Local AI coding agents
Coding agents you can point at a model on your own computer, so prompts and code can stay on your machine and there is no per-token bill.
By weekly package downloads, the leaders are OpenAI Codex (26.6M), Pi (6.4M), OpenCode (3.9M). By VS Code extension installs, the leaders are GitHub Copilot (79.4M), OpenAI Codex (16.9M), Cline (6M). By GitHub stars, the leaders are OpenCode (228K), OpenAI Codex (136K), Pi (123K). Among agents with at least 20,000 downloads in the last 30 days, the fastest grower is Crush, with downloads up 72%. As of Nov 20, 2026.
-
2
84 26.6M up 7.4% 136K 7d ago 26.6M downloads a week up 7.4% 30dGid 84 Free + $8/moOpenAI · Terminal · terminal agent + IDE extension + cloud agent
OpenAI's coding agent, offered as an open-source terminal CLI, an IDE extension, a desktop app and a cloud agent that works on repositories. Users sign in with a ChatGPT account or use an OpenAI API key; with an API key, Codex works in the CLI, SDK and IDE extension without cloud features such as GitHub code review and Slack. On the Free and Go plans Codex runs in the desktop app (subject to rollout); the web, CLI and IDE extension come with Plus and higher.
-
3
80 3.9M up 28.7% 228K 2d ago 3.9M downloads a week up 28.7% 30dWikipedia from Nov 15Gid 80 Free + $10/moAnomaly · Terminal · terminal agent + desktop app + IDE extension
Open-source coding agent from Anomaly (the team behind SST) that runs as a terminal UI, desktop app or IDE extension and connects to 75+ model providers, including local models and existing ChatGPT or GitHub Copilot logins. OpenCode 2 is now available; per the docs, OpenCode 1 and 2 both use the opencode command and cannot be installed side by side under the default command.
-
5
75 — — — — 79.4M VS Code installsGid 75 Free + $10/moGitHub · IDEs · AI coding assistant (IDE agent mode, CLI and cloud agent)
AI coding assistant from GitHub offering completions, chat and agent mode in IDEs, a terminal CLI, a desktop app and a cloud agent that works on assigned tasks in the background and opens pull requests. Models from several providers can be selected.
-
7
73 6.4M up 36.3% 123K 10d ago 6.4M downloads a week up 36.3% 30dGid 73 Free (OSS)Earendil Inc. · Terminal · terminal agent
Minimal, extensible open-source coding agent harness with interactive TUI, print/JSON, RPC and SDK modes, created by Mario Zechner and now maintained by Earendil Inc. It supports 15+ model providers through API keys or OAuth and is customized with TypeScript extensions, skills and packages. Its tools and extensions run with the permissions of the Pi process; Pi does not sandbox tool calls or ask before each one.
-
12
64 — — 95.3K 2d ago 95.3K GitHub starsGid 64 Free + $10/mo
-
20
58 84.3K up 14.9% 71.8K 25d ago 84.3K downloads a week up 14.9% 30dGid 58 Free + $9.99/moCline Bot Inc. · Terminal · VS Code extension + terminal agent
Open-source autonomous coding agent for VS Code, JetBrains, Cursor, Windsurf, Antigravity and Zed (Neovim via ACP) that also ships a CLI, a desktop app and an SDK. It works with many model providers through the user's own keys, Cline's usage-billed provider or the ClinePass subscription.
-
22
56 1.7M up 8.3% 11.4K 2d ago 1.7M downloads a week up 8.3% 30dGid 56 Free + $10/moGitHub · Terminal · terminal agent
GitHub's standalone terminal coding agent (`copilot`, npm @github/copilot) that plans and edits code, runs commands and works with GitHub from the terminal. It runs on a Copilot plan and also supports bring-your-own model providers.
