Best coding agents (2026)

1. 1.AiderAI pair programming in your terminal with any model

   Free / OSS · Apache-2.0 · 49.6k GitHub stars

   VerdictThe open-source terminal coding assistant for API-key purists.

   CatchYou are paying per-token at API rates—can exceed subscription costs on heavy use. No managed infrastructure; you are the ops team. Less polished than commercial alternatives.

2. 2.ClineVS Code extension for human-in-the-loop AI coding

   Free / OSS · Apache-2.0 · 70.5k GitHub stars

   VerdictMaximum control for developers who approve everything.

   CatchHuman-in-the-loop means constant interruptions. Large tasks require many approvals. The pace is slower than autonomous alternatives.

3. 3.OpenAI Codex CLIApache-2.0 OpenAI agent: CLI, IDE, app and cloud

   Free / OSS · Freemium · 130k GitHub stars

   VerdictOpenAI’s answer to terminal agents — open CLI, paid models; budget the plan or the API.

   CatchThe CLI is open source, but the models are not free: you pay through ChatGPT plan quotas or API spend. Codex Web runs on OpenAI’s cloud, not on your machine. The install.sh, brew and npm routes are supply-chain choices, so pin a release binary when that matters. Vs Claude Code or OpenCode: this is OpenAI’s first-party agent, tied to ChatGPT plans or OpenAI API keys.

4. 4.GooseLocal-first Rust agent with MCP at its core

   Free / OSS · OSS/MIT

   VerdictLocal-first, MCP-native, model-agnostic. Bring your own brain.

   CatchGeneral-purpose agent means you configure it for your use case. You bring the model—Goose is the runtime, not the intelligence. Documentation is still maturing.

5. 5.OpenHandsSelf-hostable autonomous software engineer

   Free / OSS · OSS/MIT

   VerdictSelf-hosted Devin alternative for teams who want control.

   CatchHeavier than lighter agents like Goose. Self-hosting autonomous agents requires infrastructure and monitoring. The autonomy that makes it powerful also makes debugging failures harder.

6. 6.Gemini CLIGoogle open-source terminal agent powered by Gemini

   Free / OSS · Apache-2.0 · 107k GitHub stars

   VerdictGoogle terminal agent. OSS CLI remains, but consumer quota was cut 2026-06-18; Antigravity (agy) is the consumer path. Gemini-bound by design.

   CatchGemini-bound lock-in—not model-agnostic like Goose or Crush. Consumer access (free/Pro/Ultra Google login) ended 2026-06-18; enterprise and API keys still work. Don't build on the old Google-login quota path. The OSS CLI remains. Antigravity CLI is the successor, not a 1:1 swap. Preview/nightly channels move fast; pin stable for production workflows. Enterprise path pushes you toward Vertex AI and Google Cloud project setup.

7. 7.CrushCharm-grade TUI coding agent with LSP and mid-session model switch

   Free / OSS · Open-Core · 28.7k GitHub stars

   VerdictThe beautiful, model-flexible TUI coding agent. Charm quality.

   CatchLicensed FSL-1.1-MIT (competing-use restriction; source-available; MIT after two years—not OSI open-source). Pre-1.0 energy—config is migrating from JSON to crushrc Bash. Not a fully autonomous SWE agent like OpenHands; you stay in the loop. Hyper is Charm own model path if you want a subscription. Metrics on by default (DO_NOT_TRACK / CRUSH_DISABLE_METRICS to opt out).

8. 8.Pydantic AITyped Python agent loop with a snap-on Coder harness

   Free / OSS · OSS/MIT · 20.7k GitHub stars

   VerdictThe typed Python agent SDK. Not the validator. ACP is still experimental.

   CatchACP (Agent Client Protocol, for editors like Zed) is experimental: it lives under pydantic_ai_harness.experimental.acp, may change or be removed in any release without a deprecation period, and importing it emits HarnessExperimentalWarning. Coder is a capability you assemble, not a turnkey SWE agent like OpenHands. Logfire is the convenient observability path; OpenTelemetry works with any backend. Do not confuse this gem with Pydantic validation.

9. 9.WorktrunkGit worktrees that spawn an agent in one switch

   Free / OSS · Apache-2.0 · 9.3k GitHub stars

   VerdictThe worktree CLI for running several agents at once. On Windows, call git-wt.

   CatchOn Windows, `wt` is Windows Terminal's command. The installer ships `git-wt` to avoid the collision (`winget install max-sixty.worktrunk` then `git-wt config shell install`); you can also disable the Terminal app-execution alias. Project hooks require first-run approval. This is worktree plumbing, not a coding agent.

10. 10.ZedGPL-3.0 GPU-accelerated editor with first-party agents

    Free / OSS · GPL-3.0 · 91.9k GitHub stars

    VerdictThe fast open editor. GPL-3.0, agents in-process, no invented Cloud giveaway.

    CatchGPL-3.0 is copyleft, not MIT. Hosted agent features that talk to Zed's models are a product on top of the editor — no Cloud free tier to count on. Web is not shipped.

