AI provider plugins let Singularity drive seven agent CLIs from one kanban: Claude Code, Codex, GitHub Copilot, Cursor, Gemini, Mistral and Ollama. Singularity prepares the ticket, opens a local process and drives the selected CLI. You keep your accounts and choose the right tool for each task.
Which providers are supported?
| Provider | Command |
|---|---|
| Claude Code | claude |
| Codex | codex |
| GitHub Copilot | copilot |
| Cursor | agent |
| Gemini CLI | gemini |
| Mistral Vibe | vibe |
| Ollama | ollama |
Each provider has its own settings tab: permissions, effort or sandbox depending on what the CLI exposes. Ollama runs through Claude Code against a local server, at http://127.0.0.1:11434 by default.
How do you configure a provider?
- Install the CLI using its publisher’s recommended method.
- Run it once in a terminal and finish its authentication.
- Ask it for a reply from a project folder: resolve anything blocking before starting a ticket.
- Open Singularity settings and adjust the provider’s tab.
How do you choose the right model day to day?
The provider setting defines the workspace’s default model. Auto lets prequalification pick an eligible model for the task. On a ticket, the agent selector overrides this choice for that run: a powerful model for a migration, a fast one for a mechanical fix. You can also exclude models from automatic selection.
Where does my data go?
Singularity does not upload your code to an AI provider by itself. The CLIs you configured talk to their services under their own terms: prompts, file excerpts and tool results may pass through them. With Ollama on the local address and a local model, inference stays on the machine; the initial model download still contacts the Ollama registry.
What are the limits?
The plugins are free, but the services may not be: your account or subscription remains the provider’s, and no per-ticket cost is guaranteed. A small local model follows long instructions and complex refactorings less well, and can become unusable without enough memory. Keep a remote provider for difficult tasks.
Frequently asked questions
Do I need an API key for each provider?
Not necessarily. Most CLIs use their own interactive sign-in. Cursor and Mistral also offer a masked field for a key or profile.
Can I use local models?
Yes, with Ollama. Download the model before selecting it. The tab recommends models based on RAM and processor.
Can I mix several providers?
Yes. A ticket’s agent selector overrides the workspace default for that run without changing it.
AI providers guide · Singularity vs Claude Code · Download Singularity