DevinTrack

Track token usage and costs for Devin sessions. A transparent wrapper sits in front of the devin CLI, logs every request's token counts to a local SQLite database, and serves a web dashboard with per-model breakdowns, session details, and cost estimates.

How it works

Devin Desktop  →  devin (wrapper)  →  devin.real (original binary)
                      │
                      ▼
                 usage.db (SQLite)
                      │
                      ▼
              Dashboard server (Python)
                      │
                      ▼
              Web UI (React + Vite)

The wrapper intercepts devin acp calls, passes them through to the real binary, and parses the JSON-RPC responses for token usage data. It captures:

• Input, output, cached read, and cached write token counts
• Model name and reasoning effort level
• Session prompts and working directory
• Request duration and stop reason

All data is stored locally. No data leaves your machine.

Platform support

Platform	Status	Wrapper	Dashboard	Remote sync
Linux	Full	Yes	Yes	Yes
macOS	Full	Yes	Yes	Yes
Windows	Partial	Experimental	Yes	N/A (no SSH)

The dashboard server and frontend work on all platforms. The wrapper works on Linux and macOS. Windows support is experimental — see Windows notes below.

Quick start

1. Build the dashboard frontend

cd dashboard/web
npm install
npm run build

This produces dashboard/web/dist/ which the Python server serves statically.

2. Install the wrapper

Linux / macOS:

sudo python3 deploy.py

Or use the bash convenience script:

sudo ./deploy.sh

This backs up the original devin binary to devin.real and installs the wrapper in its place. Restart Devin Desktop after deploying.

To remove the wrapper and restore the original binary:

sudo python3 deploy.py --undeploy
# or
sudo ./undeploy.sh

Windows:

python deploy.py

See Windows notes for details.

3. Start the dashboard

python3 dashboard/server.py

Open http://127.0.0.1:7841 in your browser.

The server auto-reprocesses wrapper.log at startup to pick up any events that were missed while it wasn't running.

Configuration

All configuration is via environment variables. No config files needed.

Variable	Default	Description
DEVIN_TRACK_PORT	7841	Dashboard server port
DEVIN_TRACK_HOST	127.0.0.1	Dashboard bind address. Set to 0.0.0.0 to expose to the local network
DEVIN_TRACK_REAL	(auto-detected)	Path to the real devin binary
DEVIN_TRACK_SOURCE	local	Source tag for locally captured entries
DEVIN_BIN_DIR	(auto-detected)	Directory containing the Devin binary (deploy script)
XDG_DATA_HOME	~/.local/share	Base directory for data storage (Linux/macOS)
LOCALAPPDATA	(system default)	Base directory for data storage (Windows)

Remote sync (optional)

If you run Devin on a remote server, sync_remote.py pulls usage data from the remote to your local dashboard via SSH.

REMOTE_HOST=my-server.com \
REMOTE_USER=root \
REMOTE_KEY=~/.ssh/id_rsa \
python3 dashboard/sync_remote.py

Variable	Default	Description
REMOTE_HOST	(required)	Remote host to SSH into
REMOTE_USER	root	SSH user
REMOTE_KEY	~/.ssh/id_rsa	SSH private key path
REMOTE_DB_PATH	~/.local/share/devin_track/usage.db	Path to usage.db on remote
DEVIN_TRACK_SOURCE	remote	Source tag for synced entries
DEVIN_TRACK_PORT	7841	Local dashboard port

Run continuously with --loop (60s interval) or once with no flags. Use --dry to preview without inserting.

Custom source tags get unique colors in the dashboard automatically — no code changes needed.

Data storage

All data is stored in ~/.local/share/devin_track/ (or %LOCALAPPDATA%\devin_track on Windows):

File	Purpose
usage.db	SQLite database with usage and sessions tables
wrapper.log	Raw log of intercepted ACP events
reprocess_state.json	Byte offset for incremental log processing

Dashboard features

• Overview cards: Total tokens, estimated cost, request count, session count
• Usage timeseries: Token trends over time (7d / 30d / all-time)
• Model breakdown: Per-model token usage with reasoning effort badges
• Session list: Browsable sessions with model, duration, and cost
• Session detail: Per-request breakdown with prompts, tokens, and timing
• Recent requests: Latest individual requests across all sessions
• Filters: By model, working directory, and time period
• Cost estimates: Based on pricing scraped from docs.devin.ai (see pricing.json)
• Dynamic source badges: Custom source tags automatically get unique colors

Pricing

Cost estimates use per-token rates in dashboard/pricing.json. To update pricing:

python3 dashboard/scrape_pricing.py

This scrapes current pricing from docs.devin.ai and writes updated rates to pricing.json.

Requirements

• Python 3.10+ (stdlib only, no pip packages)
• Node.js 18+ and npm (for building the frontend)
• Devin Desktop

Windows notes

Windows support is experimental. The wrapper is a Python script, and Windows doesn't use shebang lines. The deploy.py script handles this by:

1. Backing up devin.exe to devin.exe.bak
2. Installing devin_wrapper.py alongside the original
3. Creating a devin.cmd shim that calls Python on the wrapper

You may need to set DEVIN_TRACK_REAL to point to the backup binary:

$env:DEVIN_TRACK_REAL = "C:\path\to\devin.exe.bak"

Project structure

devin                    # Wrapper script (Python, replaces the real devin binary)
deploy.py                # Cross-platform installer (Linux, macOS, Windows)
deploy.sh                # Unix convenience wrapper for deploy.py
undeploy.sh              # Unix convenience wrapper for deploy.py --undeploy
dashboard/
  server.py              # HTTP server + SQLite database + API
  sync_remote.py         # SSH-based remote data sync
  pricing.json           # Per-token pricing rates
  scrape_pricing.py      # Pricing scraper
  web/                   # React + Vite + Tailwind frontend
    src/
      App.tsx            # Main dashboard component
      api.ts             # API client
      components/        # UI components
      ui/                # Reusable UI primitives
.github/
  workflows/ci.yml       # CI: typecheck, build, cross-platform compile
  ISSUE_TEMPLATE/        # Bug report and feature request templates
  PULL_REQUEST_TEMPLATE.md

API endpoints

Method	Path	Description
GET	/api/overview	Summary statistics
GET	/api/timeseries	Token usage over time
GET	/api/models	Per-model breakdown
GET	/api/sessions	Session list
GET	/api/sessions/{id}	Session detail with requests
GET	/api/recent	Recent requests
GET	/api/performance	Performance metrics
GET	/api/cwd_list	Distinct working directories
POST	/api/ingest	Receive remote sync data
POST	/api/reprocess	Re-process wrapper.log
DELETE	/api/reset	Clear all data

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT