OpenAI-compatible API

Any tool that lets you set a custom OpenAI base URL can talk to your agent. It answers with its knowledge, Q&A, instructions, guardrails and actions.

List models

Your agents, in OpenAI's format:

{
  "object": "list",
  "data": [
    { "id": "3f2c9a8e-…", "object": "model", "created": 1746173662, "owned_by": "anytimereply", "name": "Support" }
  ]
}

Chat completions

curl https://anytimereply.com/api/openai/v1/chat/completions \
  -H "Authorization: Bearer af_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "AGENT_ID", "user": "customer-42", "messages": [{"role": "user", "content": "Opening hours?"}]}'

{
  "id": "chatcmpl-5f1c…",
  "object": "chat.completion",
  "created": 1760865600,
  "model": "AGENT_ID",
  "choices": [
    { "index": 0, "message": { "role": "assistant", "content": "We're open every day from 10 am to 8 pm." }, "finish_reason": "stop" }
  ],
  "usage": { "prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0 }
}

How the request is used

Other OpenAI parameters (temperature, max_tokens…) are accepted but ignored — the agent's own settings apply. usage is always zero; credits are counted in your dashboard.

Your tools (handoff to your platform)

Tools you pass are offered to the agent. When it decides to use one — for example a transfer to human tool — the response has finish_reason: "tool_calls" and message.tool_calls for your app to run. Send the result back as usual (assistant message with the tool_calls, then a tool message) and the agent tells the customer what happened.

Through this API the agent never runs its own human handoff: the calling platform owns handoff. If you pass a tool whose name or description mentions a human/agent transfer, the agent is told to call it when a customer asks for a person or is upset.

Errors

Errors use OpenAI's format so SDKs handle them:

{ "error": { "message": "The model \"abc\" does not exist. Use one of your agent IDs as the model (GET /models lists them).", "type": "not_found_error", "code": null } }

type is authentication_error (401), rate_limit_error (429), not_found_error (404), invalid_request_error (400) or server_error (500).