npm install -g @crbonfree/cli

Get started

1. Install (one command, no dashboard needed)

crbon install

Shows a short code and opens your browser to approve it. It works the same on laptops, servers, VMs, and containers. No key to copy-paste. Once approved, it provisions an API key, sets up an hourly sync on your OS scheduler, and runs the first sync.

Prefer a browser handoff on this same machine? crbon install --loopback signs you in with the browser-callback flow instead: the browser returns the sign-in directly to the CLI over localhost.

2. That's it

Your coding-agent usage flows to your dashboard, where you see the CO2e behind every model and agent:

https://dashboard.crbonfree.com/my-coding-agents

Keep coding. crbon captures each session locally and syncs on its own.

Memory for your agents

crbon install asks whether to turn memory on; crbon memory on and crbon memory off change it later, and crbon status shows the answer. Once on, sessions are indexed locally under ~/.crbon/memory.

When memory turns on, choose which coding agents receive its MCP tools. Run crbon memory agents to change that choice later, or pass --agents with a comma-separated list (for example, --agents claude,codex), all, or none.

Memory content stays on the machine unless semantic search and summaries are turned on with crbon memory semantic on (paid plan). Turning them on sends the text of every indexed session to Crbon's endpoint to build the search index, and session content whenever a summary is generated; Crbon forwards it to an AI provider through Vercel's AI Gateway and does not store it. With a folder list, semantic search turns on once memory holds no sessions from outside the list; crbon memory tidy shows the ones it finds, and crbon memory tidy --yes moves them to the memory trash without touching the agents' own files. crbon memory semantic off stops it.

crbon memory sync on (paid plan) uploads the prompts and agent replies from sessions in your chosen folders, including their inline commands and file edits, so you can view and search them in the dashboard. Home paths become ~ and known secrets are scrubbed first. Crbon stores and can read synced sessions; they are not end-to-end encrypted. Use crbon memory sync off to stop future uploads while keeping the cloud copy, or crbon memory sync delete --yes to delete it. Add --session <id> to delete just one session.

Agents reach memory through crbon-mcp with three tools:

• search_content: find past sessions by text
• get_messages: read the messages around a hit
• get_memory_status: check whether memory is ready

crbon-mcp   # the memory server your agents connect to

Export one session as markdown (the default), JSON, or its original transcript. A session id comes from the memory tools' search results in your agent.

crbon memory export <session-id> --format json
crbon memory export <session-id> --format raw --out transcript.jsonl

What it does

| | | | |---|---|---| | Measure | Local agent token usage becomes CO2e on your dashboard | crbon sync | | Remember | Find past sessions by text with search_content, read messages around a hit with get_messages, and check readiness with get_memory_status | crbon-mcp | | Capture | Structured records: prompts, responses, tool calls, file diffs, models, tokens, git state | automatic | | Save tokens | Compressed output, bounded reads and searches, repeat-read caching, loop detection | Claude | | Protect | Redaction, sensitive-folder exclusion, .env blocking, secret scrubbing | always on |

Works with

Claude Code  ·  Codex  ·  Gemini CLI  ·  Cursor  ·  OpenClaw

How it works

your agents  ->  captured locally (~/.crbon)  ->  crbon sync (hourly)  ->  CrbonFree backend  ->  dashboard: CO2e
             ->  local memory (~/.crbon/memory)  ->  crbon-mcp  ->  your agents

Memory is indexed under ~/.crbon/memory and stays local unless you turn on semantic features or memory sync. Usage sync uploads structured session records for emissions. The backend computes emissions and powers the dashboard.

Commands

| Command | What it does | |---|---| | crbon install | One-command onboarding: sign-in, API key, scheduler, first sync | | crbon sync | Capture and upload new sessions now | | crbon schedule | Manage the OS scheduler (install / uninstall / status) | | crbon status | Show what's configured and what's synced | | crbon config | Manage configuration (set) | | crbon uninstall | Remove the scheduler, agent hooks, and API key before removing the package |

Installed alongside crbon and wired up automatically by crbon install: crbon-mcp (memory server), crbon-sync (historical backfill), crbon-{claude,codex,gemini,cursor,openclaw}-hooks (per-tool capture), and crbon-proxy (API capture proxy).

Scheduling

crbon install sets up an hourly sync on your OS scheduler automatically. To change the interval, run crbon schedule install --interval 30m (also 60m or 2h). crbon schedule status shows the scheduler state and the last sync outcome. crbon schedule uninstall removes the schedule.

What gets registered, per OS:

• macOS: a launchd agent named com.crbon.sync.
• Linux: a systemd user timer named crbon-sync.timer.
• Windows: a Task Scheduler task named crbon-sync.

Uninstall

1. Run crbon uninstall. This removes the OS schedule, the agent hooks (Claude Code, Gemini CLI, Codex, Cursor), and the OpenClaw plugin, and revokes the API key. Local data in ~/.crbon is kept; pass --purge to delete it too. Pass --keep-key to skip revoking the API key.

2. Remove the package:

   npm uninstall -g @crbonfree/cli

If you already removed the package without running crbon uninstall, remove the schedule manually:

• Windows: schtasks /Delete /TN "crbon-sync" /F
• macOS: launchctl bootout gui/$(id -u)/com.crbon.sync && rm ~/Library/LaunchAgents/com.crbon.sync.plist
• Linux: systemctl --user disable --now crbon-sync.timer && rm ~/.config/systemd/user/crbon-sync.{service,timer}

Configuration

State lives under ~/.crbon (config, session events, sync history, quarantined uploads). Resolution order is environment variable, then config file, then default.

| Variable | Overrides | |---|---| | CRBON_SERVER_URL | Backend base URL (default https://api.crbonfree.com) | | CRBON_API_KEY | API key (normally set by crbon install) | | CRBON_FRONTEND_URL | Dashboard base URL (default https://dashboard.crbonfree.com) |

crbon config set server-url https://api.crbonfree.com
crbon config set api-key <your-key>
crbon config set sync-folders <folder>,<folder>

Privacy

crbon is local-first. Sessions are captured on your machine and uploaded only when you sync. Built-in controls redact configured values, exclude sensitive folders, block unsafe .env reads, and scrub secrets from tool output before anything leaves your machine.

Only chosen folders count. On a machine that is also used for personal work, name the project folders that belong to your organisation:

crbon config set sync-folders ~/work/acme,~/work/acme-tools

Sessions outside those folders are never read in full, never uploaded and never counted. Memory follows the same list: only those folders are archived, and the memory tools show only their sessions. crbon install asks the same question right after sign-in; crbon install --sync-folders <list> or --all-folders answers it ahead of time. An empty list means everything counts, and crbon status shows the list.