# Open Source

Framework for LLM applications

The most widely-used framework for building LLM-powered applications, available in Python and JavaScript. Provides abstractions for chains, agents, RAG, memory, tool usage, and structured output. Integrates with 100+ LLM providers, vector stores, document loaders, and tools. LangSmith offers tracing and evaluation. LangGraph enables stateful, multi-agent workflows with cycles. 100K+ GitHub stars. The de facto standard for LLM application development despite growing alternatives like LlamaIndex.

Unified API proxy for 100+ LLMs

Drop-in OpenAI-compatible proxy supporting 100+ LLM providers with load balancing, spend tracking, rate limiting, and fallback routing. Acts as a unified gateway for all your AI model calls, letting teams switch between providers, enforce budgets, and add reliability layers without changing application code. Essential infrastructure for multi-model AI architectures.

Open-source personal AI agent for messaging apps

OpenClaw is a free, open-source AI agent framework that turns any LLM into an autonomous personal assistant accessible through messaging apps like WhatsApp, Telegram, Discord, and Signal. Running entirely on your local machine via a Node.js gateway, it connects AI models to system tools, browsers, files, and APIs for multi-step task execution with persistent memory across sessions.

Open-source search engine with vector and hybrid retrieval

OpenSearch is an Apache-2.0 distributed search engine with native vector-search support for teams that want BM25, filters, aggregations, and k-NN retrieval in the same search stack. It is distinct from Elasticsearch in the aicoolies catalog: OpenSearch is the AWS-backed open fork with its own docs, plugin path, and serverless deployment options.

Alibaba's agent framework built for the Qwen model family

Qwen-Agent is Alibaba's Apache-2.0 framework for building AI agents around the Qwen model family. It supports tool use, planning, memory, RAG, Code Interpreter, Browser Assistant, MCP extras, custom tools, and Qwen Chat backend patterns with Qwen3/Qwen3.5 examples. Best fit for teams standardizing on Qwen rather than a generic multi-agent router, with 16.5K+ GitHub stars.

Microsoft’s pytest-native red teaming framework for turning AI agent safety findings into CI tests.

RAMPART is an open-source Microsoft framework for safety and security testing of agentic AI applications. It brings red-team findings into a pytest-native workflow so teams can turn prompt injection, unsafe tool use, and behavioral boundary failures into repeatable regression tests. The strongest aicoolies angle is developer workflow: RAMPART makes agent safety part of CI/CD instead of a one-off security review.

CNCF Sandbox Kubernetes alert enrichment and automation platform

Robusta is a CNCF Sandbox project that enriches Kubernetes alerts with diagnostic context and automates remediation workflows. It intercepts Prometheus alerts, attaches relevant logs, pod status, resource metrics, and troubleshooting suggestions before delivering them to Slack, Teams, or PagerDuty. Supports custom playbooks for automated incident response and AI-powered root cause analysis.

Autonomous Kubernetes management and predictive scaling

Sedai provides an autonomous control layer for Kubernetes that right-sizes workloads, remediates anomalies, and performs predictive autoscaling ahead of traffic demand. Sedai says it manages large enterprise cloud environments for customers including Palo Alto Networks and builds behavioral models to scale pods before demand arrives rather than reacting after performance degrades.

Hybrid search and ML ranking engine at scale

Vespa is an open-source serving engine with 6K+ GitHub stars for hybrid search combining vector similarity, BM25 text ranking, and structured filtering in a single query. Built by Yahoo for web-scale, it handles billions of documents with millisecond latency. Features real-time indexing, ML model serving, tensor computation, and ACID-compliant writes. Supports custom ranking models, query federation, and geographic search. Used for recommendation systems, personalization, and RAG.

LLM observability and evaluation by Weights & Biases

W&B Weave is the LLM observability and evaluation toolkit from Weights & Biases. It provides automatic tracing of LLM calls with full input/output logging, cost and latency tracking, evaluation pipelines with custom scorers, and a trace explorer for debugging multi-step agent workflows. Integrates with OpenAI, Anthropic, LangChain, and CrewAI via simple Python/TypeScript decorators.

