Agents & collaboration
IXI: the built-in assistant
Set up IXI, work with tools, and manage memory, schedules, scripts and diagnostics.
App version 0.7.934 · Updated
Get started
- Open Agent using the brain icon or its workspace view.
- Check ChatGPT/Codex access in Settings → Agent. IXI uses your existing Codex sign-in or a ChatGPT sign-in configured there; then choose an available model.
- Give the agent a concrete task and review its visible tool calls and result.
Chat and management
IXI has its own agent chat with tools and persistent memory. Use it as a workspace view, in the right-hand panel or in a separate window. Expand management when you need memory, schedules or diagnostics.
A request to IXI belongs to this assistant. Open a Claude or Codex session for project work; enable shared chat when you want those two agents to communicate directly.
ChatGPT access, model and persona
The main IXI agent runs through ChatGPT/Codex access. An OAuth sign-in explicitly configured in IXI takes priority; otherwise IXI uses the existing native Codex sign-in. An OpenAI API key alone does not replace this subscription access. Choose model, thinking level and fast mode from the offered options.
Persona and instructions shape the assistant's responses. Smart Helpers have their own provider and model selection for supported helper tasks. That selection does not change the main IXI agent. Custom model endpoints for other session backends are covered under Models. An IXIOM license does not include a ChatGPT subscription.
Inspect memory and tools
Open management to inspect the state IXI uses. A successful tool call and a completed reply are different events; when something fails, read the tool's actual result too.
- Memory
- Inspect and manage stored material in the memory browser, keeping earlier decisions available for later tasks.
- Cron and scripts
- Inspect recurring agent jobs and existing scripts in their respective management views.
- Health and search
- Open status and search diagnostics when connections, memory retrieval or background work behave unexpectedly.
Optional background features
Features settings configure autoDream, Proactive Agent and Coordinator separately. autoDream processes stored agent knowledge; Dream Log shows runs, processed memories and detected patterns.
Proactive Agent offers checks for Git conflicts, stale dependencies, long-idle sessions and workspace anomalies. Notification level and individual checks are configurable. Coordinator splits suitable tasks across Claude Code workers, with settings for worker limit, timeout and visible output. Use the separate shared-chat feature for Claude and Codex in one timeline.