Coding agents that run in your terminal: Claude Code, Codex CLI, OpenCode, Gemini CLI, Aider and more. Compared on price, licence, models, MCP, headless use and sandboxing.
25 tools · facts checked 2026-10-28 · sorted by GitHub stars (read from GitHub 2026-10-28) · how we check
OpenCodebest for: Developers who want an open-source terminal agent they can point at any model provider, including local models. look elsewhere if: Teams that need a documented OS-level sandbox or a single vendor-supported model stack.
Claude Codebest for: Developers who already pay for Claude and want a scriptable, sandboxable agent that works across terminal, IDE, CI and phone. look elsewhere if: People who want an open-source tool or need to run local or non-Anthropic models.
Codex CLIbest for: ChatGPT subscribers or OpenAI API users who want an open-source, sandboxed terminal agent. look elsewhere if: Anyone who needs to run it on non-OpenAI or local models and wants that officially documented.
pibest for: Developers who want a small, hackable agent harness they can script, embed and customise, and who are fine adding their own isolation. look elsewhere if: Anyone who wants built-in approval prompts, sandboxing or sub-agent orchestration out of the box.
Gemini CLIbest for: Teams with a Gemini Code Assist licence or a paid Gemini/Vertex API key who want an open-source CLI agent. look elsewhere if: Individuals looking for a free tier or the newest Gemini models; Google steers them to Antigravity CLI.
Cline CLIbest for: Developers who want an open-source terminal and CI agent they can point at any provider, including local models. look elsewhere if: Users who need a documented command sandbox out of the box or a single fixed-price plan covering all model usage.
Open Interpreterbest for: Developers who want a Codex-style terminal agent they can point at cheap or local open models with sandboxing on. look elsewhere if: Anyone looking for the old Python 'let the model run code on my computer' tool or a ready-to-pay hosted service.
goosebest for: Users who want an open-source agent with both a GUI and a CLI, MCP extensions and the freedom to choose any model, including local ones. look elsewhere if: Anyone who needs a hardened, sandboxed execution environment built in.
Aiderbest for: Developers who want a git-centred, model-agnostic terminal pair programmer they can also call from shell scripts. look elsewhere if: Teams that need MCP tools, parallel subagents or a sandboxed autonomous agent.
Codewhalebest for: Developers who want an MIT-licensed Rust agent that can switch between many cloud providers and local models. look elsewhere if: Windows users who need OS-level command sandboxing.
Oh My Pibest for: Developers who want a pi-style terminal agent with LSP, debugging and subagent orchestration already included. look elsewhere if: Users who want a minimal agent or a default-deny permission model out of the box.
DeepSeek-Reasonixbest for: Developers who want a free, MIT-licensed agent with a desktop app, remote SSH sessions and a DeepSeek-friendly default. look elsewhere if: Anyone who wants a vendor-backed DeepSeek product with official support.
Crushbest for: Terminal users who like Charm's TUI style and want MCP and LSP support with a choice of cloud or local models for personal or internal use. look elsewhere if: Anyone who needs an OSI-approved open source license, e.g. to embed or resell it in a competing product.
Qwen Codebest for: Developers who want an open-source Claude Code-style agent that can point at Qwen, other cloud APIs or a local model. look elsewhere if: Anyone looking for a zero-cost hosted model out of the box, now that the free OAuth quota is gone.
Prime Agentbest for: Researchers and power users who want long-running, programmable agents with subagents and a persistent Python environment. look elsewhere if: Anyone who needs a sandboxed agent out of the box or a well-documented list of supported models.
Codebuffbest for: Developers who want a hosted multi-agent coding CLI and an SDK, and are fine paying per credit or per plan. look elsewhere if: Users who want to run their own model keys or local models with a paid-free tool; for a free option look at Freebuff or other agents.
MiMoCodebest for: Developers who want an OpenCode-based agent with built-in memory and checkpoints, using MiMo models or their own provider. look elsewhere if: Users on Terminal.app who do not want to switch terminals, or those who need a documented sandbox.
GitHub Copilot CLIbest for: Developers already on GitHub Copilot who want an agent tied to their repos, issues and pull requests. look elsewhere if: People who need open-source code or who do not want a GitHub account and subscription.
mini-SWE-agentbest for: Researchers and tinkerers who want a small, transparent agent loop for benchmarking models or hacking on their own harness. look elsewhere if: Teams wanting an out-of-the-box product with IDE integration, approvals UI and managed support.