MCP server for controlling Windows desktops through UIAutomation

Windows-MCP is an open-source MCP server for giving AI agents structured access to Windows desktop automation. It focuses on UIAutomation, snapshots, input control, and Windows-specific app workflows, making it different from general filesystem or shell MCP servers.

Open-source LLMOps platform for prompt management and evaluation

Agenta is an open-source LLMOps platform that combines prompt engineering playgrounds, prompt version management, LLM evaluation, and observability in a unified interface. It supports 50+ LLM models with side-by-side prompt comparison, A/B testing, human evaluation workflows, and OpenTelemetry-native tracing. Self-hostable with 4,000+ GitHub stars.

Multi-agent AI framework

Python framework for orchestrating autonomous AI agents that collaborate to accomplish complex tasks. Define agents with specific roles, goals, and backstories, then organize them into crews with sequential or parallel task execution. Supports tool usage (web search, file I/O, API calls), memory, delegation between agents, and human-in-the-loop input. Works with OpenAI, Anthropic, local models, and more. 25K+ GitHub stars. Leading multi-agent framework alongside LangGraph and AutoGen.

AI-driven log analysis with zero false positives

Dash0 is an AI-driven observability platform focused on log analysis that auto-structures unstructured logs, provides instant alerting with zero false positives, and delivers full-stack tracing capabilities. It uses AI to transform raw log data into structured, searchable events without requiring manual parsing configuration, making log-based debugging significantly faster for engineering teams.

Library for efficient similarity search and clustering of dense vectors at billion-scale.

FAISS is Meta AI Research's open-source library for efficient similarity search and clustering of dense vectors. It implements approximate nearest-neighbor algorithms designed to scale to billions of vectors, with optimized indexes that fit in RAM and GPU acceleration for the largest workloads. Engineering teams use FAISS as the retrieval primitive underneath custom RAG pipelines, recommendation systems, and large-scale embedding search infrastructure.

Open-source Kubernetes security platform for risk analysis and compliance

Kubescape is a CNCF-backed open-source Kubernetes security platform that scans clusters, manifests, and container images for vulnerabilities, misconfigurations, and compliance violations. It checks against NSA-CISA, MITRE ATT&CK, and CIS benchmarks, integrates into CI/CD pipelines, and provides runtime threat detection via eBPF. Supports SBOM generation and vulnerability scanning. Used by ARMO with growing enterprise adoption in cloud-native security.

Open-source observability for AI agents

Laminar is an open-source observability platform for AI agents providing tracing, evaluation, and analytics for LLM applications. It integrates with Vercel AI SDK, LangChain, OpenAI, and Anthropic with a single line of code. Features include OpenTelemetry-native SDKs, an extensible evaluation framework with CI/CD support, SQL access to traces and metrics, and a visual debugging timeline for agent reasoning and actions.

The LLM anti-framework for typed AI apps

Mirascope is an open-source Python and TypeScript toolkit for building LLM applications that prioritizes type safety, composability, and 100% test coverage. Positioned as the 'anti-framework,' it provides fine-grained control over LLM interactions using familiar language constructs rather than rigid abstractions, supporting all major providers through a unified interface.

LLM evaluation and tracing by Comet

Opik is an open-source LLM evaluation and tracing platform by Comet ML for debugging, testing, and monitoring AI applications. Provides detailed traces of LLM calls with latency, token usage, and cost tracking. Features automated evaluation with built-in and custom metrics, dataset management for regression testing, and production monitoring dashboards. Integrates with the broader Comet ML experiment tracking ecosystem. Available as both self-hosted open-source and managed cloud service.