-
32
53 155K up 69.8% 30.8K 3d ago 155K downloads a week up 69.8% 30dGid 53 Bring your keyAlibaba Cloud · Terminal · terminal agent + IDE extensions
Alibaba's open-source terminal coding agent for Qwen models that also offers a desktop app, web UI, IDE plugins (VS Code, Zed, JetBrains) and SDKs. It accepts OpenAI-, Anthropic- and Gemini-style provider APIs as well as local models.
-
33
53 60K up 65.1% 28.1K 15d ago 60K downloads a week up 65.1% 30dWikipedia from Oct 30Gid 53 Free + $19/moKilo Code · Terminal · VS Code extension + terminal agent
Open-source coding agent available as a VS Code extension, JetBrains plugin and CLI, with a gateway to 500+ models and support for bring-your-own keys and local models. Kilo launched in March 2025 and was acquired by Anaconda in July 2026.
-
40
51 3.1M down 1.7% 9,771 4mo ago 3.1M downloads a week down 1.7% 30dGid 51 Free (OSS)Princeton University & Stanford University · Terminal · terminal agent (research)
Open-source software-engineering agents from Princeton and Stanford researchers: SWE-agent (2024) and its roughly 100-line successor mini-SWE-agent, which uses only bash and works with any model through LiteLLM and similar routers. The SWE-agent README says development effort has moved to mini-SWE-agent.
-
45
50 116K up 33.4% 50.2K 15mo ago 116K downloads a week up 33.4% 30dGid 50 Free (OSS)
-
48
49 — — 56.8K 2d ago 8,142 VS Code installsWikipedia from Oct 31Gid 49 Free (OSS)Block · Terminal · terminal agent + desktop app
Open-source, general-purpose AI agent that runs locally as a desktop app or CLI and can write code, run commands and use MCP extensions with 15+ LLM providers. Created by Block, it is now hosted by the Agentic AI Foundation at the Linux Foundation.
-
54
46 721 down 24.1% 567 2d ago 721 downloads a week down 24.1% 30dGid 46 Free + $8.33/mo
-
61
45 105K — 93.8K 2d ago 105K downloads a weekWikipedia from Oct 31Gid 45 Free + $500/moOpenHands · Terminal · agent platform (web GUI + terminal UI/CLI + cloud)
Open-source AI software-development agent platform (formerly OpenDevin) with an open-source browser client (Agent Canvas; the older Docker-based local GUI is deprecated), a terminal UI/CLI, a hosted cloud and an SDK. It is model-agnostic and can use the user's own keys.
-
81
41 11.3K up 71.9% 29.1K 8d ago 11.3K downloads a week up 71.9% 30dGid 41 Free
-
137
34 206K down 20.5% 5,426 2d ago 206K downloads a week down 20.5% 30dGid 34 Free + $14.99/moMistral AI · Terminal · terminal agent + VS Code extension
Mistral's coding agent, used as the `vibe` CLI, a VS Code extension (with a JetBrains plugin and web version) and powered by Devstral models. Since May 2026 Mistral brands its whole assistant, formerly Le Chat, as Vibe.
-
168
29 — — 68.9K 9d ago 68.9K GitHub starsGid 29 Free (OSS)Open Interpreter · Terminal · coding agent (CLI)
Open Interpreter is now a Rust-based terminal coding agent, a fork of OpenAI's Codex CLI, tuned to get strong results from low-cost open models such as Kimi K3, DeepSeek, Qwen and GLM. The original Python 'let language models run code' project continues as a community-maintained fork.
-
172
28 5,095 down 20.4% 41.2K 27d ago 5,095 downloads a week down 20.4% 30dGid 28 Free
No agents match these filters.
“#” is the position among products of the same kind (agents, frameworks, tooling) by Gid Score, our 0–100 composite of adoption, community, attention and momentum. Use the filters and sorting above, or switch to the table view. Methodology · All AI agents.
