Set up your AI tools
Connect one provider, understand model discovery, and choose optional local execution modes.
On this page
Install and connect one tool
- Install Git and your chosen AI command-line tool on the connected computer. Use its official installer and sign-in instructions.
- Run that tool directly in a terminal once and finish its login/provider setup.
- In Multicode AI, open AI providers, Set up AI, or Providers inside a chat.
- Select Check connections. If a detected tool offers Connect, follow its login flow; you can also complete the CLI login in your terminal.
- Return to the room and choose its provider and model in the composer.
You only need one working tool. The app displays installation commands for missing tools and status for installed tools. It does not install every provider or purchase AI credits for you.
Available execution options
| Option | Required setup | Behavior in this app |
|---|---|---|
| Claude Code | Installed CLI and provider sign-in | Coding agent; available aliases or discovered models |
| Codex | Installed CLI and provider sign-in | Coding agent; models and reasoning efforts discovered from the CLI |
| Gemini | Installed CLI and its login/configuration | Coding agent; default or discovered models |
| OpenCode | Installed CLI and configured provider | Tool-capable text models from its provider configuration |
| Ollama | Running local server and an explicitly downloaded model | Chat-only option |
| Codex + Ollama | Codex plus a compatible model on the default Ollama server | Local coding through Codex’s OSS backend |
| Codex + LM Studio | Codex plus a compatible model served by LM Studio | Local coding through Codex’s OSS backend |
The last three are optional advanced choices. Model discovery does not download weights. Local coding quality and compatibility depend on the model, tool support, and available RAM/VRAM.
Use the live provider screen for the options actually available on your computer. A local client can still call a cloud service depending on its configuration.
Models, refreshes, and unavailable selections
Open the searchable Model picker to see the connected tool’s current inventory. The picker can show its source, refresh time, and stale or error state. Use its refresh control after installing or updating a provider.
Availability depends on the installed tool and your account. If a selected model disappears, choose a currently available model explicitly. The app does not silently substitute a different or more expensive model.
Reasoning effort choices come from the selected model. Fast uses low effort only when supported. See composer controls.
Provider help and account limits
Official setup references: Claude Code, Codex, Gemini CLI, OpenCode, Ollama, and LM Studio.
Your provider controls billing and limits. A quota, expired login, or unavailable model error must be resolved with that provider or its local configuration. Refresh connections afterward. Never paste API keys or login tokens into shared chat.