Read/write MCP bridge between AI coding agents and Figma

Talk to Figma MCP is an MIT-licensed bridge from Grab that connects Cursor, Claude Code, and other MCP-capable agents to Figma through a local MCP server, WebSocket bridge, and Figma plugin. Unlike read-only context servers, it can inspect selections, create or modify nodes, update text in bulk, and automate design operations, so teams should review permissions before enabling write access.

ByteDance's open-source LLM coding agent with multi-provider support

Trae Agent is ByteDance's open-source software engineering agent that autonomously resolves GitHub issues, fixes bugs, and implements features using any LLM provider. It supports OpenAI, Anthropic, Doubao, Azure, Ollama, and Gemini backends, making it one of the most provider-flexible coding agents available. With over 11,000 GitHub stars and a modular research-friendly architecture, it offers a strong alternative to Western-centric coding agents.

Microsoft's conversational multi-agent framework

AutoGen is an open-source programming framework from Microsoft Research for building AI agents and facilitating cooperation among multiple agents to solve complex tasks through multi-turn conversations. Pioneered conversable agents that interact, use tools, and involve humans in the loop for multi-agent workflows. v0.4 features a redesigned async event-driven architecture with stronger observability, flexible collaboration patterns, and reusable components.

Test anything that runs in a browser

JavaScript end-to-end testing framework that runs tests directly in the browser with real-time reloading and an interactive test runner UI. Features automatic waiting, time-travel debugging with DOM snapshots at each step, network request stubbing/interception, screenshot and video recording, and a dashboard service for CI analytics. Supports component testing for React, Vue, Angular, and Svelte. Known for developer experience and reliable test execution. 49K+ GitHub stars.

Self-evolution engine for AI agents with auditable updates

Evolver is an open-source self-evolution engine for AI agents that turns run logs into auditable, reviewable updates via its Genome Evolution Protocol. Instead of ad hoc prompt tweaking, teams collect traces and Evolver proposes versioned diffs to prompts, tools and workflows that engineers can approve, reject or roll back like code.

DeepSeek's optimized attention kernel for Multi-Head Latent Attention

FlashMLA is DeepSeek's MIT-licensed CUDA kernel library for optimized attention in DeepSeek-V3 and DeepSeek-V3.2-Exp style inference. It includes dense MLA decoding plus sparse attention kernels for DeepSeek Sparse Attention, with README-reported H800/CUDA metrics up to 3000 GB/s, 660 TFLOPS, and sparse 640/410 TFlops paths. It has 12.7K+ GitHub stars.

AI quality testing for bias, drift, and vulnerabilities

Giskard is an open-source testing framework for evaluating AI model quality, detecting bias, data drift, and security vulnerabilities. It provides automated test generation for LLMs and tabular models, scanning for issues like hallucination, prompt injection susceptibility, stereotypical outputs, and data leakage. Integrates with CI/CD pipelines for continuous model validation before deployment.

GitHub Copilot in your terminal

Brings GitHub Copilot to the terminal, giving developers AI assistance for shell commands, error explanations, and code generation directly from the command line. Supports natural language queries to generate complex shell commands, explains error messages in plain English, and integrates with the broader GitHub Copilot ecosystem including model selection and premium request management.

Unified API, performance, and contract testing DSL

Karate is an open-source testing framework that unifies API testing, performance testing, UI automation, and contract testing in a single BDD-style DSL. Write tests in plain Gherkin-like syntax without any Java knowledge. Built-in assertions, data-driven testing, parallel execution, and HTML reports. 8,900+ GitHub stars, MIT licensed. Mature, actively maintained project with commercial support options for comprehensive API quality assurance.

Visual IDE for React that designers can use

Onlook is an open-source visual editor for React and Next.js with 25.9K+ GitHub stars that lets designers edit real code-backed interfaces on an infinite canvas. It connects visual changes, AI chat, component and layer inspection, and versioned code workflows so teams can prototype and refine product UI without treating design files as the source of truth.

