Documentation
Enough to get a node running and a skill talking to it. The full reference — every command, all five contracts, the HTTP and WebSocket API and the security model — lives in the repository, and inside the node itself: the control plane ships the whole reference offline.
Requirements
None to run the release binary. Go 1.25+ only to build from source; Python 3.11+ to run the Python skills; Node 22+ only to rebuild the control-plane UI, which ships prebuilt and embedded. No CGO, no Docker, no database, no account.
Install
- 1 Download a signed binary from the release:
aura-linux-amd64,aura-darwin-arm64,aura-windows-amd64.exeand the other three targets, plusSHA256SUMS, its cosign signature and the SBOM. - 2 Verify:
sha256sum -c SHA256SUMS --ignore-missing, thencosign verify-blobagainst the release workflow's identity — see SECURITY.md. Check the signature before trusting the sums, not after. - 3 Or build:
git clone https://github.com/DeepAxiom/deepaxiom_aura && cd deepaxiom_aura/kernel && go build -o aura ./cmd/aura - 4 Or as a service:
docker compose up— kernel, persistent volume and the strict policy indeploy/aura.policy.yaml.
Quick start
- 1 Start it:
aura up— prints a banner with the control-plane URL and a token - 2 Connect a skill:
cd skills/echo && PYTHONPATH=../../sdk/python/src python main.py - 3 Talk to it:
aura chat --graph echo "hello aura"— needs no model at all - 4 Guard an agent you already run:
aura guard --config claude_desktop_config.json
⚠︎ Flags go before the message: aura chat --graph echo "hello" works, aura chat "hello" --graph echo silently ignores the flag. And aura chat without --graph uses the chat graph, which needs an LLM — echo is the one that works cold.
$ aura chat --graph echo "hello aura"
hello aura
# that round trip went client → kernel → executor → skill → back,
# over a WebSocket, every envelope logged before it was acknowledgedThe public registry and Studio
registry.deepaxiom.com serves the r1 API. aura add verifies and installs; AURA_REGISTRY sets the default. Publishing needs a publisher credential in the URL — AURA_REGISTRY=https://publisher:PASS@registry.deepaxiom.com aura publish my-skill/ — issued on request. Studio — studio.deepaxiom.com — is the web front: browse and download today, publishing from the browser later, reading the registry only through this API.
CLI reference (selected)
Run
| aura up | Start a node: kernel, UI, state store, ledger and witness client |
| aura chat | REPL or one-shot conversation with a graph, streaming replies |
| aura do | Natural-language goal → plan → gated execution |
| aura connect | Project an existing OpenAPI spec as skills |
Prove
| aura guard | Put existing MCP servers behind a policy checkpoint |
| aura verify | Recompute the ledger chain, tree and signatures with no node running |
| aura audit | A period report of what acted, who authorized it and under which policy |
| aura why | Narrate a session's root cause from its causal log |
| aura replay | Deterministically replay a recorded session |
| aura regress | Replay sessions against a new model and diff the effects |
| aura undo | Reverse an effect through its declared compensation port |
| aura backup | Archive the data directory — identity, ledger, secrets — and verify a restore |
Distribute
| aura publish | Sign and upload a skill to a registry |
| aura add | Download, verify hash and signature, review permissions, install |
| aura registry serve | Host a registry of your own, on its own port |
| aura federate | Resolve capabilities that physically live on another node |
The five frozen contracts
C1 manifest, C2 graph IR, C3 channel protocol, C4 effect ledger and policy, C5 attestation. Versioned, additive-only, and shipped with a black-box conformance suite that speaks the raw protocol — any implementation that passes it is conformant.
The full reference
The guide covers every command, all five contracts, the HTTP and WebSocket API, deployment, and the security model in full.