDS 0600"]
        db[("SQLite store
raw_envelopes ▸ events ▸
extractions ▸ llm_outputs")]
        mcp["aichronicles mcp-serve
stdio"]
        cli["aichronicles summarize
/ reflect / propose"]
    end

    cc -- "hook fires" --> hook
    hook -- "POST /v1/ingest" --> api
    api -- "redact + verify + persist" --> db
    cc <-- "tools/call (stdio)" --> mcp
    mcp -- "GET /v1/* (UDS)" --> api
    cli -- "GET /v1/* (UDS)" --> api
    cli -.->|"egress redact ▸ tool use"| llm
    llm -.-> cli
    cli -- "POST /v1/llm-outputs" --> api

**Legend.** Solid arrow: data flows in this direction. Dashed arrow: optional / on-demand (only fires for summarize/reflect/propose). The `local` box is the trust boundary — everything inside runs as your user, on your local disk. Cylinder is persistent state.

The diagram above is the static view (where things run). For the *dynamic* view — what fires automatically vs. what you trigger, and what happens step-by-step inside each flow — see [`docs/explanation/data-flow.md`](docs/explanation/data-flow.md).

What you get

• Searchable corpus. Every prompt, tool call, and response from your Claude Code sessions, indexed by SQLite FTS5. aichronicles search 'mongodb shutdown' return