Railtracks

Railtracks is a Python framework for building agents and agent harnesses: the loop, tools, context and controls that surround a language model. Its documentation presents it as a toolkit made by agent builders for agent builders, with the aim of giving developers control over how an agent runs rather than hiding it behind a black box. The docs cover designing agents, wrapping Python functions as tools, using agents as tools for other agents, and connecting to MCP servers. A single interface connects to models from several providers, and platform integrations include Azure AI Foundry, Ollama, HuggingFace, Portkey and Apple's on device Foundation Model on recent Macs. Calls are stateless, and an agent response returns its message history so a conversation can be continued by passing that history back in. Beyond the core loop, the framework documents middleware, verifiers, prebuilt agents, retrieval components such as chunking, embeddings and stores, and evaluation tools with metrics and evaluators. An observability section covers tracking and visualizing agent runs, and further pages cover integrations with storage and databases, tutorials, notebooks and a full API reference. Railtracks is aimed at developers who write agent logic in Python and want to inspect and control each step. The documentation site does not publish pricing.