You do not have to choose. OpenCode is a capability here.
OpenCode is an open-source agent that speaks the Agent Client Protocol, so it becomes one more provider in the chat picker: its own models, its own tools, its own config.
“The open source AI coding agent”: a terminal, desktop and IDE agent with LSP integration, parallel sessions and 75+ model providers.
OpenCode is a harness, in the same category as Claude Code and Codex: it holds the agent loop and talks to whichever model you point it at. intentic is the environment a harness runs in.
So the honest answer is not a comparison. Install the OpenCode capability and “opencode acp” becomes a provider row alongside Claude Code and Codex, inside the sandbox.
Where we agree
Both are open source and mean it: OpenCode's agent, and the intentic sandbox and CLI, MIT on GitHub. Both run on your own accounts, and neither sits between you and your model provider.
Where they differ
A harness, not a workplace
OpenCode gives you the agent: the loop, the LSP integration, the model routing. It does not give the agent a machine of its own, credentials, or an event that wakes it.
What ACP buys you here
Any ACP agent becomes a chat provider in a sandbox: OpenCode and Gemini CLI have preset cards, and the custom card takes any command from the registry. Tool calls and inline diffs stream into the same chat surface.
Model freedom, from two directions
OpenCode's answer to lock-in is 75+ providers behind one agent. intentic's is five native harnesses plus every ACP agent as a capability. Installed together you get both.
Head to head
| Dimension | intentic | OpenCode |
|---|---|---|
| What it is | the environment an agent works in | the agent itself |
| Can run the other | runs OpenCode as an ACP provider | not applicable |
| Model providers | 5 native harnesses, plus any ACP agent as a capability | ▸ 75+ providers behind one agent |
| Where it runs | a Docker sandbox on your hardware | your machine: terminal, desktop or IDE |
| Interface | a browser workspace over a private tunnel | TUI, desktop app, IDE extension |
| Reviewing changes | a diff reader over an isolated worktree, then land or discard | git, and your editor's diff |
| Credentials for your systems | capabilities, stored sandbox-side, injected per turn | your shell environment |
| Starts on an event | automations: cron, webhook, push, alert, email, chat | you start it |
| Licence | MIT sandbox and CLI; the hosted platform is closed | ▸ open source throughout |
How to add it
Capabilities → OpenCode (ACP). The card pre-fills the command, you supply the credentials its providers need, and it appears in the picker next turn. Same binary, running in the sandbox instead of your shell.
Pick OpenCode when
You want one open-source agent you can read end to end, pointed at any of seventy-five providers, and you need no sandbox, capabilities, fleet or browser around it.
Every claim above is checkable on opencode.ai and in this repo. Where we have it wrong, open an issue and it gets fixed in the next build.