OmniNode

OmniNode is a provider-agnostic multi-AI orchestration platform. It coordinates CLI-based AI agents, AI providers, persistent memory, roles, reports and planning into a unified execution system — so multiple AI systems can collaborate on complex software engineering and knowledge-work tasks.

OmniNode is not an AI provider and is not dependent on OmniHilbras (or on OpenAI, Gemini, DeepSeek, Qwen, or any other specific vendor). It orchestrates existing intelligence; it does not try to become another model.

v2.0.3 is the current stable release; v2.0.4-alpha.1 is the prerelease with the v3 audit remediation. The v2.0.0 reliability and interoperability release shipped all 24 roadmap phases; v2.0.1 – v2.0.3 were hardening and consistency releases on top of it, and v2.0.4 is the v3 audit remediation release — it fixes every confirmed critical and high finding of the full-system audit (docs/AUDIT_V3.md): the published package no longer advertises exports that do not exist, pipeline and task state is no longer lost, a store file from a newer OmniNode is never destroyed, and a misbehaving agent can no longer crash the orchestrator.

The roadmap is docs/ROADMAP_V2.md; the completed v1 plan is docs/DEVELOPMENT_PLAN.md.

The v1 plan (phases 0–11) is complete and shipped as v1.0.0 — see docs/DEVELOPMENT_PLAN.md (historical). The v2 roadmap (24 phases) is complete and shipped as v2.0.0 — see docs/ROADMAP_V2.md. Subsequent v2.0.x releases are hardening, consistency and release-quality passes over that completed architecture (the current stable release is v2.0.3; the v3 audit remediation ships as v2.0.4-alpha.1). The forward-looking v3 direction — advanced orchestration capabilities that build on the v2 foundation — is in docs/ROADMAP_V3.md.

Install the stable release with:

npm install -g @hilbras/omninode

Core architectural rule: OmniNode must run successfully with zero Hilbras dependencies. OmniHilbras and Remembera make it more powerful, but neither is required for the core engine.

npm install -g @hilbras/omninode          # stable (v2.0.3)
npm install -g @hilbras/omninode@next    # prereleases (v2.0.4-alpha.1)

Or as a library: npm install @hilbras/omninode. Node.js 20+.

Prefer a container? docker build -t omninode . then docker run --rm -v "$PWD:/work" -w /work omninode run my-pipeline "audit the repo".

omninode init                                   # writes omninode.yaml
omninode provider add --name my-gateway \
  --base-url https://api.example.com/v1 --api-key-env-var MY_API_KEY
omninode agent add --name researcher --command some-agent --input-mode stdin

Describe the work as a pipeline — in omninode.yaml:

project:
  memory:
    provider: local            # or: remembera
  planner:
    kind: model
    model: my-gateway:chatgpt
  pipelines:
    - id: repo-audit
      steps:
        - id: research         # independent analyses, run in parallel
          kind: research
          agents: [researcher, second-opinion]
        - id: plan             # findings + memory → typed implementation plan
          kind: plan
        - id: execute          # hand the plan to an execution agent
          kind: execute
          agent: researcher

…then run the whole loop in one command:

omninode run repo-audit "Audit the authentication module"

  [research] completed
  [plan] completed
  [execute] completed
  reports: 2 structured report(s)
  combined report: combined-m4f2
  plan: plan-a91
  result: Fixed JWT verification and added regression tests.
Pipeline completed. Run id: run-a91

Inspect what happened, in text or JSON:

omninode pipeline inspect run-a91
omninode report combined
omninode plan show plan-a91
omninode task inspect task-… --json | jq .
omninode audit -n 50

Prefer to start from something runnable? examples/demo/ contains a complete project and ./demo.sh that exercises the full loop with agents that need no external tools.

OmniHilbras is one provider source among many — always optional. The core architectural rule stands: OmniNode runs with zero Hilbras dependencies.

The dedicated adapter implements the plan's connect flow on top of the gateway's OpenAI-compatible API surface and tags discovered models with gateway: omnihilbras metadata:

omninode provider add --name omnihilbras --type omnihilbras \
  --base-url https://your-omnihbras-host/v1 \
  --api-key-env-var OMNIHILBRAS_API_KEY

omninode provider test omnihilbras --connect   # full connect flow + registration
omninode models omnihilbras                    # list the gateway's models

MIT