Open-source AI coding agents that actually work in 2026

Agent Governance Toolkit (AGT) is Microsoft's MIT-licensed policy enforcement, zero-trust identity, execution sandboxing, and SRE toolkit for autonomous AI agents, covering all 10 OWASP Agentic Top 10 risks with a deterministic, fail-closed policy kernel that sits outside the prompt.

jcode is an MIT-licensed terminal coding-agent harness benchmarked at 6-21x lower RAM usage and up to 245x faster startup than Claude Code, Cursor Agent, and Codex CLI, with a semantic agent-memory system, multi-agent "Swarm" collaboration, and native browser automation.

TencentDB Agent Memory is Tencent's MIT-licensed memory hub for AI agent teams, turning conversations, docs, and code into four governed, shareable memory assets — Chat Memory, Skill, Wiki, and CodeGraph — with a layered L0-L3 distillation pipeline and a reported +59% PersonaMem score.

AirLLM is an Apache-2.0 Python library that runs 70B+ LLMs — including Llama 3.1 405B on 8GB and DeepSeek-V3 671B on ~12GB — by streaming one layer (or one MoE expert) onto the GPU at a time instead of loading the whole model, no quantization, distillation, or pruning required.

aisuite is Andrew Ng's MIT-licensed Python library for building with LLMs, giving you one OpenAI-style Chat Completions API across OpenAI, Anthropic, Google, Ollama, and more, plus a first-class Agents API with tools, toolkits, policies, and native MCP support — and the engine behind the OpenWorker desktop app.

ego lite is a macOS browser (free, closed-source binary) with an MIT-licensed `ego-browser` skill that lets Claude Code, Codex, or Cursor run browser tasks in isolated Spaces alongside your own tabs, inheriting your real logins instead of driving a separate, cookie-less browser instance.

Open Code Review (the `ocr` CLI) is Alibaba's Apache-2.0 open-sourced AI code reviewer, incubated from an internal tool that reviewed millions of defects across two years of production use, combining deterministic file selection with an agent for line-precise, git-diff-aware review comments.

Kronos is an MIT-licensed, decoder-only foundation model pre-trained on K-line (OHLCV) data from 45+ global exchanges, accepted at AAAI 2026, that turns candlestick forecasting into a language-modeling problem you can run, finetune, and backtest yourself.

ZeroClaw is an open-source, Rust-built framework for running a fully autonomous AI personal assistant across 30+ messaging channels and 20+ LLM providers, with sandboxed tool execution and even GPIO/hardware support for embedded devices.

OfficeCLI is an open-source, single-binary CLI that gives AI agents full control over .docx/.xlsx/.pptx files — path-based element access, a built-in HTML/PNG rendering engine, and no Office installation required.

DeepTutor is an open-source, agent-native learning workspace from HKUDS combining chat, quizzes, research, and mastery practice on one memory system — with an arXiv paper behind it and a CLI built for other agents to drive.

Open Interpreter is now a Rust rewrite forked from OpenAI's Codex, built to emulate the exact agent harness (Claude Code, Kimi Code, Qwen Code) that gets the best results out of low-cost and open models.

OpenCut is a free, open-source (MIT) video editor for web, desktop, and mobile, positioned as a privacy-respecting CapCut alternative. Its GitHub repo is currently a ground-up rewrite; the classic codebase is archived.

BrowserOS is an AGPL-3.0 open-source Chromium fork with a built-in AI agent — and the same repo also ships BrowserClaw, a separate browser your coding agents drive using your own logged-in sessions.

screenpipe records your screen and audio locally and exposes it to AI agents via MCP and a REST API. Important distinction from most tools on dibi8 — it is source-available with a commercial license, not open source.

nanobot is an open-source, self-hosted personal AI agent framework from HKUDS with a bundled WebUI, MCP tools, long-term memory, chat-app integrations, and an OpenAI-compatible API, built around a small readable core.