OpenTelemetry-native LLM observability instrumentation

OpenLLMetry by Traceloop is an open-source instrumentation library with 7,000+ GitHub stars that adds OpenTelemetry-native tracing to LLM and AI agent applications. It captures detailed traces of model calls including latency, token usage, costs, and error rates, exporting data to any OpenTelemetry-compatible backend like Grafana, Datadog, or Jaeger for vendor-neutral AI observability.

Open-source AI-powered PR review with slash commands

PR-Agent by Qodo is the most widely adopted open-source AI code review tool with 11,500+ GitHub stars and 207 contributors. It provides automated PR description generation, intelligent line-by-line code review, and actionable improvement suggestions through intuitive slash commands like /describe, /review, and /improve, integrating natively with GitHub Actions, GitLab CI, Bitbucket, and Azure DevOps.

Open-source AI coding agent with self-hosted deployment option

Refact.ai is an open-source/on-premise oriented AI coding agent for VS Code and JetBrains that supports BYOK model routing, codebase understanding, developer-tool integrations and self-hosted deployment. Its public site now warns that Refact Cloud is shutting down soon, so teams should treat hosted availability as a migration risk and validate the current enterprise support path.

Browser automation with AI vision — no XPath or DOM parsing needed

Skyvern automates browser-based workflows using LLMs and computer vision instead of brittle XPath or CSS selectors. It understands web pages visually, navigating forms, clicking buttons, and extracting data like a human would. Achieved 85.85% success rate on WebVoyager benchmark and SOTA on WRITE tasks for RPA. 21,000+ GitHub stars, AGPL-3.0 licensed. Skyvern Cloud offers managed usage-based hosting for teams that prefer not to self-host the infrastructure.

AI-powered data quality for ML datasets

Cleanlab is a data-centric AI library that automatically detects and fixes label errors, outliers, and data quality issues in machine learning datasets. It works with any ML model and any data type including text, images, tabular, and audio by analyzing model predictions to identify mislabeled examples, near-duplicates, and ambiguous data points. Cleanlab helps teams improve model accuracy by cleaning training data rather than tuning model architecture.

Open-source dev environment management with AI integration

Daytona is secure, elastic infrastructure for running AI-generated code in isolated sandboxes. It gives agents and developer workflows programmable environments with dedicated kernel, filesystem, network, vCPU, memory, and disk, backed by OCI/Docker compatibility, SDK/API access, and under-90ms sandbox startup. The project has 72,000+ GitHub stars and is AGPL-3.0 licensed.

Run LLMs as a single portable executable file

Llamafile by Mozilla packages a complete LLM — model weights, inference engine, and OpenAI-compatible API server — into a single executable file that runs on Mac, Windows, Linux, FreeBSD, and OpenBSD with no installation. Built on llama.cpp and Cosmopolitan Libc for cross-platform portability, it delivers GPU-accelerated inference when available and falls back to optimized CPU execution. Supports GGUF models with a built-in web chat UI and REST API for integration.

Evaluation framework for RAG pipelines

RAGAS is an Apache-2.0 open-source evaluation framework with 14K+ GitHub stars that provides standardized metrics for assessing RAG pipeline quality. It measures faithfulness, answer relevancy, context precision, and context recall to identify whether retrieval, generation, or both are failing. It is framework-agnostic, supports LLM-as-judge evaluation, and its README discloses minimal anonymized Open Analytics with a RAGAS_DO_NOT_TRACK opt-out.

MIT-licensed autonomous coding-agent reference, now superseded for many new uses by mini-swe-agent.

SWE-agent is an MIT-licensed autonomous coding-agent reference from Princeton and Stanford researchers that takes GitHub issues and attempts fixes with a bring-your-own language model. Its agent-computer interface remains foundational for repository navigation, editing, and test execution. The README now says development has shifted to mini-swe-agent, which supersedes SWE-agent and is generally recommended going forward.