Kimi Code CLIbest for: Developers who already use Kimi models and want a fast single-binary terminal agent with scripting support. look elsewhere if: Users who need OS-level command sandboxing or a documented local-model workflow.
Forgebest for: Developers who live in a shell (especially ZSH) and want a free, scriptable agent with their own model keys. look elsewhere if: Users who need a documented, working one-line installer today or published pricing for the hosted offering.
Mistral Vibebest for: Developers who want an Apache-licensed terminal agent tied to Mistral/Devstral models with scriptable runs. look elsewhere if: Users who want a documented, built-in OS sandbox for every command on Windows.
Factory Droidbest for: Teams and individuals ready to pay for a managed agent platform that spans CLI, CI and chat tools. look elsewhere if: Users who want a free, open-source terminal agent or fully offline operation without a Factory account.
Ampbest for: Developers who want a hosted, multi-device coding agent they can start from a phone and leave running remotely. look elsewhere if: Anyone who needs open-source code, fully offline use with local models, or fixed predictable pricing.
Cursor CLIbest for: Existing Cursor users who want the same agent, MCP setup and models in a terminal or CI. look elsewhere if: Anyone who needs an open-source tool, local models, or bring-your-own-key billing.
What a terminal coding agent is
A terminal coding agent is a program you start in a shell. You describe a task, and it reads your files, edits code and runs commands through a language model. Every tool in our data can also run non-interactively, for scripts or CI. The tools differ mostly in who controls model access, how they are licensed and how much they isolate what they run.
How to choose
Model lock-in. Claude Code runs Claude models only, and its docs say routing to non-Claude models through a gateway is not supported. Codex CLI is tied to OpenAI models in the official docs. OpenCode lists 75+ providers, and Cline CLI supports 30+ providers plus any OpenAI-compatible API.
Licence. We count a tool as open source only when the licence file is OSI-approved. Codex CLI, Gemini CLI and Aider use Apache-2.0, and OpenCode and pi use MIT. Crush uses FSL-1.1-MIT, which bars competing commercial use. Claude Code and Copilot CLI are proprietary.
Cost. A free CLI does not mean a free model. Codex CLI is Apache-2.0, but usage goes through a ChatGPT plan or an OpenAI API key. Claude Code needs a paid Anthropic plan or an API account. Aider is free, but you pay your model provider per token.
Headless and sandboxing. Claude Code has claude -p, Codex CLI has codex exec, and Gemini CLI has -p with JSON output. Gemini CLI documents three sandbox options. pi has no built-in sandbox, and Oh My Pi defaults to an approval mode that runs tool calls without prompts.
Local models. OpenCode documents Ollama, LM Studio and llama.cpp setups, and Codewhale detects a running Ollama. For Claude Code the entry is marked false. For Gemini CLI and Kimi Code CLI we found no documented local setup.
What changed recently
Gemini CLI: per Google's blog, it stopped serving Google AI Pro and Ultra users and free Code Assist for individuals on 2026-06-18. Code Assist Standard or Enterprise licences and paid API keys keep access. Google points individuals to Antigravity CLI.
Qwen Code: the free Qwen OAuth tier was discontinued on 2026-04-15.
OpenCode: the repo moved from sst/opencode to anomalyco/opencode. Two major lines, v1 and v2, use different package names and install commands.
Kimi: the older Python Kimi CLI is archived. Kimi Code CLI, a TypeScript rewrite, replaces it, and credentials do not migrate.
Open Interpreter: now a Rust rewrite forked from Codex CLI, not the original Python project.
Aider: the last GitHub release is v0.86.0 from August 2025, and the last commit on main is from May 2026. No pause has been announced.
What we left out and why
We did not list tools that stopped being maintained. Plandex has had no release since July 2025 and its hosted cloud stopped taking new users. Continue CLI's README says it is no longer actively maintained. We also skipped tools that distribute code without the rights holder's permission, and graphical apps that are not terminal coding agents. Antigravity CLI, which Google names as the replacement for Gemini CLI, is not in our data yet.
