Network ML

A small, easy-to-use C++ library for REST API requests and LLM chat, built on libcurl. The chat client works with a local model via Ollama or cloud providers (Groq, OpenRouter, DeepSeek, OpenAI, Anthropic) behind one API.

Looking for ML::File? The filesystem utilities live in File ML now. The class is unchanged; depend on file-ml and #include "fileml.h".

It aims to make two things one-liners from C++:

// HTTP request with a rich response
ML::Requests req;
ML::Response r = req.get("https://api.example.com/data");
if (r.ok()) std::cout << r.status << "\n" << r.body;

// Talk to an LLM (local Ollama here; swap the Provider for a cloud model)
ML::Chat chat(ML::Provider::Ollama, "llama3.2:1b");
std::cout << chat.ask("Explain RAII in one sentence.");

Documentation

Full documentation, including build instructions and usage examples, lives in the project wiki:

• Getting Started: requirements, building, prebuilt binaries
• Requests: the HTTP client
• Chat: the LLM client

Features

• ML::Requests: full HTTP verbs (get/post/put/patch/del/head) returning a rich Response, with inline request bodies, multiple request headers, and per-request timeouts. Also file download (streamed to disk with optional progress), multipart upload, and opt-in auto-retries with backoff. (Legacy string-returning getReq/postReq/etc. are kept for compatibility.)
• ML::Auth / ML::Url: helpers for auth headers (bearer/basic/apiKey) and percent-encoded query URLs (Url::build).
• ML::Response: status, body, response headers (a map), and ok() (true for any 2xx).
• ML::Chat: conversational client for Ollama (local), Groq, OpenRouter, DeepSeek, OpenAI, or Anthropic:
  • one API across every provider
  • persistent system prompt and conversation memory
  • adjustable temperature and max tokens
  • token streaming via a callback
  • reasoning-token support for reasoning models (the model's "thinking" streams to a separate callback, available via lastReasoning())
  • save/load conversation history to a JSON file (saveHistory/loadHistory)
• All HTTP runs in-process via libcurl (HTTPS through Schannel on Windows), with no curl subprocess and no temp files.

License

MIT. See LICENSE.