> ## Documentation Index > Fetch the complete documentation index at: https://docs.wandb.ai/llms.txt > Use this file to discover all available pages before exploring further. # Add custom runtimes > Connect OpenAI-compatible inference endpoints to a Weave project as custom runtimes, and manage them from the UI or the Python and TypeScript SDKs. Register a custom runtime to connect an OpenAI-compatible inference endpoint to a W\&B Weave project. After registration, the endpoint’s models appear in model selectors across Weave. You and your team can then use them in the Playground, Evaluation Playground, and LLM-as-a-judge workflows without writing integration code. Use a custom runtime to connect any of the following: * An agent or model prototype your team serves behind an OpenAI-compatible API. * A self-hosted or local model (for example, one served by Ollama or vLLM). * A provider or model version that Weave doesn't have built-in support for. You can manage custom runtimes in the W\&B UI or programmatically with the Python and TypeScript SDKs. The SDK path is designed for CI/CD: registration is declarative and idempotent, so your pipeline can re-register a runtime on every deploy and Weave converges to the configuration you specify. ## Where you can use custom runtimes Models from a registered custom runtime appear in the model selector in the following places: * [Playground](/weave/guides/tools/playground): Chat with the endpoint interactively. * [Evaluation Playground](/weave/guides/tools/evaluation_playground): Evaluate and compare the endpoint against other models. * [Signals](/weave/guides/evaluation/monitors): Use the endpoint as the judge model that scores incoming traces. * Scorer iteration and Agent Signals model selectors. * The Weave completions API: Pass the model ID `custom::::` to the `completions/create` endpoint to run traced completions against your endpoint programmatically. See the [service API reference](/weave/reference/service-api). ## Add a custom runtime You can add a custom runtime to your project using code or the Weave UI. To add a custom runtime, you need the following information about your runtime: | Field | Description | | ------------------------- | ---------------------------------------------------------------------------------------------------------- | | Name | Identifies the runtime within the project. Appears as the group name in model selectors. | | Base URL | The OpenAI-compatible endpoint Weave sends completion requests to. Must be reachable from W\&B's servers. | | Runtime IDs | One or more model IDs the endpoint serves. Each ID appears as a selectable model. | | Max tokens | The maximum number of tokens the model can generate in one response. Set per runtime ID. Defaults to 4096. | | API key secret (Optional) | The name of a [team secret](/platform/secrets) that holds the endpoint's API key. | | Headers (Optional) | Custom HTTP headers sent with every request to the endpoint. | ### Add a custom runtime using code Using the runtime's information outlined in the previous section, update the appropriate fields in the following registration request, and then run it to either create or update a custom runtime in your project. Programmatic registration requires the `weave` Python package version 0.53.4 or later, or the `weave` npm package version 0.16.5 or later. ```python theme={null} import weave client = weave.init("[YOUR-TEAM-NAME]/[YOUR-PROJECT-NAME]") client.register_custom_runtime( name="support-agent", base_url="https://agent.example/v1", api_key_secret="AGENT_API_KEY", headers={"X-Tenant-ID": "customer-1"}, runtime_ids=[ "support-v12", weave.CustomRuntimeID(id="support-canary", max_tokens=8192), ], ) ``` ```typescript theme={null} import * as weave from 'weave'; const client = await weave.init('[YOUR-TEAM-NAME]/[YOUR-PROJECT-NAME]'); await client.registerCustomRuntime({ name: 'support-agent', baseUrl: 'https://agent.example/v1', apiKeySecret: 'AGENT_API_KEY', headers: {'X-Tenant-ID': 'customer-1'}, runtimeIds: ['support-v12', {id: 'support-canary', maxTokens: 8192}], }); ``` After adding a runtime, its runtime ID is addressed as `custom::::`. After the call succeeds, the runtime's models appear in model selectors under the runtime's name, the same as a runtime registered in the UI. After you update a runtime, inference may continue to use the previous configuration for up to 60 seconds while caches expire. ## Register a custom runtime in the UI To register a custom runtime from the W\&B UI: 1. In the Weave sidebar, select **Project** and then select the **AI Providers** tab. Alternatively, in the Playground model selector, click **Configure providers**. 2. Click **Add custom runtime**. This opens the configuration drawer. 3. In the drawer, enter the runtime's name, base URL, and one or more runtime IDs. Optionally add an API key secret and custom headers. 4. Save the runtime. Custom runtime configuration menu The runtime's models then appear in model selectors under a group with the runtime's name. A runtime saved without any runtime IDs remains visible in the selector with an **Add runtime ID** action that reopens the editor. To edit or remove a runtime, use the edit and delete controls in the runtime's row on the **AI Providers** tab. Deleting a runtime can't be undone.