Open-source, local-first Notion alternative with built-in AI

AppFlowy is an open-source workspace platform offering documents, wikis, databases, and project management as a privacy-first alternative to Notion. Built with Rust and Flutter for native performance, it runs fully offline with local data storage by default. AI features include writing assistance, summarization, and translation powered by configurable LLM providers. 72.5K+ GitHub stars, backed by a growing contributor community with self-hosted and cloud deployment options.

Git-friendly offline API client

Open-source API client for testing REST, GraphQL, and gRPC APIs as a lightweight, privacy-focused alternative to Postman. Stores API collections as plain files in your Git repository for easy version control and team collaboration. Features environment variables, scripting, collection runner, and a declarative syntax called Bru. Offline-first with no cloud account required. Available on macOS, Windows, and Linux. 44K+ GitHub stars and growing rapidly among developers leaving Postman.

AI-native data application framework with SQL generation and agents

DB-GPT is an open-source AI-native data app framework combining SQL generation, database chat, RAG, and multi-agent orchestration for data-centric workflows. It supports natural language to SQL conversion, automated data analysis, and custom data app development. Integrates with MySQL, PostgreSQL, SQLite, and more. 19,000+ GitHub stars, MIT licensed. Positioned as an alternative to MindsDB for teams building AI-powered data applications and internal database tools.

Drag-and-drop LLM flow builder

Open-source protocol for connecting AI models to external tools and data sources, created by Anthropic. Provides a standardized way for LLMs to interact with APIs, databases, and local files through a universal client-server architecture. Rapidly adopted across the AI ecosystem as the standard interface between AI assistants and the tools they need to be useful.

Multi-agent CLI pair programmer with FORGE, MUSE, and SAGE agents

Model-agnostic terminal coding tool with 3 specialized agents: FORGE for code editing, MUSE for planning and review, and SAGE for research. Connects to hundreds of LLM providers and models with local-first privacy and conversational Git integration. Apache 2.0 licensed. A thoughtfully designed multi-agent approach that separates concerns between coding, thinking, and information gathering for more reliable results.

Self-evolving local computer agent with a reusable skill tree

GenericAgent is a minimal, self-evolving autonomous agent from a 3.3K-line seed and ~3K core loop that gives LLMs system-level control of a local computer. It writes files, runs shell commands, browses the web, and uses keyboard/mouse/screen/mobile tools, while skill crystallization saves successful runs into a reusable skill tree that cuts token cost on repeats.

Delightful JavaScript testing

Jest is Meta's open-source JavaScript and TypeScript testing framework with batteries-included support for unit, integration, and snapshot tests. Zero-config for most projects, it bundles its own test runner, assertion library, mocking utilities, code coverage, and parallel execution. The de facto standard for React, Node.js, Next.js, and modern JS apps, and a common target for AI-generated test suites.

AI coding agent by Moonshot AI

Terminal-based AI coding agent from Moonshot AI, powered by Kimi K2.5 with a 256K context window that achieves 76.8% on SWE-Bench Verified. Reads and edits code, executes shell commands, fetches web pages, and autonomously plans multi-step development workflows through natural language. Moonshot's entry into the AI coding agent market, leveraging their strength in large-context language models.

AI agent testing and LLM evaluation platform

LangWatch is an AI agent testing, LLM evaluation, and observability platform for production AI teams. It combines traces, evaluations, scenario simulations, guardrails, prompt management, and Optimization Studio/DSPy workflows, with cloud and self-managed options for teams that need release-quality feedback loops.

Framework for evaluating LLM and agent performance

OpenAI Evals is an open-source framework and benchmark registry for evaluating LLM performance on custom tasks. It provides infrastructure for writing evaluation prompts, running them against models, and recording results in a structured format for comparison. The hosted Evals API on the OpenAI platform adds managed run tracking, dataset management, and programmatic access to evaluation pipelines. With 17,700+ GitHub stars, it serves as a foundation for systematic LLM quality measurement.