Skip to content

AuctionSI

An open-source marketplace and auction framework for autonomous agents.

Instead of a central orchestrator deciding which agent does a task, AuctionSI announces the task to a market. Capable agents bid, an auction mechanism picks the winner, and the work is contracted, executed, verified, settled and fed back into reputation.

Task → discovery → bids → auction → winner → contract → execution
     → verification → settlement → reputation

Agents can be Python functions, HTTP services, solvers, ML models, humans, LLMs or simulations. Nothing requires an LLM, an API key or a cloud service: everything runs locally, and seeded simulations are deterministic.

pip install auctionsi

What you can do with it

  • Allocate work between agents with auctions. First-price and second-price (Vickrey-style, critical-value) reverse auctions, multi-winner, open descending, forward auctions with reserve prices, bundle (combinatorial) auctions with exact winner determination, and capacity (multi-unit) procurement. See auction mechanisms.
  • Choose winners on more than price. Explainable policies weigh price, quality, latency, reliability and reputation; every score is split into named contributions. See winner selection.
  • Trust results only after checking them. Verifiers (schema, exact match, tolerance, metrics, unit tests, human approval) decide payment; settlement applies penalties and bonuses; reputation learns from delivered work. See core concepts.
  • Study markets before you run them. Simulate thousands of agents with bidding strategies (including a learning bidder), collusion, Sybil identities and misreporting, then run replicated experiments with confidence intervals and Holm-adjusted tests. See simulation and experiments and the research results.
  • Operate it. A CLI, a REST/WebSocket API with a web dashboard, SQLite or PostgreSQL storage, an append-only event log and deterministic replay of every decision. See CLI, API and storage.

Documentation map