# python

MCP, ACP and Skills support for building production coding agents — interactive or automated.

fast-agent is an Apache-licensed Python framework for building and running LLM agents with full MCP (Model Context Protocol) and ACP support. It ships with an interactive shell mode, Skills management, and multi-model routing — making it a practical platform for coding agents, workflow automation, and agent evaluation across Claude, Codex, HuggingFace, and local models.

Vectorless, reasoning-based RAG that reads documents like a human expert — no vector DB, no chunking.

PageIndex is a vectorless, reasoning-based RAG system that builds hierarchical tree indexes from long documents and uses LLMs to navigate them like a human expert would. Instead of chunking text and comparing embeddings, it constructs a table-of-contents-style structure and reasons its way to the right sections — no vector database required. Available as an open-source Python package, cloud API, MCP server, and chat platform.

Open-source post-building layer for agents — tracing, evals, and online monitoring

Judgeval is the open-source post-building layer for AI agents from Judgment Labs, providing OpenTelemetry-based tracing, hosted and custom evaluation scorers, and online behavior monitoring for LLM-powered applications. Instrument any function with a single decorator, score live production traffic against faithfulness and instruction-adherence checks, and feed real-world failures back into reinforcement learning or supervised fine-tuning loops.

Open-source observability and self-healing layer for AI agents

TraceRoot is a YC S25-backed open-source observability platform purpose-built for AI agents and LLM apps. It combines OpenTelemetry-compatible tracing with an agentic debugging runtime that reads your source code, correlates failures with recent commits, and proposes fix PRs automatically. BYOK support spans seven LLM providers; the entire stack runs self-hosted via Docker Compose, with TraceRoot Cloud available for managed deployments.

Rust-native multi-agent orchestration for production

GraphBit is a Rust-native, multi-agent orchestration framework built for production. It targets the gap between Python-first frameworks like LangGraph and the operational expectations of enterprise systems — predictable memory, low latency, deterministic concurrency, and the ability to embed an agent runtime in services that already run Rust without dragging in a Python interpreter.

Open-source toolkit for building AI SRE incident response agents

OpenSRE is an open-source Python toolkit from Tracer Cloud for building AI SRE agents that investigate and respond to production incidents. It ships with connectors to Prometheus, Grafana, Kubernetes and incident platforms, plus a simulation harness that replays past incidents so teams can benchmark agent accuracy before trusting it on live pager rotations.

Self-evolving local computer agent with a reusable skill tree

GenericAgent is a minimal, self-evolving autonomous agent in roughly 3K lines of Python that gives LLMs system-level control of a local computer. It writes files, runs shell commands, and browses the web, but its defining feature is skill crystallization: successful task runs are saved as reusable skills inside a growing skill tree that cuts token cost on repeats.

AI-powered file-type detection at Google scale

Open-source AI-powered file-type detection tool from Google that uses a custom deep-learning model under a few megabytes to identify more than 200 binary and textual content types in milliseconds, even on a single CPU. Magika ships as a CLI, Python package, JavaScript/TypeScript library, and an ONNX model, achieves around 99% accuracy on its test set, and is already used at Google scale across Gmail, Drive, and Safe Browsing as well as by VirusTotal and abuse.ch.

Human-in-the-loop web agent you can co-pilot in real time

Magentic-UI is a Microsoft Research web agent with a human-in-the-loop interface for browsing, coding, and file tasks. It plans multi-step actions, asks for approval before executing, and lets users co-pilot by taking over the browser mid-task. Built on AutoGen, it runs a team of specialized agents for web browsing, file handling, and code execution with full action transparency and safety guardrails.

Constrained generation that guarantees valid LLM outputs every time

Guidance is Microsoft's structured generation library that enforces output constraints directly within LLM decoding. It supports JSON schemas, regex patterns, grammars, and interleaved generation-and-control flow to guarantee valid outputs from any compatible model. Works with local models via llama.cpp, Transformers, and remote APIs including OpenAI and Anthropic. Eliminates retry loops and post-processing for structured data extraction.

Task runner for Python with Poetry and uv

Poethepoet (poe) is a batteries-included task runner for Python projects that integrates with Poetry and uv package managers. Define tasks in pyproject.toml, compose them in sequential, parallel, or DAG workflows, and execute with full virtual environment context. Supports shell commands, Python scripts, environment variables, .env file loading, and auto-generated shell completion across bash, zsh, and fish for streamlined development workflows.

AI-ready PDF parser with benchmark-leading accuracy

OpenDataLoader PDF is a high-performance parser that extracts structured, AI-ready data from PDFs with industry-leading 0.907 benchmark accuracy. Combines deterministic local processing with optional AI hybrid mode for complex layouts, OCR support across 80+ languages, formula extraction in LaTeX, chart descriptions, and built-in prompt injection filtering. Available as Python, Node.js, and Java SDKs for seamless RAG pipeline and data preparation integration.

