Open Source Frameworks
A software framework provides reusable structure, components, and conventions for building applications or extending existing tools. It can handle common tasks such as routing, data access, asynchronous processing, and user interface rendering, so developers can focus on an application’s specific needs. Frameworks also establish patterns for organizing code and connecting services, helping teams create software that is easier to maintain and extend.
Open source frameworks span web development, programming languages, command-line environments, and artificial intelligence. Some offer a broad foundation, while others focus on a specific task or runtime. When choosing one, consider its maturity, license, maintenance activity, documentation, system requirements, and compatibility with your languages and services. Frameworks are useful to individual developers and teams who want established building blocks, consistent architecture, and the freedom to inspect or adapt the source code.
99 repositories · updated October 17, 2026
tramAI: Govern AI Workflows on the JVM
TramAI is a Kotlin-first JVM runtime for typed AI workflows with configurable policy, provider routing, tool controls, approvals, and execution evidence. It suits teams that need governance around model use, not just a way to call an LLM.
NewHealth 45/100
shepherd: Supervise Agents with Reversible Execution Traces
Shepherd records agent work as inspectable, reversible execution traces and keeps changes as proposals for review. It is aimed at developers building systems that supervise, replay, or manage the work of other agents.
NewHealth 42/100
verl-omni: Train Multimodal Generative Models with RL
VeRL-Omni is a Python framework for reinforcement-learning post-training of diffusion, unified multimodal, and omni-modality models. It builds on verl and targets teams that need distributed training, multimodal rollouts, and configurable reward workflows.
NewHealth 68/100
uni-agent: Train Long-Horizon Agents at Scale
Uni-Agent is a Python framework for running and training long-horizon agents through a shared interaction stack. It helps agent developers connect harnesses, tasks, tools, and sandboxes for scalable inference, evaluation, and reinforcement learning.
ActiveHealth 61/100
bernstein: Govern and Orchestrate AI Agents
Bernstein coordinates AI-agent workloads through declarative policies, approval gates, and isolated task execution. It records runs for replay and offline verification, helping teams that need accountable, auditable automation.
ActiveHealth 80/100
CyberVerse: Build Self-Hosted Voice-First AI Agents
CyberVerse is a self-hosted framework for real-time voice agents with persistent character memory, tools, and retrieval-augmented generation. It can also stream digital-human video, for teams building interactive assistants with configurable model providers.
NewHealth 47/100
lenso: Build Bun-First TypeScript Plugin Applications
Lenso is a TypeScript plugin framework for Bun apps, built around async services and explicit resource ownership. Its optional packages add web, auth, database, task, and agent-operation integrations for teams assembling applications from separate capabilities.
NewHealth 40/100
declarative-agents: Define Tool-Augmented Agents in YAML
A Go runtime interprets YAML profiles that define agent tools, states, transitions, and budgets. It suits teams that want to change agent workflows through configuration, though the project is new, contributor-concentrated, and in bug-fix-only maintenance mode.
NewHealth 63/100
open-slide: Build and Present Agent-Written Slides
open-slide is a TypeScript and React framework for creating, editing, and presenting slide decks with coding agents. It provides a fixed-size canvas and tools for previewing, navigating, and exporting decks without requiring a custom slide-authoring DSL.
NewHealth 61/100
envharness: Adapt Environments for Agent Learning
EnvHarness wraps frozen interactive environments with composable layers that can target an AI agent’s weaknesses without changing benchmark code or grading. It is for researchers building and evaluating agent-training workflows across different environments.
NewHealth 27/100
json-render: Build Guardrailed Interfaces from AI-Generated JSON
json-render turns AI-generated JSON into interfaces using components and actions you define, so model output stays within your application's boundaries. It supports multiple UI frameworks and targets beyond web interfaces, including mobile, email, PDFs, video, and 3D.
ActiveHealth 51/100
fastmcp: Build MCP Servers, Clients, and Apps in Python
FastMCP is a Python framework for building Model Context Protocol servers, clients, and interactive apps. It turns Python functions into MCP tools and manages protocol details such as schemas, transport, and authentication.
ActiveHealth 73/100