architecture overview

design

every subsystem is a trait. swap implementations with a config change, zero code changes.

subsystem	trait	ships with	extend
AI models	Provider	22+ providers	custom:https://your-api.com
channels	Channel	CLI, telegram, discord, slack, iMessage, matrix, whatsapp, webhook	any messaging API
memory	Memory	SQLite hybrid search + LZ4	any persistence backend
tools	Tool	shell, file_read, file_write, memory_store, memory_recall, browser, composio	any capability
observability	Observer	noop, log, multi	prometheus, OTEL
security	SecurityPolicy	pairing, sandbox, allowlists, SSRF, encrypted secrets	-
tunnel	Tunnel	cloudflare, tailscale, ngrok, custom	any tunnel binary

modules

src/
├── agent/        # conversation loop, tool dispatch
├── channels/     # channel trait + 8 implementations
├── config/       # toml config, env var overrides
├── cron/         # scheduled tasks
├── daemon/       # full runtime with JoinSet
├── doctor/       # system diagnostics
├── gateway/      # HTTP API server
├── health/       # health check endpoints
├── heartbeat/    # periodic task engine
├── integrations/ # 74 integration definitions
├── memory/       # SQLite hybrid search
├── migration/    # legacy format migration
├── observability/# logging, metrics
├── onboard/      # setup wizard
├── providers/    # AI provider trait + implementations
├── runtime/      # native runtime adapter
├── security/     # pairing, secrets, sandbox, policy
├── service/      # OS service management
├── skills/       # skill loader
├── tools/        # tool trait + implementations
└── tunnel/       # tunnel trait + implementations

tech stack

layer	tech
language	rust, 100%
allocator	mimalloc
mutex	parking_lot (1 byte vs 40)
concurrency	tokio JoinSet
memory	SQLite + FTS5 + vector cosine + LZ4
encryption	ChaCha20-Poly1305 AEAD + DPAPI
HTTP	axum + tower
caching	DashMap concurrent hashmap
build	opt-level=z, LTO, panic=abort