# OpenAgents — inference API

> OpenAgents is an OpenAI-compatible LLM inference API, pay-per-call. Point any
> OpenAI client at the endpoint below, choose `openagents/khala`, and you get
> a completion plus a verifiable usage receipt. No subscription; you
> pay per request for exactly the tokens you use.

## What this is

OpenAgents runs the **Agent Cloud** — inference, verified coding outcomes, and
agent compute, sold per call.

- OpenAI-compatible Chat Completions endpoint: https://openagents.com/api/v1/chat/completions
- Model catalog (OpenAI `/v1/models` shape): https://openagents.com/api/v1/models
- Pay-per-call. Microtransaction-friendly. Receipt-first metering (you are
  charged for the provider-reported usage, never an estimate).

## Models

- `openagents/khala` — the one public model id. Khala routes over OpenAgents
  owned supply, including Hydralisk/GPT-OSS under the hood, and returns a
  standard OpenAI response plus the `openagents` receipt/disclosure block.
  Raw `openai/gpt-oss-*` ids and old Khala tier names are not public products
  and are not MPP-payable.

## Best fit

- An agent that wants OpenAI-compatible inference and to pay per request
  instead of managing a subscription or a prepaid key.
- Coding agents that want a completion plus a verified-outcome receipt.

## Not a fit

- If you need a long-lived dedicated deployment or fine-tuned private weights,
  that is separate Agent Cloud surface (fine-tuning / sandboxes), not Khala.

## Pricing

Per-call, derived from a published per-token model (1 credit = $0.01 USD basis;
Bitcoin-funded balances get a small funding discount). See: https://openagents.com/docs/product-promises

## Pay per call (machine payments)

The machine-payable, 402-gated endpoint is:
https://openagents.com/api/mpp/v1/chat/completions (Machine Payments Protocol / x402). An agent with a Lightning
wallet, crypto wallet, or shared payment token can pay per request with no
signup: the first request returns `402 Payment Required` with a payment
challenge, the agent's payment tool pays and retries, and the verified request
returns the completion. The default and only MPP model is
`openagents/khala`; raw GPT-OSS ids are internal supply only and are rejected
before any payment challenge is issued.
The contributor payout rail stays Bitcoin/Spark.

## Agent surfaces

- https://openagents.com/agents.md — how an agent calls Khala (and the machine-payment flow)
- https://openagents.com/ai.md — short AI instructions
- https://openagents.com/skill.md — skill / tool setup
- https://openagents.com/AGENTS.md — full agent onboarding