Tested Provider List
This page lists AI services and developer tools that Patronus can identify or observe when their traffic passes through an active Patronus protection path.
Status Legend
Section titled “Status Legend”| Status | Meaning |
|---|---|
| Supported | Patronus can identify the surface and show AI activity for supported request and response flows. |
| Detected | Patronus can identify the surface as AI-related. The amount of visible detail may depend on the app, transport, or account setup. |
| Compatible API | Patronus can observe the traffic when the tool uses a supported OpenAI-compatible, Anthropic-compatible, Gemini-compatible, or local model API shape. |
Protocols And Encodings
Section titled “Protocols And Encodings”| Protocol or encoding | Status | What customers can expect to see |
|---|---|---|
| HTTP/1.0 and HTTP/1.1 | Supported | JSON requests and responses, including content-length and chunked response flows. |
| HTTP/2 | Supported | JSON, concurrent HTTP/2 requests, large payloads, streaming JSONL responses, and compressed request or response bodies. |
| HTTPS through Patronus protection | Supported | Provider traffic is visible when the app or runtime trusts the Patronus CA and does not bypass the active protection path. |
| Server-Sent Events | Supported | Streaming AI responses from providers such as OpenAI-compatible APIs, Anthropic, Gemini, Mistral, Perplexity, Bedrock, Copilot, and similar event streams. |
| WebSocket | Supported | Text frames, UTF-8 JSON binary frames, SignalR-style frames, MessagePack frames, and selected real-time provider streams. |
| gRPC | Supported | Unary and server-streaming protobuf payloads where Patronus can decode the message shape. |
| JSON | Supported | Standard request and response bodies used by most provider APIs. |
| JSONL / NDJSON / JSON sequence | Supported | Token-style streamed responses and line-delimited generated output. |
| URL-encoded form bodies | Supported | Prompt fields and single embedded JSON fields used by some browser and provider flows. |
| Multipart form data | Supported | JSON fields inside multipart requests. Binary file content is not treated as prompt text by default. |
| Protobuf | Supported | Structured protobuf bodies and common AI-specific protobuf-like message shapes. |
| MessagePack | Supported | Binary MessagePack frames used by selected web app and real-time agent surfaces. |
| AWS EventStream | Supported | Bedrock streaming responses, including common Converse and model stream formats. |
| gzip | Supported | Compressed request and response bodies. |
| Brotli | Supported | Compressed request and response bodies. |
| Zstandard | Supported | Compressed request and response bodies. |
Hosted AI Providers
Section titled “Hosted AI Providers”| Provider | Status | What customers can expect to see |
|---|---|---|
| OpenAI API | Supported | Chat Completions, Responses API, streaming responses, tool calls, and tool results. |
| ChatGPT | Supported | ChatGPT web and Codex-related AI activity when routed through Patronus. |
| Anthropic / Claude API | Supported | Messages API traffic, streaming responses, tool use, and tool results. |
| Claude Web | Supported | Claude-related AI traffic when the app traffic is visible to Patronus. |
| Google Gemini API | Supported | Generate Content and streaming Gemini API traffic. |
| Gemini Web | Supported | Gemini web activity, including streamed response content where available. |
| Google Vertex AI | Supported | Vertex AI model requests when they use supported Gemini-style request and response shapes. |
| NotebookLM | Supported | NotebookLM generated answer streams when routed through Patronus. |
| Microsoft Copilot | Supported | Copilot web and Microsoft 365 Copilot chat activity where traffic is visible to Patronus. |
| GitHub Copilot | Supported | Copilot API and CLI-style activity, including tool-oriented events where available. |
| Azure OpenAI / Azure AI Foundry | Supported | Azure-hosted OpenAI and Azure AI endpoints. |
| Amazon Bedrock | Supported | Converse, streaming Converse, and common Bedrock model response shapes. |
| Mistral | Supported | Mistral API and Mistral web chat streams. |
| Perplexity | Supported | Perplexity API, web app streams, Sonar search activity, and agent-style search activity. |
| Groq | Supported | Groq OpenAI-compatible chat and Responses traffic, plus Groq web activity where visible. |
| xAI / Grok | Supported | xAI API chat and Responses traffic, plus Grok web activity where visible. |
| DeepSeek | Supported | DeepSeek API and web chat activity. |
| Cohere | Supported | Cohere chat API traffic for v1 and v2 chat shapes. |
| OpenRouter | Detected | OpenRouter AI API traffic. Visible detail depends on the routed model and API shape. |
| Hugging Face Router | Detected | OpenAI-compatible chat traffic through the Hugging Face router. |
| Together AI | Detected | OpenAI-compatible chat traffic through Together endpoints. |
| Fireworks AI | Detected | OpenAI-compatible chat and inference endpoints. |
| Cerebras | Detected | OpenAI-compatible chat endpoints. |
