OpenClaw is a free, MIT-licensed assistant that you run on your own computer and talk to through WhatsApp, Telegram, Slack, Discord and 20+ other chat apps. It keeps its memory in plain Markdown files and runs scheduled jobs. Claude Code is Anthropic's coding agent for the terminal, your editor, the desktop and the web; installing it is free, but using it takes a paid Claude plan or an API key. They also connect: OpenClaw can run Claude Code as its model runtime or as a coding session, and Claude Code can read and send OpenClaw messages over MCP.
Checked against OpenClaw 2026.11.3 and Claude Code 2.1.322, both released on November 10, 2026, using each project's own docs.
| Capability | OpenClawOpenClaw Foundation · MIT | Claude CodeAnthropic · proprietary |
|---|---|---|
| What it is | A personal assistant you run yourself and talk to through chat apps | Anthropic's coding agent |
| Where you use it | Chat apps, a browser dashboard, the terminal, macOS and Windows apps, and iOS and Android apps | Terminal, VS Code and JetBrains, the desktop app (macOS, Windows), the web, and the Claude mobile app |
| Working while you are away | A background service on your computer with built-in automations: cron jobs, reminders and webhooks. Its old heartbeat now runs as an ordinary automation | Cloud scheduled tasks that run with your laptop closed (research preview), plus desktop scheduled tasks and cloud sessions |
| Reaching you in chat apps | WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Microsoft Teams and 20+ more | Channels for Telegram, Discord and iMessage (research preview, only while a session is open), and Claude in Slack |
| Memory across sessions | Plain Markdown files in its workspace: AGENTS.md, SOUL.md, USER.md, MEMORY.md and a daily log | CLAUDE.md or AGENTS.md rules, plus auto memory notes it writes itself, on by default |
| Skills | SKILL.md folders, and ClawHub, an open public registry of skills and plugins | Skills, plugins, subagents, hooks and MCP servers |
| Coding tools | File edits and shell commands; its docs hand bigger coding jobs to coding agents such as Claude Code, Codex or OpenCode | Plan mode, checkpoints you can roll back with /rewind, git worktrees and subagents |
| Models | Anthropic, OpenAI, Google, OpenRouter, DeepSeek, local models through Ollama and many more | Claude models only, through Anthropic, Amazon Bedrock, Google Cloud or Microsoft Foundry |
| Plans you can sign in with | API keys, ChatGPT sign-in through Codex, and a Claude plan through your installed Claude Code (see the FAQ) | Claude Pro ($20 a month), Max ($100), Ultra ($200), Team or Enterprise, or an API key |
| Approvals | Tools run on your computer; sandboxing is off by default. Exec modes range from deny to full, and an approval binds to the exact command | Permission modes from Manual to auto; deny rules hold in every mode; a shell sandbox on macOS, Linux and WSL2 |
| Windows | A native Windows app (Windows Hub) that runs its Gateway in WSL by default, or the native PowerShell CLI | Runs natively on Windows; its shell sandbox needs WSL2 |
Neither is a complete security boundary, and both projects say so: OpenClaw treats one Gateway as one trust domain, and Claude Code's sandbox covers shell commands only. Many comparisons online are out of date: OpenClaw has retired its heartbeat, and Claude Code now has auto memory, scheduled tasks and chat channels.
This comparison was last updated on November 12, 2026; check each vendor's site for current terms.
Many people searching this end up running both. The connections go both ways: OpenClaw can run Claude Code, and Claude Code can read and send OpenClaw messages.
Sign in to Claude Code first, then point OpenClaw's Anthropic provider at it. OpenClaw runs the claude program you installed and does not read your login tokens. To bring your CLAUDE.md instructions, MCP servers and skills along, run openclaw migrate claude --dry-run first.
claude auth login
openclaw models auth login --provider anthropic --method cli --set-defaultWith the acpx plugin, /acp spawn claude starts a Claude Code session that you can bind to a chat thread and resume later. The bundled coding-agent skill is another route: it runs Claude Code in the background in its own git worktree, but stays off until you enable it.
openclaw plugins install @openclaw/acpx
openclaw config set plugins.entries.acpx.enabled true
# then, in a chat:
/acp spawn claude --bind hereopenclaw mcp serve gives an MCP client such as Claude Code tools to list, read, wait for and send messages in your OpenClaw conversations. openclaw attach starts Claude Code with a temporary grant to one Gateway session and revokes it when Claude Code exits.
openclaw mcp serve
openclaw attachOrkas runs OpenClaw and Claude Code as two local agents in the same chat, next to Codex, OpenCode and Hermes. The next table lists what Orkas can and cannot hand each of them.
