ИИ-агенты
holaOS
Local-first agentic workspace where Claude Code, Codex and a built-in agent share memory, apps and tools.
★ 11,5 тыс. звёздModified Apache-2.0
Смотреть руководство →Kun — проект с открытым исходным кодом из категории «ИИ-агенты» от KunAgent (KunAgent/Kun на GitHub). У него 6 319 звёзд на GitHub и 594 форка, он распространяется по лицензии PolyForm Noncommercial 1.0.0 и последний раз обновлялся на GitHub 7 октября 2026.
A local-first AI agent workspace for coding, writing, design, research and automation.
Pick a real task — a code change, a document, a research job — and note its constraints and acceptance criteria.
Download the desktop app for macOS, Windows or Linux, or clone the repo to run from source.
Add at least one model connection (an OpenAI-compatible endpoint or a supported provider) in the settings.
State the goal; Kun plans the steps, executes with visible tool calls and file changes, and delivers with diff and test evidence for your review.
The desktop GUI and the terminal TUI share the same Kun runtime, so you can move between visual and terminal work.
Agents work from an explicit plan and todo list; tool calls, file changes, browser/terminal results and approvals stay attached to the task.
Create a room, choose agent profiles, models and authorized local repos, and collaborate via private or group chat with coordinator or peer modes.
Each execution task runs in its own Git worktree and is delivered as an immutable version for a reviewer, with declared checks, diffs and logs kept in the room.
Requirements and plans can live inside your project, so they enter version control, code review and later recovery.
Connects to MCP tools and OpenAI-compatible model endpoints, including DeepSeek.
Clone Kun to run it from source.
git clone https://github.com/KunAgent/Kun.git
cd KunInstall dependencies (Node.js 22.19+ required) — the dev script builds the runtime.
npm installLaunch the Electron development environment, or use the TUI variant.
npm run dev
# TUI instead: npm run dev:tuiKun is maintained by the KunAgent organization.
Kun is source-available under the PolyForm Noncommercial License 1.0.0 — free for noncommercial use; commercial use is not covered by that license.
You need at least one working model connection; Kun supports OpenAI-compatible endpoints, so you can use your own provider or a local/self-hosted endpoint.
Kun is a local-first workspace spanning coding, writing, design and research, with plan-driven tasks, Git-worktree isolation and multi-agent Rooms — not just an editor or a chat window.
Install the desktop app (or npm run dev from source), connect a model, and give Kun a goal with constraints.
Local-first agentic workspace where Claude Code, Codex and a built-in agent share memory, apps and tools.
Local-first AI chat workspace for models.
ZCode — Z.ai's AI coding workspace: desktop app, browser interface and terminal agent in one open-source harness
Community-built terminal coding agent: bring your own model, keep code local.
Full-stack AI agent for coding, research, content, and multi-agent workflows.
A fast, local-first open-source agent harness written in Rust, with desktop, browser and terminal apps
Другие проекты из категории «ИИ-агенты» с наибольшим числом звёзд в каталоге LibreAI — реалистичный список альтернатив, если Kun вам не подходит.
A personal AI that really does things — on any OS, with your data staying yours.
An agentic skills framework & software development methodology that works.
Agent harness optimization system with skills, memory and security for Claude Code, Codex and more
The agent that grows with you — a Python framework for adaptive AI agents.
DeepSeek's plugin-based agent harness where everything is a plugin.
Соседние проекты из категории «ИИ-агенты» в каталоге, ближайшие по числу звёзд к Kun — продолжайте изучать полку по обе стороны от него.
A Large Action Model framework for building AI web agents that browse and act for you.
A control plane with a durable state kernel that keeps long-horizon agents moving across sessions.
Policy, identity, and sandboxing guardrails for AI agents.
Open-source LLM observability: monitor, evaluate, and experiment.
An MCP server and CLI that lets AI agents build, test and run iOS and macOS projects.
An Agent OS with interview-gated, budgeted evolution — one MCP server across 14 agent runtimes.
Бесплатно и открыто — забирайте исходный код, ставьте звезду репозиторию и начинайте создавать.