Open-source AI agents you can run on your own Claude or ChatGPT subscription

Yes, you can run an open-source AI agent on a subscription you already pay for. OpenClaw, OpenCode, Kilo Code and Hermes Agent sign in with your ChatGPT Plus or Pro plan through OpenAI's official Sign in with ChatGPT, and Cline, Zed, T3 Code and Happy run on a Claude Pro or Max plan by driving the official Claude Code you logged in to. The model work then counts toward your plan's limits instead of an API bill.

The two providers treat open source differently, and that decides which agent you can use with which plan. This guide explains the difference, lists the main open-source options with their licences, and flags the routes to avoid. Checked on 5 October 2026.

Why do Claude and ChatGPT plans work differently with open-source agents?

OpenAI welcomes them. Its developer docs say plan usage through Sign in with ChatGPT is open to open-source projects, personal projects that run locally and selected private apps. OpenAI's own partner page lists four open-source integrations by name: OpenClaw, OpenCode, Pi and T3. An open-source agent can therefore sign you in to ChatGPT itself and send requests on your plan.

Anthropic only allows its own client. The Claude Code legal page says third-party developers may not "offer Claude.ai login into their own applications, or to route requests through Free, Pro, or Max plan credentials on behalf of their users." The exception is that you may sign in to "the unmodified Claude Code binary with your own Claude subscription, including where a platform hosts Claude Code."

So on a Claude plan, every open-source option is a front end, manager or remote control for Claude Code, not a replacement engine. On a ChatGPT plan, an open-source harness can be the whole agent.

Which open-source agents run on your plan?

Licences and star counts are from each project's GitHub repository on 5 October 2026.

App	What it is	Licence	GitHub stars	Claude plan	ChatGPT plan
OpenClaw	Personal assistant in your chat apps	MIT	~391,000	Yes, via your Claude Code login	Yes, OpenAI open-source integration
Hermes Agent	Nous Research's self-improving agent	MIT	~251,000	Experimental plugin that drives Claude Code	Yes, OpenAI partner
OpenCode	Terminal coding agent, also desktop and IDE	MIT	~212,000	No	Yes, OpenAI open-source integration
Zed	Code editor with agents	Open source (see repo)	~91,000	Yes, Claude Agent or Claude Code	Yes, ChatGPT Subscription provider
Cline	Coding agent for VS Code, JetBrains, CLI	Apache-2.0	~70,000	Yes, Claude Code provider	No
AnythingLLM Mobile	Local-first agent for Android	MIT	~67,000 (AnythingLLM repo)	No	Yes, marked experimental
NanoClaw	Lightweight assistant for WhatsApp, Telegram, Slack	MIT	~31,000	No	Yes, via /add-codex
Kilo Code	Coding agent for VS Code, JetBrains, CLI, cloud	MIT	~27,000	No	Yes, OpenAI partner
T3 Code	Control surface for Claude Code, Codex, Cursor	MIT	~25,000	Yes, drives Claude Code	Yes, OpenAI open-source integration
Happy	Phone remote for Claude Code and Codex	MIT	~24,000	Yes, runs vanilla Claude Code	Yes, runs Codex
Munder Difflin	An office of agents, one terminal each	MIT	~8,400	Yes, via Claude Code CLI	Yes, via Codex CLI

More open-source apps on your plan, for narrower jobs: Graphify (codebase knowledge graphs, Claude Code backend), Daymon (scheduled tasks for Claude), Sentō (always-on agents on Claude Code), Muesli (Mac dictation and meeting notes, ChatGPT), OpenChatCut (chat-driven video editor, Codex), Open CoDesign (design tool, ChatGPT), CoWork OS (local agent OS, ChatGPT), bise (multi-agent harness, ChatGPT) and CLI-JAW (one assistant over Claude, Codex and Cursor CLIs).

What is the best open-source alternative to Claude Code?

It depends on which plan you pay for.

• On a ChatGPT plan: OpenCode is the closest like-for-like terminal agent. Run /connect and log in with ChatGPT Plus or Pro. Kilo Code is the pick if you want it inside VS Code or JetBrains.
• On a Claude plan: there is no open-source engine that may use your plan directly. The practical answer is an open-source front end over Claude Code: Cline in your editor, Zed as the editor, or T3 Code and Munder Difflin to run several Claude Code agents at once.
• On both: T3 Code and Zed switch between Claude Code and Codex without changing tools.

Cline's docs note the trade-offs of its Claude Code mode: it "uses your Claude subscription limits instead of API token billing", but "responses may not stream token-by-token" and "image uploads and prompt caching are limited in this mode."

Which open-source personal assistant should you run?

OpenClaw and Hermes Agent are the two big ones, and both are MIT licensed.

• OpenClaw lives in your chat apps. Its onboarding signs in to ChatGPT with Codex OAuth. For Claude, its Anthropic docs describe a "Claude CLI" route that reuses the Claude Code login on the same machine. Skip its community "Claude Max API proxy": that is the pattern Anthropic prohibits.
• Hermes Agent is on OpenAI's partner list, so you can log into Nous Portal with ChatGPT. Its Claude support is an official plugin, marked experimental, that "drives the unmodified official Claude Code executable" and needs Hermes 0.21.4 or newer (plugin page).

Lighter options: NanoClaw on a ChatGPT plan, or Zylos if you want an always-on assistant on a Linux server that runs Claude Code or Codex.

Which open-source routes should you avoid?

Some open-source projects add Claude plan support in ways Anthropic's terms do not allow. We do not list those routes, and you risk your account by using them.

• OpenCode with Claude Pro or Max. OpenCode removed it at Anthropic's request, and its provider docs say Anthropic explicitly prohibits the plugins that add it back.
• Pi with a Claude login. Pi is open source and on OpenAI's list for ChatGPT plans. Its own Claude login shows Anthropic's notice that third-party usage "draws from extra usage and is billed per token" (issue #3372). Community extensions that change this are unofficial.
• Proxies, setup tokens and pasted session tokens. Anthropic's page says developers "may not collect, store, or intermediate Claude.ai credentials or session tokens." If an open-source agent asks you to paste a Claude token into its config, it is not on the permitted route, however popular the repo is. A few listed apps, including ZeroClaw and Craft Agents, offer their own Claude connection alongside a ChatGPT route; we recommend their ChatGPT route.

The rule of thumb: on a Claude plan, if the app does not ask you to install and log in to Claude Code, do not give it your Claude account.

How do you set one up?

1. Sign in once at the source. Log in to Claude Code with your Claude account, or run codex login and sign in with ChatGPT. Most apps above reuse that login. Step-by-step instructions, including the API-key trap, are in how to run other apps on your Claude or ChatGPT plan.
2. Install the app and pick the subscription provider. Each BringYourSubs listing has the exact menu or command, quoted from the maker.
3. Cap it on the ChatGPT side. In ChatGPT, Settings, Usage lets you limit each connected app's weekly share of your plan. Disconnect it under Settings, Security and login when you are done.
4. Watch your Claude limits. Run /usage in Claude Code to see which tools and MCP servers are using your allowance.

Running agents on a plan is usually cheaper than paying per token; Claude subscription vs API cost has the breakeven. For closed-source apps too, see the best AI agent apps that run on your plan, and for every app grouped by plan, works with Claude and works with ChatGPT.