Recently updated open-source projects

AI SDK lets TypeScript developers call multiple model providers through a consistent API for text, structured output, agents, and generative interfaces. It works with frameworks such as Next.js, React, Svelte, and Vue, either through Vercel AI Gateway or direct provider packages. Local development requires Node.js 22 or newer.

Crush is a terminal coding assistant that can switch models during a session while preserving context and keeping multiple project sessions. It uses language servers for additional code context and connects MCP servers over stdio, HTTP, or SSE. It supports many OpenAI- and Anthropic-compatible providers as well as local models; users need to configure a working provider, credentials, or local endpoint before sending model requests.

GitHub's official MCP Server connects compatible AI clients to repositories, code, issues, pull requests, Actions, and security-related data. You can use GitHub's hosted endpoint or run the server through Docker or a local binary, then limit the exposed surface with toolsets or individual tools. Available write operations depend on the permissions granted to the OAuth session or token.

Elixir Desktop lets Phoenix LiveView developers build native-like local applications for Windows, macOS, Linux, iOS, and Android. It routes window, webview, menu, and notification operations through platform backends, using wxWidgets on desktop, a JSON bridge on mobile, or a browser backend for headless environments. An example application and setup guides provide a starting point. Core desktop and mobile execution works, but complete desktop installer generation, signing, and one-command cross-platform packaging remain roadmap items.

GPT Researcher plans subquestions, gathers evidence in parallel from the web or local documents, tracks sources, and assembles the findings into long reports with citations and export options. Its model, retriever, and MCP data sources can be customized, but typical setups require provider and search API keys. The project is experimental and explicitly does not present its output as academic advice or a recommendation for research papers.

Claurst is an open-source Rust terminal coding agent with a TUI, multiple model providers, plugins, chat forking, memory consolidation, headless use, and ACP editor integration. It can work on code through an agent tool loop without telemetry. The core agent and routing are in beta, while sustained goals and its highest-effort multi-agent workflow remain experimental.

gstack organizes product discovery, planning, design critique, code review, browser QA, security checks, release, and retrospectives as connected Claude Code slash commands. Each stage passes artifacts such as design documents and test plans to the next, giving founders and engineering teams a repeatable way to run one or many development sprints.

Google ADK is a code-first Python toolkit for building, evaluating, and deploying AI agents. Its Agent and Workflow APIs support graph-based routing, nested workflows, and structured agent-to-agent tasks, while the CLI can run an agent locally or open a web UI for an agent directory. Version 2.0 changes the agent API, event model, and session schema, so existing applications need an upgrade check.

Microsoft Agent Framework targets Python and .NET applications that need more than a single prompt-and-response loop. It supports sequential, concurrent, handoff, and group-collaboration graphs, along with checkpointing, streaming, human-in-the-loop control, OpenTelemetry tracing, and Foundry hosting samples. Its scope suits teams building restartable, observable agent workflows rather than a small one-off script.

Gemini CLI lets developers work with Gemini directly from a terminal to inspect repositories, generate or edit code, debug issues, and run tool-assisted workflows. It includes file, shell, web, and search tools, supports MCP extensions and non-interactive scripting, and offers Google sign-in, API key, or Vertex AI authentication. A personal Google account can use the documented free tier.

The MCP Python SDK turns Python type annotations into tool schemas and offers both server and client APIs for in-memory, stdio, and remote HTTP connections. It supports the standard MCP transports and is suited to building or testing protocol integrations. The documented v2 line is a breaking-change-prone prerelease, so v1.x remains the stable production choice.

Inspect is an evaluation framework created by the UK AI Security Institute for testing large language models. It includes components for prompt construction, tool use, multi-turn dialogue, and model-graded evaluation, plus more than 200 pre-built evaluations. Researchers can compose these capabilities into tasks and extend the framework through Python packages that add new elicitation or scoring techniques.

OpenAI Agents SDK lets Python developers define agents with instructions, tools, guardrails, and delegation between specialized workers. It includes session history, human approval points, tracing, sandboxed long-running agents, and realtime voice support. The framework can use OpenAI APIs or other supported model providers and requires Python 3.10 or newer.

ChatDev 2.0 expands the original software-development agent system into a zero-code multi-agent orchestration platform. Users assemble nodes, tools, and models through configuration to run workflows such as document summarization, information collection, content publishing, or data visualization. A Python SDK and modular backend also let developers add new nodes, model providers, and tools.

Daytona lets agents create isolated compute sandboxes through SDKs, an API, or a CLI, then execute code, manage files and processes, and preserve state with snapshots. Clients are available for several languages, including Python and TypeScript. This public repository has been unmaintained since June 2026 because core development moved to a private codebase, so it receives no further fixes or support.

Strands Agents is a monorepo with Python and TypeScript SDKs for building model-driven agents. The SDKs support tools, MCP, streaming, multi-agent patterns, hooks, guardrails, context management, and multiple model providers, with examples for local development and deployment. Python requires 3.10 or newer and the TypeScript package requires Node.js 20 or newer; the default quickstart uses Amazon Bedrock credentials.

OMNX Code sets up and maintains a living CLAUDE.md for React, TypeScript, Supabase, and Vercel projects, installs a security-auditor skill through a verified update flow, and keeps tasks visible. It works with Claude Code or Codex and asks for explicit confirmation before security fixes or other risk-sensitive changes.

smolagents offers a compact Python framework for code-writing agents, conventional tool-calling agents, and multi-agent hierarchies, with support for different model providers and Hub sharing. Its CodeAgent expresses actions as Python snippets. Since those snippets are executed, untrusted workloads should use a sandboxed backend; the local Python executor is explicitly not a security boundary.