| Cloudflare AI Gateway | Detected | AI gateway traffic when routed through Cloudflare’s AI endpoints. |
| Vercel AI Gateway | Detected | OpenAI-compatible traffic through Vercel AI Gateway. |
| MiniMax | Supported | MiniMax and Anthropic-compatible MiniMax message traffic. |
| Moonshot / Kimi | Detected | Kimi and Moonshot API traffic, including coding and chat endpoints. |
| Qwen / DashScope | Detected | Qwen and DashScope chat activity. |
| Zhipu / BigModel / GLM | Supported | GLM and BigModel chat activity, including streamed output where available. |
| DeepL | Detected | DeepL AI translation endpoints. |
| Many more | Detected | Patronus also detects additional AI providers and compatible provider gateways when their traffic uses recognizable AI request, response, or streaming patterns. |
Developer Apps, CLIs, And IDEs
Section titled “Developer Apps, CLIs, And IDEs”| Surface | Status | What customers can expect to see |
|---|---|---|
| Cursor | Supported | Cursor agent and chat activity, including streamed assistant output where visible. |
| Windsurf / Codeium | Supported | Local language-server chat and Windsurf cloud chat activity. |
| JetBrains AI / Junie | Supported | JetBrains AI chat and response streams. |
| Claude Code | Supported | Claude Code CLI activity when routed through Patronus. |
| OpenAI Codex CLI | Supported | Codex CLI activity when routed through Patronus. |
| Gemini CLI | Supported | Gemini CLI requests, responses, and tool-related activity. |
| GitHub Copilot CLI | Supported | Copilot CLI prompts, responses, and tool-oriented events. |
| Cline | Compatible API | Activity is visible when Cline uses a supported provider or compatible API path. |
| Amp | Compatible API | Activity is visible when Amp uses a supported provider or compatible API path. |
| Goose | Compatible API | Activity is visible when Goose uses a supported provider or local model endpoint. |
| Antigravity | Supported | Gemini-compatible Antigravity traffic when routed through Patronus. |
| Qoder / iFlow / OpenCode | Compatible API | OpenAI-compatible or provider-compatible traffic from these tools. |
| Sentry AI surfaces | Detected | AI assistant traffic in supported Sentry chat-style flows. |
| Notion AI | Supported | Notion AI request and streamed response activity where traffic is visible. |
| Microsoft 365 Copilot in Office apps | Supported | Copilot chat and AugLoop-style AI activity where traffic is visible to Patronus. |
| Many more | Detected | Patronus also detects additional AI apps, IDE extensions, and agent tools when their traffic uses recognizable AI request, response, or streaming patterns. |
SDKs And Agent Frameworks
Section titled “SDKs And Agent Frameworks”| Surface | Status | What customers can expect to see |
|---|---|---|
| OpenAI SDKs | Supported | Chat, Responses, streaming, tool calls, and tool results. |
| Anthropic SDKs | Supported | Messages, streaming, tool use, and tool results. |
| Google Gen AI SDKs | Supported | Gemini request and response activity. |
| LangChain | Compatible API | Activity is visible when LangChain uses a supported provider integration. |
| AutoGen | Compatible API | Multi-agent activity is visible when provider calls use supported API shapes. |
| CrewAI | Compatible API | Agent activity is visible when provider calls use supported API shapes. |
| OpenAI Agents SDK | Supported | Agent requests, responses, tool calls, and policy-visible activity. |
| Google ADK | Supported | Agent requests, responses, and tool-related activity through supported Gemini paths. |
| LiteLLM | Compatible API | Traffic is visible when LiteLLM uses supported provider-compatible request and response shapes. |
| Vercel AI SDK | Compatible API | Streamed responses using Vercel AI SDK formats. |
| MCP over HTTP or streamable HTTP | Supported | MCP messages, tool lists, and tool calls when routed through Patronus. |
| Many more | Compatible API | Patronus can observe additional SDKs and frameworks when they call supported providers or use compatible API shapes. |
Local Model Endpoints
Section titled “Local Model Endpoints”| Surface | Status | What customers can expect to see |
|---|---|---|
| Ollama | Supported | Local chat, generate, streaming responses, and tool calls. |
| LM Studio | Supported | Local OpenAI-compatible chat and model traffic when routed through Patronus. |
| OpenAI-compatible local servers | Compatible API | Requests and responses that use OpenAI-compatible chat or Responses formats. |
| Many more | Compatible API | Patronus can observe additional local model runtimes when they expose compatible chat, generate, or Responses-style endpoints. |
Compatibility Notes
Section titled “Compatibility Notes”- Patronus can only observe traffic that passes through an active Patronus protection path.
- Some apps encrypt, tunnel, batch, or proxy their own traffic. In those cases Patronus may show the provider or app but less message detail.
- Local CLIs and containers may require certificate trust configuration before HTTPS traffic is visible.
- Provider support can vary by app version, account configuration, transport, and model feature.