# Providers

<!-- Generated by scripts/gen-docs.mjs on 2026-08-21. Do not edit by hand. -->

anyagent knows **183 providers** and
**6673 models**,
from the [models.dev](https://models.dev) catalog plus a curated overlay.

```bash
anyagent providers                 # browse them all
anyagent providers --agent claude  # only those a given agent can use
anyagent models kimi               # search models
anyagent update                    # refresh from models.dev
```

Anything not listed still works with an explicit endpoint:

```bash
anyagent opencode --base-url https://my-gateway.internal/v1 -m my-model
```

## Anthropic-compatible providers

These serve an Anthropic Messages endpoint, which is what **Claude Code** needs.
The source column records how the endpoint was established: `probe` means the
route was confirmed to exist, `docs` means the vendor publishes it but the host
authenticates before routing so a probe cannot distinguish it, and `catalog`
means models.dev already described the provider as Anthropic-native.

| id | provider | endpoint | source |
| --- | --- | --- | --- |
| `alibaba` | Alibaba (DashScope) | https://dashscope-intl.aliyuncs.com/api/v2/apps/claude-code-proxy | docs |
| `anthropic` | Anthropic | https://api.anthropic.com | docs |
| `deepseek` | DeepSeek | https://api.deepseek.com/anthropic | docs |
| `freemodel` | FreeModel | https://cc.freemodel.dev/v1 | catalog |
| `kimi-for-coding` | Kimi For Coding | https://api.kimi.com/coding/v1 | catalog |
| `minimax` | MiniMax (minimax.io) | https://api.minimax.io/anthropic | probe |
| `minimax-cn` | MiniMax (minimaxi.com) | https://api.minimaxi.com/anthropic/v1 | catalog |
| `minimax-cn-coding-plan` | MiniMax Token Plan (minimaxi.com) | https://api.minimaxi.com/anthropic/v1 | catalog |
| `minimax-coding-plan` | MiniMax Token Plan (minimax.io) | https://api.minimax.io/anthropic/v1 | catalog |
| `moonshotai` | Moonshot AI | https://api.moonshot.ai/anthropic | probe |
| `moonshotai-cn` | Moonshot AI (China) | https://api.moonshot.cn/anthropic | docs |
| `novita-ai` | NovitaAI | https://api.novita.ai/anthropic | probe |
| `ollama` | Ollama (local) | http://127.0.0.1:11434 | catalog |
| `openrouter` | OpenRouter | https://openrouter.ai/api | probe |
| `siliconflow` | SiliconFlow | https://api.siliconflow.com | probe |
| `subconscious` | Subconscious | https://api.subconscious.dev/v1 | catalog |
| `thinkingmachines` | Thinking Machines | https://tinker.thinkingmachines.dev/services/tinker-prod/anthropic/api/v1 | catalog |
| `xai` | xAI | https://api.x.ai | probe |
| `zai` | Z.ai | https://api.z.ai/api/anthropic | docs |
| `zai-coding-plan` | Z.ai Coding Plan | https://api.z.ai/api/anthropic | docs |
| `zhipuai` | Zhipu AI | https://open.bigmodel.cn/api/anthropic | probe |

## OpenAI Responses providers

These serve `/responses`, which is what **Codex** and **Copilot CLI** need.
Codex removed `wire_api = "chat"`, so a chat-only provider cannot drive it.

| id | provider | endpoint | source |
| --- | --- | --- | --- |
| `deepseek` | DeepSeek | https://api.deepseek.com/v1 | docs |
| `fireworks-ai` | Fireworks AI | https://api.fireworks.ai/inference/v1 | probe |
| `groq` | Groq | https://api.groq.com/openai/v1 | probe |
| `meta` | Meta | https://api.meta.ai/v1 | catalog |
| `moonshotai` | Moonshot AI | https://api.moonshot.ai/v1 | probe |
| `openai` | OpenAI | https://api.openai.com/v1 | catalog |
| `openrouter` | OpenRouter | https://openrouter.ai/api/v1 | probe |
| `perplexity-agent` | Perplexity Agent | https://api.perplexity.ai/v1 | catalog |
| `togetherai` | Together AI | https://api.together.xyz/v1 | probe |
| `vercel` | Vercel AI Gateway | https://ai-gateway.vercel.sh/v1 | probe |
| `vivgrid` | Vivgrid | https://api.vivgrid.com/v1 | catalog |
| `xai` | xAI | https://api.x.ai/v1 | probe |
| `zai` | Z.ai | https://api.z.ai/api/paas/v4 | docs |

## Everything else

Every other catalogued provider serves OpenAI Chat Completions and works with
OpenCode, Droid, DeepSeek Harness, Pi, OpenClaw, Hermes, Cline, Qwen Code and Pool.

## Local runtimes

| id | endpoint | notes |
| --- | --- | --- |
| `ollama` | `http://127.0.0.1:11434` | serves both OpenAI- and Anthropic-compatible APIs, so it drives every agent |
| `lmstudio` | `http://127.0.0.1:1234/v1` | OpenAI-compatible |
| `llamacpp` | `http://127.0.0.1:8080/v1` | OpenAI-compatible |
| `vllm` | `http://127.0.0.1:8000/v1` | OpenAI-compatible |

No key is required for any of them.

## Providers anyagent cannot drive

| id | reason |
| --- | --- |
| `google-vertex` | Vertex AI uses Google application default credentials. |
| `google-vertex-anthropic` | Vertex AI uses Google application default credentials. |
| `amazon-bedrock` | Bedrock uses SigV4 request signing, not bearer tokens. |
| `azure-cognitive-services` | Azure OpenAI needs a per-deployment endpoint; use --base-url. |
| `azure` | Azure OpenAI needs a per-deployment endpoint; use --base-url. |

These authenticate with signed requests or cloud credentials rather than a bearer
token, which a base URL cannot express. Use the vendor's own gateway, or point
anyagent at a proxy that accepts a bearer token with `--base-url`.