Interactive demo — ChorusGraph in 90 seconds

Click through the launch story: native graph engine, Route Ledger audit, cache savings, and verified benchmarks. Steps 1–3 run locally with no API key.

LLM-free demo CLI Apache-2.0 pip install chorusgraph

    Step 1 of 6

    The problem

    Production agents usually glue six systems together — orchestration, cache, vector DB, checkpoints, observability, audit. Repeat questions burn tokens; compliance asks why the agent answered, and logs are not enough.

    Token burn

    Every similar intent re-hits the model.

    Audit gap

    No per-hop decision trace.

    Integration tax

    LangGraph + Redis + Pinecone + …

    Step 2 of 6

    Install — one package

    ChorusGraph is a native runtime (not a LangGraph wrapper). Python 3.11+.

    $ pip install chorusgraph $ python -c "import chorusgraph; print(chorusgraph.__version__)" 1.0.3

    Optional: pip install "chorusgraph[retrieval,gemini]" for PrismRAG + live Gemini examples.

    Step 3 of 6

    Run the demo — no API key

    chorusgraph-demo exercises routing, envelopes, and conditional edges on the native Graph engine.

    $ chorusgraph-demo === short path (message='hi') === { "result": { "response": "short:hi", "route": "short_path", ... }, "ledger": { "run_id": "...", "steps": [ ... ] }, "persisted_match": true } === long path (message='hello world!') === { "result": { "response": "long:hello world!", "route": "long_path", ... }, ... }

    Same graph, two routes — scored by message length. This is the native BSP scheduler, not LangGraph.

    Step 4 of 6

    Route Ledger — replay every hop

    Every node publishes artifacts with rule_chain and category. The ledger persists for audit and replay.

    "steps": [ { "hop": "analyze", "category_slug": "general", ... }, { "hop": "route_decision", "rule_chain": ["score_gt_5", "route=long_path"], ... }, { "hop": "long_path", "category_slug": "long_path", ... } ]

    Production teams use this for “why did the agent say that?” — sovereignty-friendly, no third-party trace cloud required.

    Step 5 of 6

    Measure cache savings (offline)

    chorusgraph-audit simulates semantic cache hit rate from a query log — no live LLM calls.

    $ chorusgraph-audit --log your_queries.jsonl Estimated cache hit rate: 42% Projected LLM call reduction: 38% ...

    Pair with staging traffic before you flip cache on in production.

    Step 6 of 6

    Verified benchmarks

    Paired Azure runs vs LangGraph baselines — same tasks, only orchestration differs. Canonical mid tier: mid_20260708_111539.

    ScenarioLangGraphChorusGraph
    Finance single87%98%
    Finance multi87%94%
    Healthcare single74%79%
    Healthcare multi59%85%

    Full tables: benchmarks page · latency summary

    Star on GitHub Install guide Book an audit
    Product Hunt / embed: Use this URL as your interactive demo link — . Optional: record this walkthrough in Supademo or Arcade and embed on this page later.