OpenClaw vs Claude Code: purpose, models, license, security

OpenClaw vs Claude Code compares two different kinds of tool: OpenClaw is an MIT-licensed, self-hosted personal assistant that runs a Gateway on your machine and answers you in chat apps such as WhatsApp, Telegram, Slack and Discord, with swappable model providers; Claude Code is Anthropic’s proprietary coding agent for the terminal, IDEs, desktop and web, running Claude models on a Claude plan or API key.

Applies to: OpenClaw 2026.9.7 (npm) and Claude Code 2.1.285, docs as of October 2, 2026. Last checked: October 2, 2026.

On this page

What each one is

OpenClaw. The project README describes “an open-source AI assistant that runs on your own computer and meets you in the channels you already use: Discord, iMessage, Slack, Teams, Telegram, WhatsApp, and 20+ more, plus native apps for macOS, iOS, Android, Windows, and Linux” (openclaw/openclaw). The docs call it “a self-hosted gateway that connects your favorite chat apps … to AI coding agents” (OpenClaw docs). The official site says it is stewarded by the OpenClaw Foundation, an independent US non-profit, with no paid tier, hosted service or token.

Claude Code. Anthropic describes it as “an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools”, available in the terminal, IDEs, a desktop app and the browser (overview).

Install

OpenClaw, macOS, Linux and WSL2 (README, getting started):

curl -fsSL https://openclaw.ai/install.sh | bash

Windows PowerShell:

iwr -useb https://openclaw.ai/install.ps1 | iex

Or with npm (Node 24.16+ or 26.1+):

npm install -g openclaw@latest --allow-scripts=openclaw

Then:

openclaw onboard --install-daemon
openclaw gateway status
openclaw dashboard

The getting-started page says you need either an existing Claude Code or Codex CLI login or a provider API key, that the Gateway listens on port 18789, and that openclaw triage runs setup diagnostics.

Claude Code, macOS, Linux and WSL (advanced setup):

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell:

irm https://claude.ai/install.ps1 | iex

Then run claude in a project. Claude Code needs a Pro, Max, Team, Enterprise or Console account; the free Claude plan does not include it.

Models and providers

OpenClaw’s official provider plugins include OpenAI (OPENAI_API_KEY, default model openai/gpt-6-astra), Anthropic (ANTHROPIC_API_KEY, example anthropic/claude-opus-5-5), Google Gemini, OpenRouter and Ollama. Its local models page lists Ollama, LM Studio, vLLM/SGLang, llama.cpp and llmman.

For Anthropic, the OpenClaw provider docs describe two routes: an Anthropic Console API key (pay-as-you-go), or reusing an installed, logged-in Claude Code, where “OpenClaw communicates directly with the installed Claude Code executable” and usage draws from that subscription’s limits. OpenClaw recommends API-key auth “for shared automation or predictable production spend”.

Claude Code’s aliases are opus, sonnet, haiku, fable, best and others; on the Anthropic API, opus resolves to Opus 5.5 and sonnet to Sonnet 5.5 (model configuration).

Warning

Anthropic’s legal and compliance page says OAuth (subscription) authentication is “intended exclusively for purchasers of Claude Free, Pro, Max, Team, and Enterprise subscription plans” for ordinary use of Claude Code and native Anthropic apps, and that third-party developers may not route requests through Free, Pro or Max credentials on behalf of their users. Read that page before pointing any automation at a subscription login; a Console API key is the route both projects document for shared or automated use.

Plans and cost

OpenClaw has no plans; the site says there is “no subscription”, “no hosted tier” and “no token”. Your cost is whatever your model provider charges.

Claude Code is included from Claude Pro ($17/month with annual billing, $20 billed monthly) upward, through Max and Team seats, or billed per token through a Console API account (pricing, costs). See the plans tracker for every tier.

Skills and MCP

Sources: OpenClaw skills, OpenClaw MCP, Claude Code skills, Claude Code MCP. OpenClaw’s skills docs mark skills from skills.sh as “Not scanned by ClawHub”, so check where a skill comes from before installing it; the skills directory lists each collection’s source repository. OpenClaw also has an openclaw attach command that “launches Claude Code with a strict temporary MCP config bound to one Gateway session” (attach).

Security guidance

OpenClaw’s security guide defines “one trusted boundary per gateway: a single operator, or a team whose members trust each other” and says it is not built for adversarial multi-tenant use. Its documented defaults and checks:

• The Gateway binds to loopback by default; container images expose the bind and need auth set up.
• Unknown DM senders get a pairing code instead of being processed; approve with openclaw pairing approve.
• Group access is allowlisted behind mention gates.
• openclaw security audit reports drift from recommended settings.
• The README says to “treat inbound messages as untrusted input” and that tools run on the host unless sandboxing is configured.

Sandboxing is off by default and set per agent or globally with modes off, non-main and all, using Docker, SSH, Podman and other backends; the docs say it “is not a perfect security boundary, but it materially limits filesystem and process access” (sandboxing).

Warning

OpenClaw’s local-models page says “Local models do not provide hosted providers’ safety filters. Keep tool permissions and prompt-injection defenses appropriate for the model and task.” Because the agent reads messages from chat channels, enable a sandbox mode and keep DM pairing and allowlists on before connecting it to accounts other people can message.

Claude Code starts in default permission mode, which asks before edits and commands; other modes are acceptEdits, plan, auto, dontAsk and bypassPermissions. Its Bash sandbox is off by default and turned on with /sandbox (permission modes, sandboxing). Anthropic says to use bypassPermissions “only in isolated environments like containers, VMs, or dev containers without internet access”.

Which to pick

• If you need an assistant you message from WhatsApp, Telegram, iMessage or Slack, the docs show that is OpenClaw’s design; Claude Code is used from a terminal, IDE, desktop app or browser.
• If you need MIT-licensed code you can self-host and modify, the docs show OpenClaw is MIT; Claude Code is proprietary.
• If you need non-Claude or local models, the docs show OpenClaw provider plugins for OpenAI, Google, OpenRouter and local runtimes such as Ollama.
• If you need an agent focused on editing a codebase, with IDE diffs and GitHub integration, the docs show Claude Code’s IDE extensions, GitHub Actions and cloud sessions.
• If you need both, OpenClaw documents reusing a logged-in Claude Code and launching it with openclaw attach; check Anthropic’s authentication terms first.