Skip to content

Latest commit

 

History

History
203 lines (174 loc) · 9.79 KB

PROVIDER_SUPPORT_MATRIX.md

File metadata and controls

203 lines (174 loc) · 9.79 KB

Provider Support Matrix (Rust Runtime)

Last audited: 2026-02-25

This matrix is generated from the runtime implementation, not marketing labels:

  • src/gateway.rs: normalize_provider_id, provider_runtime_defaults, provider_runtime_bridge_defaults, OAUTH_PROVIDER_CATALOG
  • src/website_bridge.rs: guest bridge routing (zai, qwen, inception) and bridge request path selection

Status legend:

  • Built-in default: runtime has a canonical default base_url + API mode and env-var auth lookup.
  • Alias + config: alias is normalized, but you must set endpoint/auth in config (models.providers.<id>).
  • Bridge default: runtime seeds website bridge hints (websiteUrl and/or bridgeBaseUrls).
  • OAuth catalog: appears in auth.oauth.providers/start/wait/complete/logout/import.

Runtime security note (v1.7.0):

  • Provider routing/output now flows through the expanded tool runtime safety stack (security.wasm, dynamic WIT registry, credential leak redaction, and SafetyLayer output controls).
  • models.dev naming variants are normalized to canonical runtime IDs for configured providers, including:
    • fireworks-ai -> fireworks
    • moonshotai / moonshotai-cn -> moonshot
    • novita-ai -> novita
    • opencode-go -> opencode
    • kimi-for-coding -> kimi-coding
    • inference -> inference-net

Requested Coverage Matrix

Provider Runtime status Canonical id
Ollama Built-in default ollama
LM Studio Built-in default lmstudio
vLLM Built-in default vllm
llama.cpp server Built-in default llamacpp
LocalAI Built-in default localai
Hugging Face TGI Built-in default tgi
GPT4All Built-in default gpt4all
KoboldCPP Built-in default koboldcpp
Oobabooga Built-in default oobabooga
Groq Built-in default groq
Google Gemini (AI Studio OpenAI-compat) Built-in default google
DeepSeek Built-in default deepseek
Mistral Built-in default mistral
Fireworks Built-in default fireworks
Together AI Built-in default together
DeepInfra Built-in default deepinfra
Cerebras Built-in default cerebras
SiliconFlow Built-in default siliconflow
SambaNova Built-in default sambanova
Novita Built-in default novita
Hyperbolic Built-in default hyperbolic
Nebius Built-in default nebius
Inference.net Built-in default inference-net
OpenRouter Built-in default openrouter
Inception Mercury Built-in default inception
AIMLAPI Built-in default aimlapi
Vercel AI Gateway Alias + config vercel-ai-gateway
ShareAI Alias + config shareai
Bifrost (Maxim) Alias + config bifrost
OpenAI Built-in default openai
Anthropic (Claude) Alias + config, OAuth catalog anthropic
Azure OpenAI Alias + config azure-openai
Google Vertex AI Alias + config vertex-ai
Amazon Bedrock Alias + config bedrock
Cohere Built-in default (compat endpoint) cohere
xAI Grok Built-in default xai
NVIDIA NIM Built-in default nvidia
GitHub Models Alias + config github-models
OpenCode Zen Built-in default + bridge default opencode
OpenRouter free models Built-in default + model id support openrouter

Notes:

  • zai / zhipuai guest website bridge is implemented (chat.z.ai path support).
  • qwen-portal guest website bridge fallback is implemented (chat.qwen.ai path support).
  • inception / mercury guest website bridge fallback is implemented (chat.inceptionlabs.ai path support).
  • kimi-coding has a built-in API endpoint default and OAuth catalog support, but guest bridge is login/session-gated in practice.

Exhaustive Built-in Runtime Defaults (Canonical IDs)

All entries below resolve to api_mode = openai-completions in runtime.

