Ship production agents without the integration tax.

You're paying to re-answer the same question — and you still can't replay why the agent said it. ChorusGraph is one runtime: semantic cache, retrieval, memory, checkpoints, and Route Ledger audit.

+26 pp healthcare multi-agent 98% finance single 329+ tests, no API keys pip install chorusgraph
Try the interactive demo Book a 30-Min Agent Stack Audit Read the Whitepaper View on GitHub

The pain

Token burn

LLM line items grow faster than revenue when every repeat intent re-hits the model.

Audit gap

Compliance asks for replay; you have logs, not decision traces per hop.

Integration tax

LangGraph + Redis + Pinecone + checkpoints + audit = three teams and six repos.

Proof (verified Azure run mid_20260708_111539)

100 tasks/scenario · real Gemini · benchmark-only methodology fixes (no library release)

ScenarioLangGraphChorusGraph
Finance single87.0%98.0%
Finance multi87.0%94.0%
Healthcare single74.0%79.0%
Healthcare multi59.0%85.0%

LLM calls −48% to −76% on finance; full latency tables: summary · benchmarks · fairness disclosure

Plan — three steps to staging

  1. Install — pip install "chorusgraph[retrieval]"
  2. Wire — ChorusStack.with_retrieval(PrismRAGRetrievalBackend(...))
  3. Measure — Route Ledger + cache hit report on staging traffic

Or let your AI IDE do it

Paste our Cursor/Claude Code prompt — it installs and wires ChorusGraph in your project. AI IDE prompts

pip install chorusgraph
python -c "import chorusgraph; from chorusgraph import Graph, ChorusStack; print(chorusgraph.__version__)"

Five swappable ports

Cache · Memory · Tools · Retrieval · Persistence (SQLite free; Postgres Enterprise with offline license)