# Endpoint names and requirements

Checked against official provider documentation on 2026-10-17. This is a configuration matrix, not a claim of account access or successful inference. Provider models and access policies can change; use the authenticated model picker for your account. Selecting an endpoint must not silently choose a paid model.

Type `/endpoint` to inspect choices or `/endpoint <name>` for a registered provider. Names and aliases are case-insensitive. An existing configured endpoint for the same provider retains its custom URL/port and exact-scope credentials. Switching providers clears the old credential/key pool; required missing keys use the existing secret input. Explicit URLs and custom protocol configuration remain available.

For example, use `/endpoint ollama`, `/endpoint openai`, `/endpoint claude` or `/endpoint gemini`. Type a prefix such as `/endpoint cl` and choose a suggestion with Tab or the pointer; completion fills the input without connecting. Enter applies the command. Add `--local` to save an endpoint only for the current project. Required keys are entered through the secret prompt, rather than placed in the command history.

| Name / aliases | Default base URL | Protocol and credential requirement | Official source |
| --- | --- | --- | --- |
| `ollama` | `http://127.0.0.1:11434` | Native Ollama. Local API has no default authentication; an authenticated proxy may require its own token. Server and selected model must exist. Cloud models have separate Ollama sign-in/key requirements. | [Authentication](https://docs.ollama.com/api/authentication) |
| `lmstudio`, `lm-studio` | `http://127.0.0.1:1234/v1` | OpenAI-compatible local server; start its server and select a model. Send a bearer token when server authentication is enabled. | [Compatibility](https://lmstudio.ai/docs/developer/openai-compat), [authentication](https://lmstudio.ai/docs/developer/core/authentication) |
| `vllm`, `v-llm` | `http://127.0.0.1:8000/v1` | OpenAI-compatible serving endpoint. Use the server's bearer key if configured; exact model, chat template and tool parser depend on that deployment. Setup does not start a CUDA service. | [Server](https://docs.vllm.ai/en/latest/serving/online_serving/openai_compatible_server/), [server flags](https://docs.vllm.ai/en/latest/cli/serve/) |
| `openai` | `https://api.openai.com/v1` | OpenAI Chat Completions; your API bearer credential and authorized model. Organization/project selection can require additional account-scoping headers. | [API authentication](https://developers.openai.com/api/reference/overview), [chat API](https://developers.openai.com/api/reference/resources/chat/subresources/completions/methods/create) |
| `anthropic`, `claude` | `https://api.anthropic.com/v1` | Anthropic Messages; `x-api-key` plus `anthropic-version: 2023-06-01`. This preset uses workspace-bound API keys; multi-workspace keys require an additional workspace header and are not implied by entering a key alone. | [API overview](https://platform.claude.com/docs/en/api/overview) |
| `gemini`, `google`, `google-ai` | `https://generativelanguage.googleapis.com/v1beta/openai` | Gemini's OpenAI-compatible API, with your Gemini API key as bearer credential; not its distinct native request schema. | [OpenAI compatibility](https://ai.google.dev/gemini-api/docs/openai) |
| `openrouter` | `https://openrouter.ai/api/v1` | OpenAI-compatible bearer API key. Referral/title headers are optional. Account credit/limits and selected model pricing remain authoritative. | [Authentication](https://openrouter.ai/docs/api_reference/authentication) |
| `groq` | `https://api.groq.com/openai/v1` | OpenAI-compatible API with your Groq API key. Provider compatibility and model capabilities still apply. | [Compatibility](https://console.groq.com/docs/openai) |
| `deepinfra` | `https://api.deepinfra.com/v1/openai` | OpenAI-compatible API with your DeepInfra token and an exact catalog model. | [Chat API](https://docs.deepinfra.com/chat/overview) |
| `fireworks`, `fireworks-ai` | `https://api.fireworks.ai/inference/v1` | OpenAI-compatible API with bearer API key and accessible model/deployment. | [Chat endpoint](https://docs.fireworks.ai/api-reference/post-chatcompletions), [authentication](https://docs.fireworks.ai/api-reference/introduction) |
| `together`, `together-ai` | `https://api.together.ai/v1` | OpenAI-compatible API with your Together API key. Recognize previously saved `api.together.xyz` URLs without silently replacing an explicit configured endpoint. | [Compatibility](https://docs.together.ai/docs/inference/openai-compatibility) |
| `chutes`, `chutes-ai` | `https://llm.chutes.ai/v1` | OpenAI-compatible API with bearer Chutes key. Use concrete catalog model IDs; saved routing aliases need account-side configuration. | [Agent connection requirements](https://chutes.ai/agents/connect) |
| `mistral`, `mistral-ai` | `https://api.mistral.ai/v1` | OpenAI-compatible chat API and your Mistral bearer API key; account mode, model access and rate limits apply. | [API key setup](https://docs.mistral.ai/getting-started/quickstarts/studio/activate-and-generate-api-key) |
| `cerebras` | `https://api.cerebras.ai/v1` | OpenAI-compatible API with your Cerebras key. Shared inference uses exact model IDs; dedicated inference can use an endpoint ID. | [Compatibility](https://inference-docs.cerebras.ai/resources/openai) |
| `sambanova`, `samba-nova` | `https://api.sambanova.ai/v1` | SambaCloud OpenAI-compatible bearer API key. Dedicated endpoints may use a different base URL; legacy SambaStudio APIs have different requirements and are not this preset. | [Official integration](https://docs.sambanova.ai/docs/en/integrations/vscode) |
| `nvidia`, `nvidia-nim`, `nim` | `https://integrate.api.nvidia.com/v1` | NVIDIA hosted OpenAI-compatible chat API and bearer NVIDIA API key with model access/credit. Self-hosted NIM endpoints are explicitly configured URLs. | [Hosted chat reference](https://docs.api.nvidia.com/nim/reference/stepfun-ai-step-5-preview-infer) |
| `builtin-free` | `https://opencode.ai/inference/openai/v1` | Published public access remains subject to provider permission. Discovery is separately pinned to `/inference/v1/models`. Current independent Omnius and genuine OpenCode controls returned403; catalog membership does not establish usable generation. | [Inference API](https://opencode.ai/v2/docs/console/api/inference/) |
| `mesh` | Configured loopback gateway; default `http://127.0.0.1:11436/v1` | Omnius OpenAI-compatible sponsored inference. On explicit selection, read the enabled owner-private default mesh configuration and its gateway token. Existing explicitly configured mesh URLs retain their own credential scope. The gateway must already be running; endpoint selection starts no service. | [Setup and trust boundaries](sponsored-inference-mesh.md) |
| `hyperbolic` | Retired serverless preset | Official documentation says the old inference API and playground are retired, and old requests fail even with valid keys. Do not offer this as a working shortcut. Operator-hosted replacement servers use an explicit custom URL/protocol. | [Retirement notice](https://www.hyperbolic.ai/docs/faq/inference-models) |

Custom endpoints require a URL and an explicit supported protocol when detection is ambiguous: `ollama`, `openai-chat`, or `anthropic-messages`. A local server's optional bearer authentication is not a reason to discard a supplied credential. No API key is inferred from a different provider, and endpoint discovery does not perform generation, model downloads or hardware deployment.

The standard presets cover the auth forms above. They do not claim every provider's OAuth, federation, multi-workspace, tenant or administrative API is implemented. Configure a supported scoped API credential or an explicitly authorized compatible gateway for those deployments.
