Python

A local sandbox for running LLM-generated code from Python or Node with no cloud, no daemon, and strong isolation controls.

Open-source execution runtime for production agents. It records tool and LLM calls as durable steps, supports resume after interruption, and lets teams allow, deny, or approve steps with human oversight.

Bastiontrace is a Python library for analyzing AI agent tool-call traces to locate prompt injections, identify where they landed, and map the resulting blast radius.

Reference implementation for building and understanding AI evaluation systems, including rule-based evaluation, human review, LLM-as-a-Judge, online evaluation, and experiment comparison for a flight search app.

A two-part self-study course for building a tiny LLM stack from scalar autograd to a working chat assistant, with modules covering transformers, pretraining, instruction tuning, retrieval, tool use, and agent loops.

A Python framework for building AI agents as Python objects, with typed methods, stateful fields, and LLM-driven agentic methods. The repo includes installation steps, a quick start, examples, and notes on supported models and safety boundaries.

fenic is an open-source Python framework for semantic DataFrames, letting you use LLM-powered operators like extract, classify, summarize, and embed inside query pipelines.

A local-first Python tool that audits Git history to quantify rework in AI-generated engineering work, including PR correction rates, churn clusters, superseded work, and line survival.

Subarr is an open-source coordination layer for the *arr subtitle stack. It adds subtitle coverage tracking, audio-language verification, provider success tracking, queue management, Whisper tuning, and post-transcription quality checks for Bazarr/Subgen workflows.

A Python CLI that packages data science repositories and data-heavy project files into LLM-friendly prompts, with sampling for CSV, SQL, Excel, and notebooks.

A Python framework for building OAuth 2.0 authorization servers for MCP servers, with token storage, PKCE, device flow, introspection, and rate limiting.

Python package for recording real LLM API responses and replaying them in tests so LLM-driven code can run deterministically without live API calls.

Semble is a local-first code search tool for agents. It offers natural-language search, an MCP server, bash/AGENTS.md integration, and a Python API, with claims of faster indexing and much lower token usage than grep+read.

Self-hosted voice agent runtime that streams speech-to-text, LLM output, and text-to-speech together in one process. The page shows quickstart commands, a minimal agent example, browser/client embedding, authentication notes, and performance benchmarks.

Anypick is an open library for selecting the best LLM across providers using filters for capabilities, pricing, and benchmarks. It includes Python and TypeScript implementations, provider support for OpenRouter and Vercel AI Gateway, and examples showing how to pick a model by constraints and strategy.

Codeknow turns a codebase into a queryable knowledge graph for architecture analysis, impact detection, drift tracking, onboarding tours, and refactoring guidance. It runs without LLM keys, supports 25+ languages, and exposes a broad CLI plus export formats and interactive views.

Self-hosted multiplayer AI workspace in Python for team-based agent collaboration, with governed actions, real-time shared rooms, and a tamper-evident audit log.

Open-source agent skill and Python service for stripping AI provenance marks from text and files. It supports multiple hosts, includes install scripts, hooks, and a service-backed workflow for cleaning watermarks across many file types.

An open-source reinforcement learning training loop for LLMs, spanning small CPU-friendly experiments like CartPole to async RLVR runs on GPU clusters. The repo highlights a compact codebase, quickstart commands, task customization, and support for algorithms like REINFORCE, PPO, GRPO, and RLOO.

A research repo for reducing hallucinations in LLM-generated code using semantic triangulation, with setup, benchmarking, experimentation, and reproducibility instructions.

Hubmesh is a Python library for multi-hop GraphRAG retrieval planning on top of existing vector databases. It adds centrality-aware seed selection, budget-aware context packing, entity graph building, and an MCP server for agent/tool integrations.

Open-source Python library for AI agent context governance with tamper-evident audit trails, provenance validation, freshness checks, and adapter support for common agent stacks.

A Python package and CLI for checking how closely an AI-generated answer reuses its source text, with support for verbatim matching, subsequence matching, quotation verification, and pipeline gating.

A local terminal UI for exploring AI coding spend across tools like OpenCode, Claude Code, Codex, GitHub Copilot, and more. It shows costs by month, day, project, session, and model, with drill-down views, heatmaps, exports, and a web browser mode.