Compatible Language Providers

Stuff Stash can run its inventory conversation through an OpenAI-compatible Chat Completions endpoint. The model must support function calling. Inventory permissions and approval of changes remain enforced by Stuff Stash.

Allow an Endpoint

Set STUFF_STASH_COMPATIBLE_PROVIDER_ENDPOINTS on the API to the exact API base URL, including its path. For example, a local runtime may expose http://model-server:8080/v1. The adapter appends /chat/completions. Multiple base URLs can be separated by commas. Restart the API after changing the allowlist. Choose operator-controlled hosts; this setting grants network access to those destinations. Redirects are rejected.

Remote openai_compatible profiles require HTTPS. local_http profiles may use HTTP for an explicitly allowed endpoint on your own network. Endpoint and model names have no implicit defaults.

Create and Test a Profile

As a household administrator, create a Language inference provider profile:

1. Choose OpenAI compatible or Local HTTP.
2. Enter the same base URL the operator allowed and a model the endpoint serves.
3. Save an API key or OAuth bearer credential through the profile’s credential action. For a local runtime that ignores authentication, save a non-secret marker; the adapter still sends it as a Bearer value.
4. Run the provider test, which sends a short synthetic function-call request.
5. Select the tested profile as the language provider in your voice configuration or conversation workflow.

An optional runtime setting such as {"httpTimeout":"60s"} controls the request timeout. It uses Go duration syntax and must be positive. Requests and responses are each limited to 1 MiB. Provider errors and malformed tool output fail safely.

Capability Limits

This adapter provides language inference. Keep separate speech recognition and speech synthesis profiles for voice sessions. A runtime that serves plain chat but cannot return function calls will fail the diagnostic. Protocol compatibility does not establish a model’s inventory reasoning quality; test your normal read and approval workflows with the chosen model before relying on it.

Stuff Stash uses the Chat Completions function-calling contract. It does not require provider-hosted tools or automatically execute model output.