Supported Agents
Argus finds and indexes the sessions from your AI agents on your computer, and includes every one of them by default.
| Agent | Description |
|---|---|
| Claude Cowork | Claude's agent for knowledge work in the Claude desktop app |
| Claude Chat | Chat conversations on claude.ai (via desktop, see below) |
| Claude Code | Claude's coding agent for the terminal, IDE and desktop app |
| ChatGPT Work | OpenAI's agent for longer, multi-step knowledge work |
| Codex | OpenAI's agent for software development and technical work |
| Gemini CLI | Google's coding agent that runs in your terminal |
ChatGPT Work and Codex are two experiences inside ChatGPT rather than two names for the same thing. They write their sessions to the same place, so Argus reads both together.
Locations
Argus automatically indexes sessions from every agent it finds on your computer without any additional configuration. By default, Argus uses these locations to find your existing agent sessions:
| Agent | macOS | Windows |
|---|---|---|
| Claude Cowork | ~/Library/Application Support/Claude/local-agent-mode-sessions | not yet supported |
| Claude Chat | ~/Library/Application Support/Claude/Cache/Cache_Data | %APPDATA%\Claude\Cache\Cache_Data |
| Claude Code | ~/.claude/projects | %USERPROFILE%\.claude\projects |
| ChatGPT Work and Codex | ~/.codex/sessions | %USERPROFILE%\.codex\sessions |
| Gemini CLI | ~/.gemini | %USERPROFILE%\.gemini |
If you've moved an agent's data with its own setting (for example CLAUDE_CONFIG_DIR, CODEX_HOME or GEMINI_CLI_HOME), Argus follows that same setting. For where Argus keeps its own data, see the CLI Reference.
Claude Chat sessions require the desktop app
Argus finds Claude Chat in the Claude desktop app, which keeps a local copy of your conversations. It can't include chats that only happened on claude.ai and were never synced to the desktop app, so two limits apply:
- You need the desktop app installed and signed in, and only conversations that have synced to it are included.
- Claude Chat usage is estimated, not exact, because Claude Chat doesn't report token counts the way the command-line agents do.