Set one API key. Commander classifies the task, picks the right topology, and streams every agent decision to your terminal. No black boxes. 6700+ tests, no graph builders, no YAML.
Deliberation, topology, tools, and quality gates stream live — not after the fact. Copy, set a key, and watch agents work.
git clone https://github.com/PStarH/Commander.git
cd Commander && pnpm install
export OPENAI_API_KEY=sk-...
npx tsx packages/core/src/cliEntry.ts run "audit this repo" --streamSix core capabilities, each backed by production-grade infrastructure and audited in CI.
Every agent thought, tool call, and decision streams in real time via SSE.
Read more →Classifies CODING / RESEARCH / ANALYSIS / FACTUAL and picks SINGLE · CHAIN · DISPATCH · ORCHESTRATOR · REVIEW.
Read more →One API key. Auto-detect + failover across cloud and local models.
Read more →Five-layer verification: hallucination, consistency, completeness, accuracy, safety.
Read more →Thompson Sampling + Reflexion meta-learner. Activates after 5+ experiences.
Read more →Circuit breakers, DLQ, WAL checkpoints, semantic cache, saga, supervision tree.
Read more →Built for engineers who want to see what their AI is actually doing.
Other frameworks hide agent graphs behind YAML and hope. Commander streams every thought, tool call, and quality gate so you can debug and trust the run.
Describe the task in plain language. Deliberation classifies it, picks a topology, and scales agents 1–20. Simple tasks stay simple.
Circuit breakers, DLQ, saga compensation, SQLite WAL checkpoints, semantic cache, multi-tenancy — the same discipline as a distributed system.
Set any API key. Auto-detect + failover across cloud and local models (Ollama, vLLM). Swap providers without rewriting your workflow.
Three paths into the same engine.
One command, 25 providers, every decision visible. Set an API key and go.