Recently updated open-source projects

MetaGPT organizes product managers, architects, project managers, and engineers into a coordinated multi-agent workflow. From a one-line requirement, it can produce user stories, requirements, data structures, APIs, documentation, and a code repository. It runs from the command line or as a Python library, and requires Python 3.9 through 3.11 plus Node.js and pnpm for actual use.

Mistral Vibe is Mistral's open-source command-line coding assistant for exploring, searching, and editing projects through natural language while running shell commands when needed. It supports delegated subagent tasks, reusable skills, and selectable agent profiles; voice input is experimental, and Unix environments are the official target even though Windows can run it.

Composio creates a separate session for each user and gives an agent runtime access to app authentication, tool discovery, and execution without loading hundreds of definitions into context. It offers TypeScript and Python SDKs, a CLI, framework adapters, and hosted MCP endpoints for more than 1,000 app toolkits. Setup requires a Composio API key and the relevant connected user accounts.

AutoGPT’s current platform uses a low-code block builder to create agents and workflows, manage deployments, and run automations continuously from external triggers. The repository also retains classic tooling such as Forge and an agent benchmark. Self-hosting is a technical Docker-based setup, the hosted service remains a closed beta, and licensing differs between the Polyform Shield platform directory and the MIT-licensed classic components.

OpenAgents is developing Sarah as an inspectable supervision layer for bounded plans distributed across named Codex, Claude, and Grok accounts, using local desktops or optional cloud capacity. The repository is early and changing quickly, and Sarah does not yet compose the existing fleet components into the complete daily workflow. Its current milestone is to manage at least three useful work streams at once.

Mem0 stores user, session, and agent state so assistants, support bots, and automated systems can recall preferences and prior interactions. Developers can integrate it through Python or JavaScript libraries, a CLI, a self-hosted server, or the managed platform. Its core workflow depends on an LLM and an embedding model, and some managed-platform optimizations are not included in the open-source SDK.

Flowise provides a visual environment for assembling AI agents and workflows, backed by a Node server, React interface, third-party integration nodes, and generated API documentation. It can be launched locally with Node.js, run in Docker, or deployed across several cloud providers, with a hosted Flowise Cloud option also available. The repository is licensed under Apache 2.0.

AutoGen provides multi-agent orchestration, tool use, and a visual studio for prototyping workflows in which agents work autonomously or alongside people. The project is now in maintenance mode and will not receive new features. Existing users can continue using it or follow the migration guide, while new projects are directed to Microsoft Agent Framework.

Hermes Agent is designed to retain useful context across sessions by creating skills from experience, searching past conversations, and nudging knowledge to persist. It can run on a small VPS, a GPU cluster, or serverless infrastructure, and Telegram access lets you interact with a cloud instance while it works away from your laptop.

Mastra gives TypeScript developers one framework for model access, autonomous agents, graph-based workflows, and AI application development. It can integrate with React, Next.js, and Node or run as a standalone server, supporting a path from prototypes to deployed products through a common interface for many model providers. Most of the core uses Apache 2.0, but enterprise features under `ee/` require a commercial license for production.

Dify combines visual AI workflows, model management, prompt testing, document-based RAG, agent tools, and operational logs in one application-development platform. Each application is exposed through APIs for integration with existing products. Teams can use Dify Cloud or self-host the Community Edition with Docker Compose; the repository uses the Dify Open Source License, based on Apache 2.0 with additional conditions.

Agno combines an SDK, the AgentOS runtime, and a control plane for teams building agent products in their own cloud. It keeps data, context, tools, permissions, memory, and human-review loops under the operator’s control, while providing production APIs, persistent storage, integrations, and observability for running the platform.

Aider pairs a language model with an existing Git repository, building a map of the codebase so it can make coordinated changes across larger projects. It automatically records edits in Git and can run linters and tests after changes, fixing issues they expose. It supports many hosted models as well as local ones, and can take context from IDE comments, images, web pages, or voice.

Oh My Pi is a terminal coding agent that works across many model providers and uses optimized reading, search, and hash-anchored editing tools. It brings LSP operations, real debugger sessions, browser access, persistent Python and Bun execution, and isolated subagents into one interface. It suits developers who want complex coding, review, and automation workflows without leaving the command line.

OpenClaw is a personal AI assistant that runs on your own devices and connects to common messaging channels such as WhatsApp, Telegram, Slack, and WeChat. It also supports voice and a controllable Canvas across Apple and Android surfaces. The onboarding command guides setup of the gateway, workspace, channels, and skills on macOS, Linux, or Windows.

LoopTroop turns a complex coding ticket into an interview-led plan, small execution beads, isolated worktree changes, and a final review. Its LLM council compares planning drafts, while failed implementation attempts can reset and retry with fresh context. It is designed for long-running feature work rather than quick edits, and the project recommends running unattended agents inside a disposable VM or sandbox.

Superpowers is a composable skill framework and development methodology for coding agents. It prompts clarification before implementation, confirms a specification in readable chunks, and guides worktrees, plans, TDD, subagent execution, two-stage reviews, and branch finishing. It supports several agent harnesses, including Claude Code, Codex, Cursor, and OpenCode, but must be installed separately for each one.

Deep Agents is an agent harness for long-running, multi-step work, with built-in planning, sub-agents, filesystem access, shell use, and persistence. Each part can be overridden or replaced without forking, and the framework can work with tool-calling frontier, open-weight, or local models.