Agents & collaboration
Providers, models and thinking
Choose an agent and configure model, reasoning, speed and access for your task.
App version 0.7.934 · Updated
Get started
- Choose a provider or plain terminal in the new-session dialog.
- Connect it in Settings > Agent and check the detected program path for external agents.
- Choose an available model and the thinking level offered by that backend.
- Set Fast and work mode where available, then save your defaults.
- Use the model controls in the chat for later changes and check what your provider account supports.
Distinguish agent from model
IXIOM offers sessions with Claude Code, Codex, Antigravity, Kimi and Grok. External agents through ACP include OpenCode, GitHub Copilot CLI, Grok Build, DeepSeek Harness, Qwen Code, MiniMax Code and Cursor CLI. CLI-based providers need an installed and appropriately authorized program.
Terminal, Linux/WSL and, where supported, Hermes are separate terminal entry points. Model selection happens within those environments if offered. Older Gemini CLI sessions are migrated to Antigravity; a Gemini API connection is separate.
Choose supported options
Model controls follow the backend. Codex offers model-specific reasoning levels and a Fast service tier; Claude uses its available thinking levels; supported Kimi models offer thinking on or off. Antigravity exposes its CLI models and available effort and work modes.
Model catalogs can change. The choices in your installed IXIOM version and the access granted to your connected account determine what you can use. More reasoning can mean longer responses and more usage. Fast is a provider option, not a universal free speed boost.
New sessions, presets and team roles can inherit defaults or set their own values. Check the selected provider and the controls actually displayed rather than inferring the model from a preset’s name.
Local models and additional connections
Settings > Agent lists external agents and model providers separately. Assign a configured provider to compatible agents there. Profiles include Ollama Local, Ollama Cloud, DeepSeek API, MiniMax API and GLM Coding Plan.
Ollama Local requires an accessible local Ollama service. Enter its address and model; you can start a model download from settings. Cloud and API profiles need the appropriate access of your own. Not every agent supports every assignment.
The IXI assistant uses the ChatGPT/Codex connection and has its own model and thinking controls. Image Studio, voice and external services may need additional, separate connections.