mrkeyoor.com_
Fri 16 Oct 18:53 UTC

Best Open-Source AI Coding Agents (2026) — measured, not ranked by stars

We cloned and ran every agent on this page in a clean container; the numbers are ours. Each one got the same treatment: a fresh checkout, the project’s own install command, then its own build and test commands, on a 3-CPU, 8 GB sandbox with a 900-second cap per step. Where a project exposes no build or test target, we say so instead of guessing. The order below is our usability score first, stars only as a tie-break — which is why a 27k-star agent outranks a 119k-star one.

Twelve terminal and editor agents, each cloned and run on our own machine — install time, dependency count, build and test results.

01

crush

charmbracelet/crush · ★ 29k ▲76/7d · Go

Crush is a terminal coding agent that can edit a project, run tools, use language servers, and connect to several model providers. It gives developers one text interface for coding sessions without tying the workflow to a single model company.

Measured sandbox golang:1.24-bookworm · 3 cpu / 8g · install 166s · 446 deps · build 211s · tests 118/118 go test · 167,608 LOC · 642s total

SetupDocsCommunityMaturity

Verdict Our Crush run passed all 118 tests, but install, build, and test took 627 seconds combined, so this polished terminal client still asks for patience when you work on its source. Use it if model choice, MCP connections, and an interactive terminal matter to you.

Pick it if Developers who want an agentic coding interface inside the terminal.

Full review →
02

codex

openai/codex · ★ 129k ▲846/7d · Rust

Codex CLI is OpenAI's coding agent for working on a local repository from the terminal. It reads files, edits code, and runs commands under configurable approval and sandbox rules, using either a ChatGPT account or an OpenAI API key for model access.

Measured sandbox lab-node:22 · 3 cpu / 8g · install 27s · 1 deps · 170 MB · no build target · no test target · 2,070,775 LOC · 66s total

SetupDocsCommunityMaturity

Verdict Our Codex run installed one pnpm package in 27 seconds and used 170 MB, yet it never reached a build or test target, so the released CLI is the sensible path for users and the source tree is work for specialists. Use Codex when OpenAI model access, local repository tools, and explicit approval rules fit your development process.

Pick it if Developers who want an agent to inspect, edit, explain, and test a real local codebase.

Full review →
03

gemini-cli

google-gemini/gemini-cli · ★ 107k ▲112/7d · TypeScript

Gemini CLI is Google's open-source terminal agent for reading code, editing files, running shell commands, searching the web, and calling Gemini models. It also works in scripts, connects to MCP servers, reads project instructions, saves checkpoints, and links to an IDE companion.

Measured sandbox lab-node:22 · 3 cpu / 8g · install 46s · 1,336 deps · 798 MB · build 95s · tests 7,187/7,201 vitest · 703,120 LOC · 1,055s total

SetupDocsCommunityMaturity

Verdict Our Gemini CLI run installed 1,336 packages and built successfully, but 10 of 7,201 tests failed before the suite hit 900 seconds and npm audit reported 73 vulnerabilities. Use the stable package if your team already relies on Gemini and can confine file, shell, web, and MCP access.

Pick it if Developers who already use Gemini and want an agent inside the terminal.

Full review →
04

goose

block/goose · ★ 55k ▲181/7d · Rust

Goose is a local AI agent available as a desktop app, command-line tool, and embeddable API. It can edit code, run shell tools, automate other work, and connect to outside capabilities through MCP extensions while letting you choose the model provider.

Measured sandbox lab-rust:1 · 3 cpu / 12g · install 66s · 1,389 deps · build FAILED after 900s (900s cap) · tests FAILED after 900s (900s cap) · 429,650 LOC · 1,911s total

SetupDocsCommunityMaturity

Verdict Our Goose run installed 1,389 packages in 66 seconds, but both build and tests reached their 900-second caps while Rust was still compiling, so source contributors should expect a slow first feedback loop. The packaged app is the better route for people who want a provider-flexible local agent with MCP tools.

Pick it if Developers who want one local agent across a desktop app, terminal, and API.

Full review →
05

opencode

sst/opencode · ★ 214k ▲1.4k/7d · TypeScript

OpenCode is an open-source coding agent that works in a terminal and, in beta, as a desktop app. It gives developers a read-only planning mode and a full-access build mode, so the same tool can inspect a codebase or change it under different permission rules.

Measured sandbox lab-node:22 · 3 cpu / 8g · install FAILED after 900s (timeout) · deps: not measured · build: not measured · tests: not measured · 688,859 LOC · 915s total

SetupDocsCommunityMaturity

Verdict Our OpenCode install reached the 900-second cap without finishing, so contributors should prove the Bun workspace on their own machine before committing to it. The product is still a strong candidate for developers who want open source, model choice, and a clear split between planning and editing.

Pick it if Developers who want an open-source coding agent in the terminal.

Full review →
06

cline

cline/cline · ★ 70k ▲244/7d · TypeScript

Cline is a coding agent that reads and changes project files, runs terminal commands, and connects tools through MCP. It gives you approval controls and a choice of model providers across an editor, terminal, desktop app, or SDK.

Measured sandbox lab-node:22 · 3 cpu / 8g · install 20s · 2,446 deps · 3,897 MB · build 96s · tests 150/150 vitest · 857,522 LOC · 451s total

SetupDocsCommunityMaturity

Verdict Our Cline build passed in 96 seconds, but the full test command ended with 2 daemon shutdown failures, so source contributors should expect debugging before a clean release gate. Use it for approval-controlled coding with a provider you choose.

Pick it if Developers who want to approve file edits and shell commands while an agent works across a repository.

