Skip to content

provider-matrix.md

Latest commit

 

History

History
77 lines (63 loc) · 3.49 KB

provider-matrix.md

File metadata and controls

77 lines (63 loc) · 3.49 KB
Raw

Provider Compatibility Matrix

LinuxAgent uses LangChain provider wrappers. Most non-Anthropic providers use the OpenAI-compatible wire format through langchain-openai; Anthropic-format providers require the optional anthropic extra.

This matrix documents the intended configuration path. A row marked compatible path means the provider should work through the listed protocol, but maintainers still want real endpoint reports before calling it fully verified.

Provider Protocol Typical base_url Token parameter API key Status
deepseek OpenAI-compatible https://api.deepseek.com/v1 max_completion_tokens Required default path
openai OpenAI https://api.openai.com/v1 max_completion_tokens Required supported
openai_compatible OpenAI-compatible relay relay /v1 URL often max_tokens Required supported
local OpenAI-compatible local http://127.0.0.1:8000/v1 max_tokens Optional compatible path
ollama OpenAI-compatible local http://127.0.0.1:11434/v1 max_tokens Optional compatible path
vllm OpenAI-compatible local http://127.0.0.1:8000/v1 max_tokens Optional compatible path
lmstudio OpenAI-compatible local http://127.0.0.1:1234/v1 max_tokens Optional compatible path
qwen OpenAI-compatible https://dashscope.aliyuncs.com/compatible-mode/v1 max_tokens Required compatible path
kimi OpenAI-compatible https://api.moonshot.ai/v1 max_tokens Required compatible path
glm OpenAI-compatible https://open.bigmodel.cn/api/paas/v4 max_tokens Required compatible path
minimax OpenAI-compatible https://api.minimax.io/v1 max_tokens Required compatible path
gemini OpenAI-compatible https://generativelanguage.googleapis.com/v1beta/openai/ max_tokens Required compatible path
hunyuan OpenAI-compatible https://api.hunyuan.cloud.tencent.com/v1 max_tokens Required compatible path
anthropic Anthropic provider default n/a Required optional extra
anthropic_compatible Anthropic-compatible relay relay URL n/a Required optional extra
xiaomi_mimo Anthropic-compatible relay relay URL n/a Required optional extra

Local Models

Local providers are shortcuts for OpenAI-compatible HTTP servers. They do not require a real API key; LinuxAgent passes a placeholder to satisfy client construction.

api:
  provider: ollama
  base_url: http://127.0.0.1:11434/v1
  model: llama3.1
  api_key: ""
  token_parameter: max_tokens

API Relays

When using a relay, prefer the generic path first:

api:
  provider: openai_compatible
  base_url: https://relay.example.com/v1
  model: relay-model
  api_key: "sk-..."
  token_parameter: max_tokens

If requests fail with an unknown parameter error, change api.token_parameter between max_tokens and max_completion_tokens.

Provider exceptions are kept as machine diagnostics. Runtime language localizes LinuxAgent-owned wrappers, summaries, and UI labels, but it does not translate raw provider error text, model output, prompt templates, or tool/MCP protocol fields.

Compatibility Reports

When reporting a provider as working or broken, include:

  • provider name and endpoint type
  • model name
  • token_parameter
  • whether streaming worked
  • whether command planning returned valid JSON
  • whether tool calls were used
  • sanitized error output if it failed