These rows come from Orkas's adapter code for each CLI, which is the same in Orkas 2.0 and the next release. They describe what a host can do with each agent, not how well either one works.
| In Orkas | OpenClaw | Claude Code |
|---|---|---|
| How Orkas starts it | openclaw agent --local, the embedded agent, so no Gateway is needed | claude -p, with streaming JSON in and out |
| Project folder | Works from the Orkas workspace | Bound to the project folder you choose |
| The next turn | The same OpenClaw session, picked up by its session ID | Continues the same session |
| Permission level | OpenClaw's own approval settings; Orkas offers no other level | Use CLI default, Ask for permission (you approve in Orkas) or Full access |
| Orkas skills and the services you connected | Available in every run | Available in every run, plus the agent's Orkas memory |
| While it runs | No live progress; the reply arrives when the turn ends, and new messages wait | Shows progress as it works, and you can send a message mid-run |
OpenClaw's local agent returns its result only at the end of each turn, so Orkas has nothing to show until then. Orkas checks for Claude Code 2.0.0 or later and OpenClaw 2026.4.11 or later, finds both on your PATH, can update either one, and installs neither. OpenClaw's chat channels and automations keep running in its own Gateway; Orkas does not start or manage it.
Orkas is a free, MIT-licensed desktop app for macOS and Windows. It runs the OpenClaw and Claude Code you already have, plus Codex, OpenCode and Hermes, as agents in one chat.
Install Claude Code and OpenClaw first and sign each one in to a model. Orkas finds them on your PATH and needs OpenClaw 2026.4.11 or later.
Open Connect agents and add both.
Set Claude Code's approvals in AI Team > Claude Code > Runtime settings > Permission level, then give each agent its part of the task in the same chat.
What Orkas does not do here: it does not install either CLI, show OpenClaw's progress mid-run or change OpenClaw's approval settings. OpenClaw's chat channels and automations stay in its own Gateway.
It can edit files and run commands, but its own docs send bigger coding jobs to a coding agent. It can run Claude Code, Codex or OpenCode as a session from a chat, and its bundled coding-agent skill hands work to them in a separate git worktree. Most people use OpenClaw to start and track the work, and a coding CLI to write the code.
Yes, in three documented ways. It can use your installed Claude Code as its model runtime after you run openclaw models auth login --provider anthropic --method cli once. It can open a Claude Code session from a chat with the acpx plugin and the /acp spawn claude command. It can also delegate through the coding-agent skill, which is off until you enable it. All three need Claude Code installed and signed in on the same computer.
OpenClaw documents a route that reuses your Claude Code login: it runs the claude program you installed and does not read the tokens. Anthropic's support page says Agent SDK, claude -p and third-party app usage draw from your plan's limits, and its terms do not stop you signing in to the unmodified Claude Code program with your own plan. But in April 2026 Anthropic told subscribers that third-party harnesses including OpenClaw could no longer use plan limits, as TechCrunch reported. Check Anthropic's current terms, or use an API key.
It depends on how you set it up. Its tools run on your computer unless you turn on sandboxing, and OpenClaw advises keeping the Gateway on loopback or behind authenticated private access and running openclaw security audit --deep to check. Treat ClawHub skills as code you choose to trust: in February 2026 Koi Security found 341 malicious skills among the 2,857 listed there. OpenClaw also advises against giving it full autonomy over your personal messages.
Yes. Orkas runs both as local agents in one desktop window, next to Codex, OpenCode and Hermes. Claude Code works in the project folder you choose, shows its progress and continues its own session. OpenClaw runs its embedded agent in the Orkas workspace, keeps its session between turns and replies when each turn ends.
Orkas is free and MIT licensed. It drives Claude Code, Codex, OpenCode, OpenClaw and Hermes as local sessions on your own machine.