Generic OpenAI-compatible
Anything exposing the OpenAI API shape can be added as a generic "OpenAI-compatible" provider. This includes:
- DeepSeek:
https://api.deepseek.com/v1 - Qwen (Alibaba):
https://dashscope-intl.aliyuncs.com/compatible-mode/v1 - Groq:
https://api.groq.com/openai/v1 - Together AI:
https://api.together.xyz/v1 - Fireworks:
https://api.fireworks.ai/inference/v1 - Cerebras:
https://api.cerebras.ai/v1 - LM Studio (self-hosted):
http://localhost:1234/v1 - vLLM (self-hosted):
http://your-server:8000/v1 - Ollama:
http://localhost:11434/v1
Adding one
Setyenv → Agent → Providers → Add Provider → Generic OpenAI-compatible.
Required:
- Base URL: e.g.
https://api.deepseek.com/v1. - API key: provider-specific. Self-hosted endpoints often accept any string or
EMPTY. - Model identifier: e.g.
deepseek-v4-flash,llama-3.1-70b-instruct.
Test connection probes POST {base_url}/chat/completions with a 1-token request. If it returns a valid choice, you're good.
Known caveats
- Tool use varies. Not every OpenAI-compatible endpoint implements
toolscorrectly. If the agent's tool calls come back as plain text instead of structured calls, that provider doesn't support tool use. Pick a different model. - Streaming: most do; some don't. The plugin auto-detects and falls back to non-streaming if needed.
- Token counting: some providers don't return usage stats. Budget tracking degrades to "best effort" for those.