Skip to main content

Agents

factorai runs Claude Code and Codex. Install either or both and log each in yourself.

What factorai knows about each​

Settings → Agents with the Claude Code card open: its detected binary and the Override path field

Settings → Agents has one card per agent: ACTIVE with its version, or NOT DETECTED. Expand it for:

  • Detected binary: the path factorai found.
  • Override path: use a different binary. Applies from the next session; leave empty to auto-detect.

Nothing found? See Troubleshooting.

Which agent a project runs​

The billing-api project's right-click menu with New session with open, offering Claude Code and Codex, Codex highlighted
  • Each profile belongs to one agent. A project runs its profile's agent, or the starred default's. Both are set in Settings → Profiles; with two or more profiles the project's right-click menu has a Profile submenu.
  • To start the other agent once: the chevron beside + on the project row, or New session with ▸ in the project's menu (only when both are installed).

What is the same, and what differs​

Claude CodeCodex
Transcripts read from~/.claude/projects/~/.codex/sessions/
Isolated byCLAUDE_CONFIG_DIRCODEX_HOME
Session titleyour /rename, else Claude's own, else the first promptCodex's own thread name, else the first prompt
Statusthe title glyphthe title's run state
New session idassigned by factorai before the process startsCodex's, adopted at the first turn
Add to agent contextyes, into the composeryes, as a new turn (codex queue), once the thread has had its first message
Viewer follows the files the agent readsyesnot yet

Tools for the agent​

Both agents get factorai's MCP tools at launch: they can list, create and update the project's routines, and switch one off or on, but not delete one.

Codex hides MCP tools behind its tool search, so name them: "use the factorai MCP server to list the routines".