Also reviewed (57)

• Continue — Apache-2.0 open-source coding assistant — GitHub is source of truth
• OpenCode — MIT OSS coding agent: CLI, TUI, desktop; BYOK/local
• Kilo Code — MIT OSS IDE+CLI agent; active Cline/Roo line + MCP marketplace
• SWE-agent — MIT research OSS coding agent; BYOK; mini is successor
• Tabby — Apache-2.0 self-hosted coding assistant; IDE plugins, your GPU
• Open Interpreter — Apache-2.0 terminal coding agent; Rust Codex fork for open models
• Spec Kit — MIT Spec-Driven Development toolkit — any AI coding agent (GitHub)
• Open Code Review — Apache-2.0 hybrid deterministic+LLM review agent (MCP/plugins)
• OpenResearch — MIT coding agents → research agents (Rust)
• Atlas (pacifio) — MIT multi-agent change tracking — source control for agents
• OpenSpec — MIT spec-driven change tooling for coding agents
• Security Audit Skill — MIT coding-agent skill for multi-phase security audits
• Bend — Parallel language + proof-gated agent coding
• gawkbot — Source-available always-on bots that build microapps — local Grok Bot alternative
• Dexto — Elastic-2.0 agent harness OS — state, tools, recovery, ships a coding agent
• Orca — Parallel coding-agent ADE with worktrees, mobile control, and design mode
• T3 Code — Remote control surface for already-authenticated agent CLIs
• Caveman — Viral skill and proxy that forces coding agents to spend fewer tokens
• Jev Review — Local MCP that scores coding-agent diffs with TypeSafe Jev
• Prime Agent — Persistent REPL agent with recursive LMs
• GenericAgent — Tiny self-evolving agent seed with full system and Android control
• Codex-X — Codex-specific desktop and CLI workspace for visual configuration and MCP skills
• Pi — MIT agent harness with coding CLI, unified LLM API, and TUI
• Claude Code Templates — MIT npx catalog of Claude Code agents, hooks, settings, and MCPs
• Jev MCP — Ten TypeSafe Jev judgment tools over local stdio MCP
• ZCode — Apache coding workspace — desktop, web, and zcode Agent CLI
• AgentVerse OS — Apache personal cloud OS for you + coding agents on one Ubuntu box
• CLI-Anything — Apache hub that makes arbitrary software agent-native via CLI surfaces
• ARTEMIS — Apache Android NL automation + MCP for IDE coding agents
• HarnessRouter CE — Apache self-hosted unified API over Codex, Claude Code, DSH, and peers
• Whiteboard — MIT desktop canvas where humans + coding agents share specs
• CLI Proxy API — MIT local OpenAI/Claude/Gemini-compatible API in front of coding CLIs
• Multica — Source-available board where coding agents pick up human issues
• OpenAI Tunnel Client — Apache customer daemon for Secure MCP Tunnel to ChatGPT/Codex
• fx — Apache-2.0 tiny Zig Unix-like coding agent CLI (Vercel Labs)
• Qwen Code — Apache-2.0 multi-surface coding agent (Qwen lineage)
• Oh My Pi — MIT Stencil Labs Pi fork with IDE wired in (LSP/DAP)
• OpenRig — Apache-2.0 multi-harness agent team — Claude Code + Codex in one YAML rig
• DwarfStar 4 — MIT in-process coding agent — project quants only, shared KV cache
• Claude Code Action — MIT GitHub Action — Claude Code on your runner, model calls still bill
• awwwards-mcp — MIT design-inspiration MCP — Awwwards search, DNA, screenshots
• MCP for Blender — MIT Blender MCP — prompt-driven 3D scenes from any agent client
• Plannotator — MIT/Apache local browser review for agent plans, docs and diffs — feedback goes straight back to the agent
• Omnigent — Apache-2.0 meta-harness over Claude Code, Codex, Cursor, Pi and custom agents — sessions on any device, policies, sandboxes
• Dyad — Local, open-source AI app builder — a v0/Lovable/Bolt alternative on your own machine
• Codync — Apache-2.0 bots for Claude Code, Codex, Cursor and 40+ ACP agents — message them from phone or desktop
• Kandev — AGPL-3.0 multi-agent kanban with git worktrees — orchestrate, review diffs, open PRs
• Phone Harness — MIT CLI that lets coding agents drive a real phone
• Hex-Rays IDA MCP — Official MIT MCP server for the IDA disassembler
• Apache Maka — Apache-2.0 local-first agent workspace that logs every step
• ripwire — Apache-2.0 offline code map and quality checks for coding agents (CLI + MCP)
• OpenClaw 2.0 — Coding and personal agent across 29 messaging channels
• Antigravity CLI — Google terminal agent (agy) — Gemini CLI consumer successor
• Claude Code — Terminal-first coding agent with deep repository context
• Cursor — The AI-first code editor that understands your codebase
• Devin — Autonomous AI engineer for defined engineering tasks
• Amp — Sourcegraph coding agent — free with BYOK, orbs for long-running work