Full review →
07

qwen-code

QwenLM/qwen-code · ★ 29k ▲99/7d · TypeScript

Qwen Code is an open-source coding agent that runs in a terminal, IDE, desktop shell, daemon, or programmatic SDK. It can use Qwen or other model providers to inspect a repository, edit files, run tools, coordinate subagents, and connect to MCP servers.

Measured sandbox lab-node:22 · 3 cpu / 8g · install 816s · 1 deps · 1,398 MB · build 361s · tests FAILED after 900s (900s cap) · 4,383,804 LOC · 2,110s total

SetupDocsCommunityMaturity

Verdict Our Qwen Code run installed 1,937 packages in 348 seconds, used 1,475 MB, and built successfully, but tests were still running at the 900-second cap and npm audit found 9 known vulnerabilities. Try it if provider choice, MCP, and several client surfaces outweigh the size and release pace.

Pick it if Developers who want a terminal agent while retaining a choice of Qwen, OpenAI, Anthropic, Gemini, or local model endpoints.

Full review →
08

OpenHands

All-Hands-AI/OpenHands · ★ 91k ▲481/7d · TypeScript

OpenHands Agent Canvas is a browser control center for coding agents that read and change files on your laptop, a remote server, or a cloud backend. It lets you switch agents and execution locations and set up recurring work without managing each through a separate interface.

Measured sandbox lab-node:22 · 3 cpu / 8g · install 47s · 1,959 deps · 1,057 MB · build 24s · tests FAILED after 900s (900s cap) · 416,090 LOC · 980s total

SetupDocsCommunityMaturity

Verdict Our OpenHands run built successfully in 24 seconds, but tests timed out at 900 seconds and npm audit reported 42 known vulnerabilities, so adopt it only with time for dependency and runtime review. Agent Canvas is worth a restricted trial when switching agents and execution backends solves a daily problem.

Pick it if Developers who want OpenHands and external agents such as Claude Code in a shared browser interface.

Full review →
09

kilocode

Kilo-Org/kilocode · ★ 28k ▲67/7d · TypeScript

Kilo Code is an open-source AI coding agent that works in VS Code, JetBrains IDEs, the terminal, and a web-hosted cloud agent. It handles code generation, repository questions, debugging, review, autocomplete, and command execution, so developers do not need a different assistant for each interface.

Measured sandbox lab-node:22 · 3 cpu / 8g · install 80s · 1,996 deps · 2,629 MB · no build target · tests FAILED after 12s · 1,378,588 LOC · 105s total

SetupDocsCommunityMaturity

Verdict Our install completed in 80 seconds, but it pulled 1,964 packages and the root test command failed by design after 18 seconds. Kilo Code is worth trying if you genuinely want the same agent across VS Code, JetBrains, CLI, cloud, and automated review, especially when model choice matters.

Pick it if Developers who want one coding agent across VS Code, JetBrains, and the CLI

Full review →
10

aider

Aider-AI/aider · ★ 50k ▲109/7d · Python

Aider is a terminal coding assistant that sends selected repository context to a language model, edits files, and records changes in Git. It gives developers a shell-based alternative to an AI editor while letting them choose hosted or local models.

Measured sandbox ghcr.io/astral-sh/uv:python3.12-bookworm · 3 cpu / 8g · install 154s · 171 deps · 772 MB · build 8s · tests 474/492 pytest · 42,721 LOC · 778s total

SetupDocsCommunityMaturity

Verdict Our Aider run installed 171 packages and reached 772 MB, then pytest stopped with 9 failures and 9 setup errors, so adopting it carries real review and maintenance work. Use it when you want model choice and Git-visible edits in the terminal, with a developer watching every diff.

Pick it if Developers who prefer terminal sessions, Git diffs, and small reviewable commits over an editor sidebar.

Full review →
11

continue

continuedev/continue · ★ 36k ▲57/7d · TypeScript

Continue is an Apache-licensed coding agent available as a command-line tool, a VS Code extension, and a JetBrains plugin. It lets developers connect a coding workflow to language models, but its maintainers now describe the repository as read-only and version 2.0.0 as the final release.

Measured sandbox lab-node:22 · 3 cpu / 8g · install 35s · 319 deps · 107 MB · no build target · no test target · 331,015 LOC · 54s total

SetupDocsCommunityMaturity

Verdict Our Continue install took 16 seconds and reported 0 known npm vulnerabilities, but the maintainers now call 2.0.0 the final release and the repository read-only. Existing users have a usable endpoint to pin, while new users would be adopting 3,001 files of agent code without normal upstream maintenance.

Pick it if Developers who want an open codebase for a coding agent across the terminal and two major IDE families.

Full review →
12

plandex

plandex-ai/plandex · ★ 16k ▲29/7d · Go

Plandex is a terminal coding agent for changes that span many files and several steps. It builds a plan, keeps generated edits in a separate cumulative diff, and lets you review or revise the result before applying it to your project.

Measured sandbox golang:1.24-bookworm · 3 cpu / 8g · install 35s · 360 deps · build 44s · tests passed in 10s (0 collected) · 81,449 LOC · 115s total

SetupDocsCommunityMaturity

Verdict Our Plandex run installed 360 packages and built in 44 seconds, but its successful test command found 0 tests, so the clean build is weak evidence for unattended use. Try it only in an isolated local setup with manual apply review, especially while the path-containment reports against commit e2d7720 remain open.

Pick it if Developers tackling cross-file changes who want a written plan before edits reach the working tree.

Full review →

Also run, not ranked: Roo-Code installed 2,665 packages and passed build and tests, but the extension was shut down and the repository archived on 15 May 2026, so it is not a 2026 pick.

See also