Local API reference
Connect an API client while keeping model requests separate from full agent execution.
Your space → Developer Tools → Server
A compatible Mellow build with the relevant capability configured.
Read the address
Copy the server address displayed in Mellow. Do not assume an example port matches your installation. Keep the server local unless you intend to configure remote access.
List available models
Send a GET request to the server’s /v1/models endpoint. If authentication is enabled, use the credential configured for that server.
Send a small request
Set model to a returned ID and messages to a short user message. Check the HTTP status and response before enabling streaming or tool calls.
Choose the execution contract
Use raw model endpoints when your client manages context and tools. Use an authorized agent endpoint when you need the configured Mellow agent behavior.
Controls & reference
| Control or concept | How to use it |
|---|---|
| GET /v1/models | Lists the models available through the server. Use a returned identifier rather than guessing a model name. |
| POST /v1/chat/completions | Chat-completions requests; stream: true requests a streamed response. The client executes returned tool calls. |
| POST /v1/responses | Responses-compatible requests for clients using that protocol. |
| POST /agents/{id}/run | Server-side agent execution subject to its access and tool policies. |
| GET /admin/config/export | Export configuration through a loopback-only administrative route. |
| POST /admin/config/plan | Preview a configuration change without applying it. |
| POST /admin/config/apply | Apply a validated configuration change; high-risk changes require explicit confirmation. |
Try it with a small task
Example request body for /v1/chat/completions: {"model":"<returned-model-id>","messages":[{"role":"user","content":"Say hello in one sentence."}],"stream":false}
If it does not work
- External HTTP callers do not automatically receive permission to perform file writes or shell actions.
- In-app memory and skills are not automatically injected into a raw model request.
- Administrative configuration routes are restricted to loopback; a remote key does not remove that restriction.
Good to know
- External HTTP callers do not automatically receive permission to perform file writes or shell actions.
- In-app memory and skills are not automatically injected into a raw model request.
- Administrative configuration routes are restricted to loopback; a remote key does not remove that restriction.