GuidesHosting choices

OpenCode vs OpenClaw: a coding agent and a chat assistant compared

OpenCode is a coding agent for your repository; OpenClaw is an assistant you message from chat apps. What each does, runs on and signs in with, and using both.

November 17, 2026The Everpod team
The short answer

They do different jobs and only the names are alike. OpenCode is an open-source coding agent: you open it in a code repository, in a terminal, its desktop app or a web page, and it reads, edits and runs the code there. OpenClaw is an open-source personal assistant: one always-on program, the Gateway, runs on your computer or a server, and you talk to it from WhatsApp, Telegram, Slack, Discord and other chat apps, where it remembers you and does things for you on a schedule. If someone told you about an AI that writes code with you, it was OpenCode; if it was an AI you text that runs your errands, it was OpenClaw. Both are MIT-licensed, both bring no model of their own, and they work together: OpenClaw can start OpenCode as a coding session from a chat.

Side by side

As of November 2026, with OpenCode 2.3.0 (released November 13) and OpenClaw 2026.11.5 (released November 17), from each project’s own docs:

OpenCodeOpenClaw
What it is
“An open source coding agent”, made by Anomaly“An open-source AI assistant that runs on your own hardware and meets you in every chat app you already use”
What it does
Works on the code in the folder you open it in: reads it, edits files, runs commandsKeeps a persistent assistant with its own memory and persona, answers you in chat, runs jobs on a schedule, from webhooks and hooks
Where you talk to it
A terminal interface, a desktop app, a web app and a phone appChat apps (Discord, Google Chat, iMessage, Matrix, Microsoft Teams, Signal, Slack, Telegram, WhatsApp, Zalo and more by plugin), a browser dashboard, and apps for macOS, Windows, iOS and Android
What it runs on
macOS, Windows or Linux, as one binarymacOS, Linux or Windows, on Node 24 or 26; the Gateway stays running as a background service
How you install it
curl -fsSL https://opencode.ai/v2/install | bash, or npm install -g @opencode/clicurl -fsSL https://openclaw.ai/install.sh | bash, which installs Node if needed and starts onboarding
Models
“200+ LLM providers”, local models includedMany providers, local models through Ollama, LM Studio, vLLM and others
Sign-ins instead of a key
ChatGPT Plus or Pro, GitHub Copilot, OpenCode Go; a Claude plan through your own installed Claude CodeChatGPT through Codex sign-in, GitHub Copilot, OpenCode Go; a Claude plan through your own installed Claude Code
What it remembers
A project’s rules in AGENTS.md at its root, written for you by /initA workspace at ~/.openclaw/workspace: persona in SOUL.md, long-term memory in MEMORY.md, a dated log for each day
Choose it when
The work is software in a repositoryYou want an assistant you message from your phone that keeps working while you are away

OpenCode has two published lines: version 2 (@opencode/cli, the one its home page installs) and version 1, still updated (1.19.6 on November 17). Much of what is written about OpenCode describes version 1; installing OpenCode on Linux covers which install line gives you which. OpenClaw publishes releases every few days (2026.11.4 and 2026.11.5 came out a day apart), so a tutorial a few months old may name commands that have since moved.

Why the two get mixed up

Beyond the names, both are free, open source, run on your own machine, and take a model from almost any provider, so a list of “open-source AI agents” puts them side by side. The difference is the centre of each. OpenCode’s centre is a code repository and you, at a screen, asking for changes; a session starts when you open it in a project. OpenClaw’s centre is the Gateway, a process that runs all the time whether or not anyone is looking, holds the conversations and the connections to your chat apps, and answers a message on Telegram at three in the morning. OpenClaw can write code, and OpenCode can be reached from a phone, but neither is built around the other’s job.

One overlap trips people up in setup: OpenCode is also the name of a model service. OpenClaw lists “OpenCode” and “OpenCode Go” among its model providers, meaning OpenCode’s hosted model catalogues, reached with an OpenCode API key (openclaw onboard --auth-choice opencode-go). Choosing that provider in OpenClaw’s onboarding does not install the OpenCode coding agent; it only buys the model from OpenCode’s service. Go is a paid subscription of its own, and its key is what both tools take.

Models and sign-ins

Neither project ships a model. Both take an API key from the big labs or a router, both run local models, and both accept a few subscriptions in place of a key. A ChatGPT Plus or Pro plan signs in to either. A Claude plan reaches both only the same way: through your own Claude Code, installed and signed in on the same machine. OpenCode’s Claude Agent provider needs Claude Code 2.1.248 or later and runs it underneath, and OpenClaw’s Anthropic provider can reuse “an existing Claude Code login through the installed executable on the same host”, with use drawing on the plan’s own limits. Whether that sits within Anthropic’s terms for a given use is Anthropic’s call, set out in your Claude subscription in Hermes, Pi or OpenCode. Which models OpenCode offers out of the box, and how to set a default, is in OpenCode models.

Running both

They do not compete for the same place, and where both are used the split follows the jobs: OpenClaw on an always-on server, where its Gateway can answer chat at any hour, and OpenCode wherever the code is, on a laptop or on the same server. On one machine they share nothing by default: separate installs, separate configuration, separate sign-ins, though a ChatGPT plan or an OpenCode Go subscription can serve both.

They also connect directly. OpenClaw’s ACP agents let it “run external coding harnesses”, and OpenCode is on its list beside Claude Code, Cursor and Gemini CLI: from a chat, OpenClaw starts a coding session with /acp spawn and hands the work to OpenCode, so you can ask for a code change from your phone and let the coding agent do it. The same pattern with Claude Code in OpenCode’s place is in OpenClaw vs Claude Code, and how OpenCode itself compares with Claude Code is in OpenCode vs Claude Code.

Run Claude Code, Codex, OpenCode, Pi, Hermes or OpenClaw on an always-on developer pod.

A developer pod is a permanent cloud computer (VPS), set up and secured for you, with your pick of Claude Code, Codex, OpenCode, Pi, Hermes and OpenClaw installed. Run more agents at once than your laptop can hold, and keep them working when it’s closed. From $24 a month, built in about ten minutes.