Familiar-Connect

An AI "familiar" that joins Discord voice channels, listens, understands speech, and talks back with real AI voices. Written in Rust.

Full documentation lives at docs/. The architecture pages describe the system as designed; the implementation is the Rust workspace in this repo (ported from the original Python prototype in July 2026 — see docs/rust-port/ for the port design, per-subsystem specs, and the review log).

Quickstart

cp .env.example .env                    # fill in your tokens
cargo run --release --features discord -- run --familiar <id> -v

Feature flags select the integration surface (defaults cover storage, HTTP, and images):

# Text-only Discord bot
cargo build --release --features discord

# Voice (DAVE E2EE via songbird) + Deepgram streaming STT
cargo build --release --features discord,discord-voice,stt-deepgram

# Local ML extras (ONNX turn detection, local embeddings)
cargo build --release --features local-turn,local-embed

See docs/rust-port/DAVE-RUNBOOK.md for prerequisites (Discord bot token + intents, OpenRouter key, optional TTS/STT keys), the on-disk <familiars-root>/<id>/ shape, platform notes (Windows voice builds need CMake), and a staged smoke ladder for voice.

Where things are

• Architecture — the bot shell, event bus, memory strategies, voice pipeline.
• Port history — module map, decision log, conventions; specs/ holds the per-subsystem behavioral contracts and the review log.
• Contributing — dev workflow and conventions.

Development commands

cargo build                                    # compile (default features)
cargo test                                     # run the test suite (~2,000 tests)
cargo clippy --all-targets -- -D warnings      # lint (pedantic; warnings are errors)
cargo fmt                                      # format
cargo run -- diagnose <logfile>                # span/latency report from a captured log

The pre-commit hook at .githooks/pre-commit runs the same gates (git config core.hooksPath .githooks to enable).