Show HN: Padwan AI, an LLM Client for Python with HTTP/2 and HTTP/3

A developer released Padwan AI, an open-source async Python client for large language model APIs that depends only on the niquests HTTP library and supports OpenAI, Gemini, Mistral, Grok, Anthropic, Voyage AI, ElevenLabs (experimental), Gradium (experimental) and any OpenAI-compatible endpoint. Padwan AI negotiates HTTP/2 and HTTP/3 automatically, types all requests and responses as plain TypedDicts with no validation framework, and includes an AgentSession agentic loop with tool dispatch, parallel execution, approval hooks and snapshot persistence, plus Model Context Protocol tool servers over streamable HTTP and stdio transports. The client also streams thinking tokens from Gemini, Claude, Mistral, Grok and vLLM through an on_thought callback and runs realtime speech-to-speech WebSocket sessions for OpenAI Realtime, Gemini Live and Grok Voice; it installs via pip install padwan-ai or uv add padwan-ai.

Padwan AI Lightweight async client for OpenAI, Gemini, Mistral, Grok, Anthropic, ElevenLabs experimental , Gradium experimental , TypeSafe JEV, experimental , and any OpenAI-compatible API. Why Most LLM client libraries pull in a stack of dependencies and lock you into a single provider's SDK. Padwan AI takes a different approach: - One dependency — niquests https://github.com/jawah/niquests , nothing else. No provider SDKs: Padwan talks to each API directly. - TypedDict-only — all request/response types are plain TypedDict s, no validation framework required. No runtime cost, full editor support. - Multi-provider, extensible — supports the major providers OpenAI, Gemini, Mistral, Grok, Anthropic with a shared base class that makes adding new ones straightforward. Features - Unified interface - Single API for multiple LLM providers - Async-first - Built on async/await for high performance - HTTP/2 and HTTP/3 - Automatic protocol negotiation via niquests https://github.com/jawah/niquests - Fully typed - Complete type hints with Python 3.13+ generics - Streaming support - Real-time token streaming for all providers - Conversation management - Built-in conversation history handling - Agentic loop - AgentSession drives multi-turn conversations with tool dispatch, parallel execution, approval hooks, and snapshot persistence - MCP support - Streamable HTTP and stdio transports for Model Context Protocol https://modelcontextprotocol.io/ tool servers - Thinking models - Stream Gemini, Claude, Mistral, Grok and vLLM thought tokens separately via an on thought callback - Realtime voice - Speech-to-speech WebSocket sessions OpenAI Realtime, Gemini Live, Grok Voice via RealtimeClient - Text-to-speech - Single and two-speaker Gemini TTS via GeminiClient.generate speech details clients/gemini/ text-to-speech , ElevenLabs and Gradium TTS REST, streaming and WebSocket , transcription and speech-to-speech, plus ElevenLabs sound effects and music, via ElevenLabsClient clients/elevenlabs/ and GradiumClient clients/gradium/ Supported Providers | Provider | Chat | Streaming | Batch | Realtime | Transcription | TTS | Embeddings | |---|---|---|---|---|---|---|---| | OpenAI | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | | Gemini | ✅ | ✅ | ✅ | ✅ | ➖ | ✅ | ✅ | | Mistral | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ | | Grok | ✅ | ✅ | ✅ | ✅ | ➖ | ❌ | ➖ | | Anthropic | ✅ | ✅ | ❌ | ➖ | ➖ | ➖ | ➖ | | Voyage AI clients/voyage/ | ➖ | ➖ | ➖ | ➖ | ➖ | ➖ | ✅ | | ElevenLabs experimental clients/elevenlabs/ | ➖ | ➖ | ➖ | ✅ | ✅ | ✅ | ➖ | | Gradium experimental clients/gradium/ | ➖ | ➖ | ➖ | ✅ | ✅ | ✅ | ➖ | | OpenAI-Compatible clients/openai-compatible/ | ✅ | ✅ | ➕ | ➕ | ➕ | ❌ | ➕ | - ❌ = the provider offers it, Padwan does not implement it yet. - ➖ = the provider has no such API Gemini transcribes through chat audio input; Anthropic recommends Voyage AI for embeddings . - ➕ = depends on the endpoint: the request is sent in OpenAI's wire format; support is up to the server. - Realtime: Gradium's is speech-to-speech translation, not a voice-agent session; ElevenLabs and Gradium also stream TTS and STT over WebSocket. Quick Example export OPENAI API KEY="sk-..." uvx padwan-ai "Hello " -m gpt-5.4-mini python from padwan ai import LLMClient async with LLMClient model="gpt-5.5" as client: response, usage = await client.complete chat {"role": "user", "content": "Hello "} print response "content" Installation pip install padwan-ai Or with uv: uv add padwan-ai Agentic loop AgentSession wraps a conversation with the loop that calls the LLM, dispatches any tool calls, feeds the results back, and repeats until the model produces a plain text answer. It accepts individual McpTool instances and whole McpTransport servers in the same list, transports are entered as part of the session lifecycle: python from padwan ai import AgentSession, LLMClient, McpStdio async with AgentSession client=LLMClient model="gpt-5.5" , mcp tools= McpStdio command="uvx", args= "weather-mcp" , system="You have access to weather tools.", as session: text = await session.send "What is the weather in Paris?" See the Agents page agents/ for approval hooks, parallel execution, and snapshot persistence. MCP Model Context Protocol Connect to MCP tool servers over streamable HTTP or stdio: python from padwan ai import McpStreamable, McpStdio async with McpStreamable url="https://mcp.example.com/mcp", token="sk-..." as mcp: result = await mcp.tools 0 .handler {"query": "test"} See the MCP page mcp/ for architecture diagrams, the dual-channel design of McpStreamable , the stdio reader model, and the reconnection flow. CLI / TUI The interactive CLI/TUI is available as a separate package: padwan-cli . One-shot prompt uvx padwan-cli "Explain Python decorators" -m gemini-3.8-flash Interactive chat uvx padwan-cli chat -m gpt-5.5