3D reconstruction without camera parameters

DUSt3R is Naver's breakthrough 3D reconstruction method that generates dense 3D scenes from unconstrained image pairs without known camera intrinsics or extrinsics. It casts pairwise reconstruction as pointmap regression, removing hard geometric constraints of projective camera models. Supports multi-view alignment, depth estimation, visual localization, and extends to MASt3R and MUSt3R for large-scale applications.

Declarative multimodal AI data infrastructure

Pixeltable is a declarative data infrastructure for multimodal AI that stores video, audio, images, and documents as first-class column types. Define Python computed columns for inference and transformations, and Pixeltable auto-orchestrates execution with incremental updates. Built-in vector search eliminates the need for separate vector databases while supporting RAG and semantic search workflows.

Declarative code-first ELT data integration

Meltano is a declarative, code-first data integration engine with 500+ Singer connectors for building ELT pipelines. It replaces custom API integration code with configuration-driven pipeline definitions that live in version control alongside application code. Integrates with dbt for transformation, supports scheduling and monitoring through a unified CLI, and powers production pipelines at scale.

Cross-platform on-device AI model runtime

Nexa SDK enables running frontier LLMs and multimodal models locally across PC, mobile, IoT, and wearables with automatic hardware acceleration for GPU, NPU, and CPU. It supports Qwen, Gemma, Llama, DeepSeek models with Python/C++ desktop SDKs, Android/iOS mobile SDKs, and Docker for edge deployment. Includes an OpenAI-compatible API server with chat and function calling support.

Data quality validation framework for Python

Great Expectations is an open-source Python framework for validating, documenting, and profiling data quality. Teams define expectations as expressive unit tests for their data using an intuitive API, then validate datasets against those rules in CI/CD pipelines or production workflows. It connects to pandas, Spark, and SQL sources, generates data documentation automatically, and integrates with orchestrators like Airflow and Prefect for continuous data quality monitoring.

Fast memory-efficient GPU attention kernels

FlashAttention is a fast and memory-efficient exact attention implementation that reduces GPU memory usage from quadratic to linear in sequence length. Created by Tri Dao, it achieves 3-4x speedups over baseline implementations through IO-aware tiling that minimizes HBM reads and writes. Versions include FlashAttention-2 with improved parallelism, FlashAttention-3 optimized for Hopper H100 GPUs, and FlashAttention-4 targeting Hopper and Blackwell architectures.

Open-source deep learning text-to-speech toolkit

Coqui TTS is an open-source deep learning toolkit for text-to-speech synthesis, originally built by former Mozilla TTS engineers. It supports multi-speaker and multilingual synthesis, voice cloning from just six seconds of audio, and ships pre-trained models for 20+ languages. After Coqui shut down in 2023, the Idiap Research Institute forked and actively maintains it. With 45K+ GitHub stars, it remains the most popular open-source TTS framework in Python.

Container-native local AI model serving with Podman

RamaLama is an open-source tool that containerizes AI model inference using Podman or Docker, eliminating host system configuration complexity. It auto-detects GPUs (NVIDIA, AMD, Intel, Apple Silicon), pulls models from HuggingFace, Ollama, and OCI registries, and runs them in isolated rootless containers with read-only mounts and network isolation. Developed under the Containers project (Red Hat ecosystem), it brings familiar container workflows to local LLM serving.

Open-source real-time voice, video, and AI agent infrastructure

LiveKit is an open-source WebRTC infrastructure platform for building real-time voice, video, and data applications. It powers ChatGPT Advanced Voice Mode and serves as the foundation for AI agent voice interactions. LiveKit provides server-side SDKs for Go, Python, Node.js, and Rust alongside client SDKs for every major platform. The Agents framework lets developers build voice-enabled AI assistants with speech-to-text, LLM processing, and text-to-speech pipelines in Python or Node.js.

Production-grade web scraping and browser automation library

Crawlee is an open-source web scraping and browser automation library for Node.js and Python that handles the hard parts of building reliable crawlers. It manages proxy rotation, request queuing, automatic retries, session management, and fingerprint spoofing out of the box. Supports Puppeteer, Playwright, Cheerio, and HTTP-based crawling with a unified API. Built by Apify, it includes persistent storage, autoscaling concurrency, and TypeScript-first design for production deployments.

Open-source product analytics, session replay, and feature flags

PostHog is an all-in-one open-source product analytics platform that combines event tracking, session replay, feature flags, A/B testing, error tracking, and user surveys in a single tool. With a generous free tier covering 1 million events per month, it replaces the need for separate tools like Mixpanel, Amplitude, LaunchDarkly, and Hotjar. Self-host on your own infrastructure for complete data ownership, or use the managed cloud with warehousing and SQL query access.

Instant GraphQL and REST APIs on any database

Hasura auto-generates real-time GraphQL and REST APIs directly from your database schema—PostgreSQL, MySQL, SQL Server, MongoDB, and more. It provides fine-grained row-level and column-level access control, event triggers on database changes, remote schema stitching, and real-time subscriptions out of the box. Available as a managed cloud service or self-hosted, Hasura eliminates weeks of boilerplate API development while maintaining full control over authorization logic.