Spacebot

Spacebot is an opinionated AI agent harness written in Rust, designed from the ground up for teams, communities, and multi-user environments. Its core architectural bet is that agent state belongs in structured storage, not markdown files the LLM manages: memory lives in a typed graph in SQLite, conversation history compacts into durable navigable 'chronicles,' and autonomy runs on a task state machine linked to goals rather than a heartbeat file. The system splits the monolithic agent loop into specialized processes — conversation, compaction, memory retrieval, and execution each run independently, so the agent never goes dark while doing background work.

A standout feature is continuous self-improvement: after complex tasks, the agent captures reusable skills, and a background process silently persists worthwhile memories when conversations idle. Recent development shows serious engineering rigor — a prompt inspector that records and decomposes every LLM request into block maps, per-model context ceiling enforcement with learned refusals, and careful handling of fallback model routing. It ships with one-click hosted deployment via spacebot.sh connecting Discord, Slack, Telegram, and Twitch, while remaining self-hostable.

Compared to OpenClaw, Spacebot targets a different niche: true concurrent multi-user operation with shared memory across channels, versus OpenClaw's personal-assistant focus. It is licensed under FSL-1.1-ALv2 (a delayed open-source license), which may matter for some adopters. The project is young but backed by the Spacedrive team with very active, disciplined development.