Recently updated open-source projects

LlamaIndex OSS supplies connectors, indexes, retrieval, and query components for grounding LLM and agent applications in documents, APIs, databases, and other private data. Teams can start with a bundled package or assemble the core library with selected model, embedding, and vector-store integrations. LlamaParse is a separate commercial platform for document parsing, OCR, extraction, and hosted document agents.

smelt combines an AI coding agent, a Vim-style editor, and its own terminal renderer in a compact CLI. Lua plugins can add keymaps, commands, tools, and modes, while provider support covers subscriptions, OpenAI-compatible APIs, and local Ollama models. The project is under active refactoring, so the latest prerelease is recommended over the older stable build.

E2B provides JavaScript and Python SDKs for starting isolated cloud sandboxes where agents can execute commands or generated code without running directly on the host. The managed service requires an E2B API key, and a separate Code Interpreter package adds higher-level code execution. Self-hosting is documented through Terraform, with current provider support centered on AWS and Google Cloud.

camofox-browser wraps Camoufox in a REST service built for agent browsing, using compact accessibility snapshots and stable element references for navigation, clicking, typing, and extraction. It isolates cookies and storage by user and supports proxies, imported login state, traces, and several deployment targets. The first install downloads roughly 300 MB, and anonymized crash telemetry is enabled by default but can be disabled.

Ralph TUI connects coding agents such as Claude Code, Codex, and OpenCode to PRD or Beads task trackers, then selects work, builds prompts, runs the agent, and detects completion in a loop. Its terminal interface exposes live output, subagent activity, pause and resume controls, and persistent sessions. It can also monitor and control remote instances from the same TUI.

This skill lets Claude create and execute Playwright automation for page tests, multi-step flows, and browser validation instead of relying on fixed scripts. The browser is visible by default, common interactions have helpers, and a universal executor handles module resolution; it can be installed as a Claude Code plugin or standalone skill.

Letta Code is a stateful agent harness built around persistent memory, identity, and experience over time. Agents can revise their own memory, skills, and prompts as they learn, while remaining accessible through a CLI, desktop app, browser, or messaging channels. Message search, subagents, and schedules support work that continues across sessions.

Agent Lightning connects existing agents to optimization workflows with minimal code changes by collecting prompts, tool calls, and rewards as structured traces. It can selectively optimize one or more agents built with LangChain, the OpenAI Agent SDK, AutoGen, CrewAI, Microsoft Agent Framework, or plain Python. Supported approaches include reinforcement learning, automatic prompt optimization, and supervised fine-tuning.

Khoj turns local or online language models into a personal assistant that can search documents, answer from the web, and create agents with custom knowledge, tools, and personas. It supports research automations and semantic search across formats such as PDF, Markdown, Notion, and Word. You can self-host it privately or use the hosted app from browsers and several productivity clients.

Cloudflare Agents uses Durable Objects to give each agent its own persistent state, storage, and lifecycle. The SDK includes callable methods, WebSockets, scheduling, AI chat, MCP, and workflow integration, with client helpers for real-time synchronization. Developers can start from the provided template or add the package to an existing Cloudflare application.

OpenCastle generates project-specific agents, skills, rules, and MCP settings for tools such as Copilot, Cursor, Claude Code, OpenCode, Windsurf, and Codex. You can use the team interactively in an IDE or define a Convoy for parallel batch work, with isolated Git worktrees, resumable execution, validation gates, and an observability dashboard.

Agent of Empires runs multiple coding agents in isolated Git worktrees, with optional Docker sandboxes and tmux-backed sessions that survive terminal closure. Its TUI and web dashboard show which agents are active, blocked, or waiting for input, while exposing terminals and diffs from a laptop, phone, or tablet. It is designed for Linux and macOS.

Semantic Kernel is a model-agnostic SDK for building, orchestrating, and deploying AI agents and multi-agent systems across Python, .NET, and Java. It supports multiple model providers, plugins, structured workflows, and vector-database integrations. The project now identifies Microsoft Agent Framework as its successor and provides a migration guide, which is relevant when maintaining or starting an application.

SWE-agent gives a selected language model tools to inspect a repository, edit code, and attempt a fix from a GitHub issue. It also supports custom coding tasks, SWE-bench experiments, and offensive-security challenges. The maintainers recommend the simpler mini-SWE-agent for most new users, while the EnIGMA mode currently directs users to SWE-agent 0.7.

R2R ingests documents, images, audio, and other content into a self-hosted knowledge base, then exposes hybrid search, cited RAG, knowledge graphs, and a multi-step research agent through a REST API. Python and JavaScript clients handle document management and retrieval. It can run in a lightweight mode or a fuller Docker deployment, and normally requires an API key for the selected language model.

Plandex is a terminal coding agent designed for large projects and real development tasks. It loads only the context needed for each step, generates and reviews changes across many files, and can debug builds, tests, deployments, or scripts. Users can choose between full autonomy and fine-grained control or combine models from different providers; Windows is supported only through WSL.

Toad combines a traditional shell with a terminal UI for working with coding agents such as Claude, Gemini, and Codex. Its agent catalog lets you find, install, and run compatible tools from one interface, while Agent Client Protocol support gives developers a way to add more agents.

AgentBox moves a project and supported agent settings into isolated environments so Claude Code, Codex, OpenCode, and similar tools can run in parallel on local Docker or supported cloud providers. Each box offers persistent shells plus browser and editor access, can resume quickly from checkpoints, and keeps Git credentials on the host. Local use requires macOS or Linux, Docker, and Node 20.10 or newer.