Open source · self-hosted AI agents · written in Zig
Ten small open-source tools for running AI agents on your own hardware: an agent you can message from Telegram or Slack, plus the orchestration, task queue, shared memory and tracing to grow it intoa fleet. Each one is a single Zig binary — install NullHub to manage them all, or take any piece alone.
a fleet. Each one is a single Zig binary — install NullHub to manage them all, or take any piece alone.
counted in the source at v2026.5.29 · everything pre-1.0
01 · Get started
NullHub is one binary with an embedded dashboard. Run it with Docker, or grab a release binary for your platform.
docker run --rm -p 19800:19800 \
-v nullhub-data:/nullhub-data \
ghcr.io/nullclaw/nullhub:latest
# dashboard → http://nullhub.localhost:19800The install wizard fetches, configures and links each component. Skip what you don't need — every component runs independently.
nullhub install nullclaw # the agent
nullhub install nullboiler # orchestration
nullhub install nulltickets # durable tasks
nullhub install nullwatch # observabilityOne command brings the stack up; the dashboard supervises processes, health, logs and updates from ~/.nullhub.
nullhub start-all && nullhub status02 · The stack
03 · Core tools
also in the ecosystem nullwatchnullpantrynllclwnulldesknullcapnullbuilder
04 · Design principles
State lives in ~/.null* on your disk. Nothing phones home; there is no hosted service to depend on.
Every component is one static Zig binary. No Node, no Python, no containers required — kilobytes, not gigabytes.
Components compose over plain JSON and HTTP, and every one works alone. Use one tool or combine the full stack.
Pre-1.0, CalVer-versioned; the numbers on these pages come from the source tree and the release assets.
05 · Documentation
Task-oriented guides for real setups — a personal assistant, a durable backlog, an observable multi-agent stack. The site also ships an llms.txt index that points agents to the canonical documentation.