Supported platforms
The coding agents and operating systems bmor supports, what it reads from each agent, and known gaps.
#Coding agents
bmor works with Claude Code, Codex, and Cursor. It reads the session logs each agent already keeps on your machine. For Cursor, that covers both the editor's agent and the Cursor CLI agent.
| Agent | Where it reads sessions | Also reads |
|---|---|---|
| Claude Code | ~/.claude/ | The model, cost, context window, and rate-limit data Claude Code passes to its status line |
| Codex | ~/.codex/ | Nothing else from Codex |
| Cursor | Editor agent transcripts in ~/.cursor/, and Cursor CLI sessions in ~/.cursor/ | The editor's local database, state.vscdb, for conversations and details the transcripts don't have. On macOS and WSL, see Known gaps. |
These are the default locations. To point bmor at other Claude Code or Codex session folders, or at another Cursor database, see Where bmor finds sessions.
Whichever agent a session comes from, bmor also reads your Claude Code account details from ~/.claude.json, if you have one, and adds them to its uploads. What Branchmore collects lists everything we collect.
#Operating systems
bmor is a single binary, built for each system below. Installation options lists which install methods work on each.
| System | Architectures | Status |
|---|---|---|
| macOS | Apple silicon (arm64), Intel (x86_64) | Supported |
| Linux | x86_64, arm64 | Supported |
| Windows, inside WSL | Same as Linux | Supported |
#Windows
On Windows, bmor runs inside WSL, on both x64 and ARM. Install it in the same WSL distro as your agents, and it works the same as on Linux, except for the Cursor editor: the editor runs on Windows even when you open a WSL folder, so its database stays on the Windows side. bmor only sees agents running in WSL, so it can't pick up sessions from agents running directly on Windows.
Running bmor directly on Windows isn't supported. If you'd like it to be, let us know.
#Known gaps
Things that don't yet work as well as the rest of these docs describe. The main ones are with Cursor. We'll update this section as things change.
#Cursor
Most of these come from Cursor keeping less on your machine than Claude Code and Codex do.
- Token counts and cost. Cursor doesn't save token usage locally, so most Cursor sessions arrive without token counts or cost. To add them, export your usage CSV from the Usage tab at cursor.com/settings, then upload it with Cursor Tokens on the Model and Token Usage page in the app. Upload a new export to cover later sessions; rows you've already uploaded aren't counted twice.
- Some reports leave Cursor out. The Long sessions and Parallel leverage reports need per-message token counts or timestamps that Branchmore doesn't get from Cursor, so they cover Claude Code and Codex only. Cursor sessions also have no git branch.
- The editor's database on macOS and WSL.
bmoronly looks for the Cursor editor's database,state.vscdb, where Cursor keeps it on Linux. On macOS until you set its path, and in WSL, where the editor keeps it on the Windows side, conversations that exist only in the database aren't uploaded, and other Cursor sessions miss details the database adds, such as the model. Even with the path set, conversations found only in the database can arrive without a project name, and they're skipped if you've set a directory allowlist or denylist.
To set the path on macOS, see collector.cursor.db_path.
#Request an agent
We can usually add an agent if it keeps a local record of its sessions that we can read, and every request moves an agent up the list. Tell us what you use on Discord, on the feedback page in the app, or through the contact form.