Canonical provider Default base URL API key optional
aimlapi https://api.aimlapi.com/v1 no
byteplus https://api.byteplus.com/v1 no
cerebras https://api.cerebras.ai/v1 no
cohere https://api.cohere.com/compatibility/v1 no
deepinfra https://api.deepinfra.com/v1/openai no
deepseek https://api.deepseek.com/v1 no
fireworks https://api.fireworks.ai/inference/v1 no
google https://generativelanguage.googleapis.com/v1beta/openai no
gpt4all http://127.0.0.1:4891/v1 yes
groq https://api.groq.com/openai/v1 no
huggingface https://api-inference.huggingface.co/v1 no
hyperbolic https://api.hyperbolic.xyz/v1 no
inference-net https://api.inference.net/v1 no
inception https://api.inceptionlabs.ai/v1 yes (bridge fallback)
kimi-coding https://api.kimi.com/coding no
koboldcpp http://127.0.0.1:5001/v1 yes
litellm http://127.0.0.1:4000/v1 yes
llamacpp http://127.0.0.1:8080/v1 yes
lmstudio http://127.0.0.1:1234/v1 yes
localai http://127.0.0.1:8080/v1 yes
mistral https://api.mistral.ai/v1 no
moonshot https://api.moonshot.ai/v1 no
nebius https://api.studio.nebius.com/v1 no
novita https://api.novita.ai/v3/openai no
nvidia https://integrate.api.nvidia.com/v1 no
ollama http://127.0.0.1:11434/v1 yes
oobabooga http://127.0.0.1:5000/v1 yes
openai https://api.openai.com/v1 no
openai-codex https://api.openai.com/v1 no
opencode https://opencode.ai/zen/v1 yes
openrouter https://openrouter.ai/api/v1 no
perplexity https://api.perplexity.ai no
qianfan https://qianfan.baidubce.com/v2 no
qwen-portal https://portal.qwen.ai/v1 yes (bridge fallback)
sambanova https://api.sambanova.ai/v1 no
siliconflow https://api.siliconflow.cn/v1 no
tgi http://127.0.0.1:8080/v1 yes
together https://api.together.xyz/v1 no
vllm http://127.0.0.1:8000/v1 yes
volcengine https://ark.cn-beijing.volces.com/api/v3 no
xai https://api.x.ai/v1 no
zai https://api.z.ai/v1 yes (bridge fallback)
zhipuai https://open.bigmodel.cn/api/paas/v4 yes (bridge fallback)
zhipuai-coding https://open.bigmodel.cn/api/coding/paas/v4 yes (bridge fallback)

Alias + Config Required Providers

These canonical IDs are normalized from aliases, but do not have built-in default base URLs in provider_runtime_defaults.

Canonical provider Example aliases
anthropic claude, claude-code, claude-cli
azure-openai azure, azure-openai-service
vertex-ai vertex, google-vertex
bedrock amazon-bedrock
github-models github-model
vercel-ai-gateway vercel-ai, ai-gateway
shareai share-ai
bifrost maxim-bifrost, bifrost-maxim

Website Bridge Defaults

Canonical provider Default websiteUrl Default bridge candidates
opencode https://opencode.ai https://opencode.ai/zen/v1, https://api.opencode.ai/v1
qwen-portal https://chat.qwen.ai https://chat.qwen.ai
zai https://chat.z.ai https://chat.z.ai
zhipuai https://chat.z.ai https://chat.z.ai
zhipuai-coding https://chat.z.ai https://chat.z.ai
kimi-coding https://www.kimi.com none pre-seeded
minimax-portal https://chat.minimax.io none pre-seeded
inception https://chat.inceptionlabs.ai https://api.inceptionlabs.ai/v1

Implementation detail:

  • Zhipu/Z.ai, Qwen, and Inception guest bridge fallbacks are explicitly recognized in src/website_bridge.rs.
  • Kimi/Minimax website hints are surfaced for configured bridge mode, but practical usage is login/session dependent.

OAuth Provider Catalog (RPC Surface)

auth.oauth.providers currently advertises these provider IDs:

  • openai
  • openai-codex
  • anthropic
  • google-gemini-cli
  • qwen-portal
  • minimax-portal
  • kimi-coding
  • opencode
  • zhipuai

Endpoint Documentation References (Audited)

The following official documentation sets were reviewed in parallel for endpoint/auth alignment:

For providers that are alias-only/config-required in this runtime, these docs should be used to supply models.providers.<id>.baseUrl, auth headers, and any provider-specific request defaults.