Free evaluation: November 18–December 2, 2026, 16:00 UTC. No payments or subscription. Read the evaluation policy before use.

AgentFoundry

Deterministic JSON argument repair for AI agents. Correct limited formatting and type errors before retrying a model. Ambiguous changes are refused; no tool is executed.

Try the evaluation API

The evaluation uses a signed, one-time challenge and a scoped session. Signing grants no payment authority. Ten lifetime attempts per principal apply, with global limits. Private keys never leave the caller.

  1. POST an address to /v1/evaluation/challenge.
  2. Sign the exact returned message using EIP-191.
  3. POST challenge_id and signature to /v1/evaluation/session.
  4. Send the returned bearer token with a repair request.
{"input":"{\"count\":\"12\"}","schema":{"type":"object","properties":{"count":{"type":"integer"}},"required":["count"],"additionalProperties":false}}

Expected correction: {"count":12}. The value "0012" is refused because leading zeroes may matter. Schema validity does not prove user intent.

Machine-readable documentation

OpenAPI · Service catalog · Pricing status

Value and limits

Compare total workflow cost and correctness against your existing native retries. No savings or paying demand has been demonstrated. $0.01/call is an unvalidated price hypothesis, not a commercial offer. Evaluation cannot automatically become paid usage.

Business and privacy status

Brand: AgentFoundry. Operator: HML Services, Indiana, United States. Support and privacy: HMLTrucks@gmail.com. Service: agentfoundry-eval.hmltrucks.workers.dev. Privacy, retention, support and refund policy. Use fictional data only; do not submit credentials or confidential material. This evaluation stores hashed identifiers and quota/outcome metadata, not request arguments.

Start in a few minutes

Runnable Node.js client and MCP setup · Agent instructions. No funded wallet or payment is needed. Server-side integrations only.