How each agent connects to a local model
Only agents whose own documentation describes a local setup are listed; each link goes to that page. Checked Oct 6, 2026.
| Agent | How it connects | Setup docs |
|---|---|---|
| OpenAI Codex | --oss mode with Ollama or LM Studio |
official docs |
| OpenCode | Ollama, LM Studio, llama.cpp or any OpenAI-compatible local server | official docs |
| GitHub Copilot | VS Code bring-your-own-model with the Ollama extension | official docs |
| Pi | llama.cpp, Ollama, LM Studio and vLLM via a custom base URL | official docs |
| Zed | Ollama, LM Studio, llama.cpp and other local OpenAI-compatible servers | official docs |
| Cline | Ollama and LM Studio providers | official docs |
| GitHub Copilot CLI | Bring your own model: Ollama, vLLM, Foundry Local or any OpenAI-compatible endpoint | official docs |
| Qwen Code | Ollama or vLLM through an OpenAI-compatible provider | official docs |
| Kilo Code | Ollama and LM Studio providers | official docs |
| mini-SWE-agent | Local servers such as vLLM through LiteLLM | official docs |
| Aider | Ollama, or any OpenAI-compatible local server | official docs |
| Goose | Ollama, Docker Model Runner and Ramalama providers | official docs |
| Junie | Junie CLI with LM Studio or Ollama | official docs |
| OpenHands | LM Studio, Ollama, vLLM and SGLang | official docs |
| Crush | Auto-discovers Ollama, LM Studio and llama.cpp servers | official docs |
| Mistral Vibe | Any OpenAI-compatible local server (vLLM, llama.cpp, LM Studio, Ollama) | official docs |
| Open Interpreter | Built-in Ollama and LM Studio providers | official docs |
| CodeWhale | Local runners such as Ollama, vLLM and SGLang, with no API key | official docs |
Why run a coding agent locally
Privacy. With a local model, your code and prompts are not sent to a model provider. (Check the agent's own telemetry settings too.) That matters for client code, regulated work, or a company policy that forbids sending source to third parties.
Cost. There is no per-token bill. You pay once, in hardware, and in electricity.
Offline work. Once the model is downloaded, open-source agents can keep working without a connection. Extensions of commercial products may still need you to be signed in.
What to expect
Local models are smaller than the frontier models behind commercial agents, and coding agents are demanding: they read many files, keep a long context and call tools in loops. In practice that means:
- Hardware decides what you can run. A laptop with 16 GB of memory runs small models that can handle simple edits and explanations. Models that can drive a multi-step agent loop reliably need much more memory, typically a workstation GPU or a recent Mac with a large amount of unified memory.
- Context length matters as much as model size. Local servers often start with a short default context window; agents need a long one to see enough of the codebase. Check your server's context setting before judging a model.
- Tool calling must be supported by the model and the server, or the agent cannot run commands or edit files.
A common setup is hybrid: a local model for routine or sensitive work, a hosted model for the hard tasks. Agents that let you switch models mid-session make that easy.
Agents that do not run locally
Some popular agents work only with their vendor's hosted models, or reach a model on your machine only through a public tunnel. For example, Claude Code documents that it does not route to non-Claude models, and Gemini CLI connects only to Google's backends. Their pages on this site list which models they support.
Questions about local AI coding agents
Which AI coding agents can run with a local model?
18 agents tracked here document local-model support, including OpenAI Codex, OpenCode, GitHub Copilot, Pi, Zed. Most connect through Ollama, LM Studio or another OpenAI-compatible server running on your machine.
Is a local coding agent completely free?
The open-source agents are free to install, and a local model has no per-token cost. You still need hardware capable of running a useful model, and some agents (such as editor extensions of commercial products) may require an account or plan for other features.
Can Claude Code or Cursor use a local model?
Not as a documented local setup. Claude Code's documentation says it does not route to non-Claude models. For Cursor, staff on its official forum explain that requests go through Cursor's servers, so a model on your machine would need a public HTTPS tunnel.