Multi-Agent Wiki
Decision

Market / Auction / Contract Net

Tasks and resources are allocated via bidding, pricing, or contract-net protocol.

Live visualizationAnimated topology — press Space to play / pause
·Diagram
Live
Variant
1 / 6

—

Definition

Agents allocate tasks and resources through bidding, pricing, or the Contract Net Protocol.

Category: Decision

When to use

Resource scheduling, tool-cost optimization, robotic task allocation, multi-agent load balancing.

When not to use

When bids can't be trustworthy estimates, tasks are tiny, or the negotiation cost outweighs execution.

How to implement

  1. The manager publishes a task announcement with goal, constraints, budget, and acceptance criteria.
  2. Agents return bids: cost, ETA, confidence, required permissions.
  3. The manager picks by a scoring function.
  4. Update agent reputation after completion — discourages chronic low-balling.

Minimal pseudocode

TypeScript
const bids = await Promise.all(agents.map(a => a.bid(task)));
const winner = rank(bids, b => b.confidence / Math.max(b.cost, 1))[0];
const result = await winner.agent.run(task);
reputation.update(winner.agent, result);

Recommended trace events

  • auction.announced
  • auction.bid.received
  • auction.awarded
  • auction.completed

Common failure modes

  • Agents underestimate cost.
  • The scoring function gets gamed instead of solving the real task.
  • Negotiation tokens dominate the budget.

Implementation checklist

  • Input/output schemas defined.
  • Each agent's permission boundary defined.
  • Every agent call carries a run id / trace id.
  • Failure, timeout, cancel, and retry strategies defined.
  • Context passed is the minimum required, not the full history.
  • High-risk actions are gated by approval or a verifier.

References