Providers and endpoints
Every model provider OHarness speaks to, with the base URL, the key and the protocol for each. Hosted vendors, gateways, and models running on your own machine.
This page is written in English. The terms it explains are the ones people search for in English, and a translated term of art is a changed term.
27 endpoints, read from the harness’s own catalogue rather than listed here by hand. Each speaks one of four wire formats, and every one of them drives the same agent loop, the same tools and the same permission rules — switching between them mid-session converts nothing.
Hosted
| Provider | Protocol | Key |
|---|---|---|
| Anthropic | Anthropic Messages API | ANTHROPIC_API_KEY |
| DeepSeek | OpenAI-compatible chat completions | DEEPSEEK_API_KEY |
| Google Gemini | Google Gemini API | GEMINI_API_KEY |
| Moonshot (Kimi) | OpenAI-compatible chat completions | MOONSHOT_API_KEY |
| OHarness | OpenAI-compatible chat completions | OHARNESS_API_KEY |
| OpenAI | OpenAI Responses API | OPENAI_API_KEY |
| OpenRouter | OpenAI-compatible chat completions | OPENROUTER_API_KEY |
| OttoPort | OpenAI-compatible chat completions | OTTOPORT_API_KEY |
| Volcengine (Doubao) | OpenAI-compatible chat completions | ARK_API_KEY |
| Z.ai (GLM) | OpenAI-compatible chat completions | ZAI_API_KEY |
| Zhipu (GLM) | OpenAI-compatible chat completions | ZHIPU_API_KEY |
| Cerebras | OpenAI-compatible chat completions | CEREBRAS_API_KEY |
| DeepInfra | OpenAI-compatible chat completions | DEEPINFRA_API_KEY |
| Fireworks | OpenAI-compatible chat completions | FIREWORKS_API_KEY |
| Groq | OpenAI-compatible chat completions | GROQ_API_KEY |
| Hyperbolic | OpenAI-compatible chat completions | HYPERBOLIC_API_KEY |
| MiniMax | OpenAI-compatible chat completions | MINIMAX_API_KEY |
| Mistral | OpenAI-compatible chat completions | MISTRAL_API_KEY |
| Novita | OpenAI-compatible chat completions | NOVITA_API_KEY |
| NVIDIA NIM | OpenAI-compatible chat completions | NVIDIA_API_KEY |
| Perplexity | OpenAI-compatible chat completions | PERPLEXITY_API_KEY |
| SiliconFlow | OpenAI-compatible chat completions | SILICONFLOW_API_KEY |
| Together AI | OpenAI-compatible chat completions | TOGETHER_API_KEY |
| xAI | OpenAI-compatible chat completions | XAI_API_KEY |
On your own machine
No key, no account, and nothing billed by anyone. Start the server and the harness asks it for its own model list.
| Runtime | Default address |
|---|---|
| LM Studio | http://localhost:1234/v1 |
| Ollama | http://localhost:11434/v1 |
| vLLM | http://localhost:8000/v1 |
Anything else
The list above is what the harness ships knowing about, not what it can reach. Any endpoint speaking the OpenAI chat format works by pointing a base URL at it, which is also how a gateway or a self-hosted server is connected. The providers reference covers the catalogue and its override layers, and what an agent harness is covers why the model underneath is meant to be replaceable.