OpenCode and DeepSeek Harness
Two coding agents that run on the model providers you choose, including models on your own machines.
Add either with Set up AI… in the Ask chip, or Add an agent… in the chat's model picker. Circuitry installs it.
OpenCode
Desktop appYour own AI
OpenCode signs in to the model providers you already use, or takes their API keys.
- Add OpenCode. It opens in a terminal.
- The first time, type
/connectand pick your provider. - Local models you have set up in Circuitry appear in its model list too.
Chats keep their place: come back to a chat and it continues the same OpenCode session.
A button in the chat header switches between OpenCode's terminal and its own web app, on the same conversation.
To choose the model in the chat view, use the model chip under the composer, which lists what OpenCode can run for you. Your choice is remembered for your next OpenCode chat on this device. With nothing chosen, it uses the model you last picked inside OpenCode.
DeepSeek Harness
Desktop appYour own AI
DeepSeek Harness runs on DeepSeek's models or your own model server, such as Ollama or LM Studio. It has no terminal: it opens its own app in the chat pane.
- Add DeepSeek Harness. Its app opens here.
- The first time, give it a model in its Settings → Models: a DeepSeek key, a DeepSeek sign-in, or your own model server.
Circuitry's tools come with every run, so there is nothing to connect.
To run a chat on another model, use the model chip under the composer. It lists your default, DeepSeek's model and your local models.