An open-source, self-hosted autonomous AI command center.
horus-os lets one person run a personal team of AI agents from a single workstation. Agents take instructions through a CLI or a local web chat, execute tasks against a persistent knowledge base, and surface results in a local dashboard. The whole stack runs on the user's own hardware. AI calls are billed to the user's own API keys.
Alpha, v0.1.0 (2026-05-23). See CHANGELOG.md for what's in this release and ROADMAP.md for what's next.
git clone https://github.com/Ridou/horus-os.git
cd horus-os
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install '.[all]'
# Initialize, then walk through API key onboarding
horus-os init --interactive
# Or initialize silently and set env vars yourself
horus-os init
export ANTHROPIC_API_KEY=sk-ant-... # and/or GEMINI_API_KEY
# Ask the agent something. Default tools include reading and writing
# your notes folder, plus reading any file under it. Tokens stream
# to stdout as they arrive; pass --no-stream for the v0.1 behavior.
horus-os run "Summarize the notes in my notes directory."
# Drive the agent through a named profile (the `default` profile is
# bootstrapped on init; create more with `horus-os agents create`).
horus-os agents list
horus-os run --agent default "Summarize today's notes."
# Browse traces or open the local dashboard
horus-os traces
horus-os serve # http://127.0.0.1:8765
- Two providers, your keys. Anthropic Claude and Google Gemini, called via the official SDKs. No abstraction layer.
- Tool registry. Built-in
read_file, plus six memory tools that read and write a markdown notes folder of your choice. - SQLite persistence. Every agent run produces a trace. Every memory write lands in an audit table.
- Local web dashboard. Single-page chat, traces explorer, writes audit view. No npm, no Node, no build step.
- CLI surface.
init,run,traces,servesubcommands. Tested on macOS, Ubuntu, and Windows. - Setup wizard.
horus-os init --interactivevalidates your API keys against the live providers before saving anything.
The active milestone ships multi-agent orchestration, live streaming
responses, and the third-party adapter plugin contract. Highlights:
named agent profiles in SQLite, a delegate_to_agent tool, a
run_agent_stream async generator, streaming output on the CLI and
dashboard, an Adapter Protocol with entry-point discovery, and a
reference HTTP webhook adapter.
- Migration guide:
docs/MIGRATION-v0.1-to-v0.2.md - Runnable examples:
examples/ - v0.2 milestone detail:
ROADMAP.md
CHANGELOG.md, release notesPROJECT.md, project intent, core values, what is in and out of scopeROADMAP.md, current milestone and what comes nextARCHITECTURE.md, technical shapedocs/MIGRATION-v0.1-to-v0.2.md, upgrade notes for v0.1 usersexamples/, runnable scripts for multi-agent, streaming, and custom adaptersCONTRIBUTING.md, dev setup, workflow, and code styleSECURITY.md, how to report vulnerabilitiesCODE_OF_CONDUCT.md, Contributor Covenant 2.1
Issues and pull requests are welcome. Start with CONTRIBUTING.md for
dev setup and the workflow, and PROJECT.md for scope. New
contributors should look for the good first issue and help wanted
labels on the issues page.
Apache 2.0. See LICENSE.