hmanoor/kageops-core: Open-source autonomous AI dev team: Sensei orchestrates specialist agents from idea to shipped product. The engine behind KageOps.

Open-source autonomous AI dev team: Sensei orchestrates specialist agents from idea to shipped product. The engine behind KageOps. - hmanoor/kageops-core

Summary

KageOps-core is the open-source engine behind KageOps: a local-first, multi-agent system where an orchestrator called Sensei routes work to eight specialist agents through a six-phase product lifecycle, then runs real build and acceptance gates before declaring anything done. It targets engineers who want to automate end-to-end product work without sending code to a hosted service. The repository is AGPL-3.0 licensed TypeScript that ships with a Command Center Electron UI and a headless CLI.

What it is useful for

This is for engineers who want to explore autonomous, multi-agent code generation with explicit cost guardrails and real verification, rather than another chat wrapper. Concrete use cases drawn from the README:

• Prototyping full apps from a one-line brief — the README's worked example is a kanban board with drag-and-drop columns and localStorage.
• Previewing a deterministic plan at zero spend with --dry-run, which prints the phase plan, per-phase task count, and a per-preset cost estimate without making any LLM calls.
• Multi-provider experiments through a single adapter: Claude, OpenRouter, OpenAI, Gemini, or Ollama for fully local runs at $0.
• Self-hosted orchestration with no Docker and no external database — PGlite (Postgres compiled to WASM) boots in-process on first run.
• Operating from a GUI (the Command Center) for live supervision, or from CI/scripts via the headless runner.

It is less useful if you only need single-agent code completion, cannot run Node 20+ locally, or want a managed cloud tier — the README notes that KageOps Cloud (managed compute, team collaboration, hosted identity, billing) lives in a separate private repository.

How engineers can use it

The documented setup is:

1. Clone and build:

git clone https://github.com/hmanoor/kageops-core.git && cd kageops-core
npm install && npm run build

2. Preview the plan with no API spend:

npx tsx src/cli/headless-runner.ts --dry-run \
  --name "TaskFlow" \
  --description "A kanban board with drag-and-drop columns and localStorage. Include #board, #add-task, #columns."

3. Run for real under a hard USD cap:

export OPENROUTER_API_KEY=sk-...
KAGEOPS_PRESET=openrouter_budget KAGEOPS_MAX_RUN_USD=0.50 \
  npm run run:headless -- --name "TaskFlow" --description "..."

Every real run requires an explicit KAGEOPS_MAX_RUN_USD; a budget poller checks spend every three seconds and cancels the run if it approaches the cap.

4. Launch the desktop Command Center:

npm run dev

The BuildVerificationGate runs real npm install/build/test, and the AcceptanceGate checks the output against the brief (MUST/SHOULD severity); failures trigger a bounded retry-with-repair loop handled by the Forge agent. The docs/how-to/README.md directory in the repo contains narrated screen-capture walkthroughs if you want to see the GUI in action before installing.