At a glance
JetBrains' coding agent runs in the terminal, CI, JetBrains IDEs, and a tethered remote web UI with subscriptions, BYOK and local models, MCP, approval controls, Agent Skills, parallel sessions, Git worktrees, and specialized subagents.
Good choice if
- JetBrains users who want one agent across an IDE, terminal, and CI
- Large repositories that benefit from IDE diagnostics, planning, and repository intelligence
- Parallel workflows using multiple sessions, Git worktrees, and specialized subagents
Check before choosing
- Subagent model-policy controls are EAP, invocation is automatic rather than directly callable, and available models or aliases vary by environment
- Worktree isolation prevents file conflicts but is not an OS security boundary
- No built-in OS or container sandbox and no first-class file rollback mechanism are documented for the CLI
See 5 more considerations
- Brave On removes sensitive-action approvals; Auto relies on a command safety classifier rather than process isolation
- Project-trust enforcement for non-interactive JSON, ACP, and Gateway launches is currently behind a disabled rollout toggle, so those modes retain the previous trusted behavior instead of automatically entering restricted mode
- Only the last ten sessions retain full context, while guidelines and prompt history are explicit context rather than learned durable memory
- Remote mode is a synchronized UI over a running local CLI: the machine must remain awake, terminal-only actions stay unavailable in the browser, and JetBrains AI Enterprise authentication is unsupported
- The public GitHub repository distributes installers, registries, templates, and a small test surface rather than the proprietary agent implementation or eval suite