💸 agentcost

Monitor what your AI coding agents are spending. In real-time.

A lightweight local proxy that sits between your AI tools and the API providers, logging every token and dollar. Comes with a terminal dashboard (TUI) and a web dashboard — both update live as requests flow through.

Works with Claude Code, Cursor, Copilot, Continue, Aider, OpenAI API, Anthropic API, Google AI, DeepSeek — anything that calls an LLM API over HTTP.

  ╔══════════════════════════════════════════════════════════════════════════╗
  ║  agentcost — AI Agent Spend Monitor                                     ║
  ║                                                                         ║
  ║  Today: $3.4721 (142 reqs, 847.2K tokens)  |  All Time: $28.91 (1,204) ║
  ╠══════════════════════════════════════════════════════════════════════════╣
  ║  Today's Spend by Hour ($)        ║  By Provider                       ║
  ║  ▆                                ║  ╭──────────╮                      ║
  ║  █ ▃                              ║  │ ████░░░░ │ Anthropic  72%       ║
  ║  █ █ ▅                            ║  │ ██░░░░░░ │ OpenAI     21%       ║
  ║  █ █ █ ▂ ▁                        ║  │ █░░░░░░░ │ Google      7%       ║
  ║  █ █ █ █ █ ▁ · · · · · ·         ║  ╰──────────╯                      ║
  ║  09 10 11 12 13 14                ║                                    ║
  ╠═══════════════════════════════════╬════════════════════════════════════╣
  ║  Top Models Today                 ║  Live Request Log                  ║
  ║  ─────────────────────────────    ║  claude-sonnet-4 | 12.4K | $0.05  ║
  ║  claude-sonnet-4      89  $2.41   ║  gpt-4o-mini     | 1.2K  | $0.00  ║
  ║  gpt-4o               31  $0.82   ║  claude-sonnet-4 | 8.7K  | $0.03  ║
  ║  gemini-2.5-flash     18  $0.11   ║  gpt-4o          | 4.1K  | $0.02  ║
  ║  claude-3-5-haiku      4  $0.09   ║  claude-sonnet-4 | 15.2K | $0.07  ║
  ╠═══════════════════════════════════╩════════════════════════════════════╣
  ║  Last 30 Days ($)                                                      ║
  ║  ▁▂▃▂▄▅▆▅▃▅▆▇▆▅▆█▇▆▅▇▆▇█▇▆▅▆▇█                                     ║
  ╚════════════════════════════════════════════════════════════════════════╝

Why?

AI coding agents are brilliant but expensive. They make hundreds of API calls per session, and it's nearly impossible to know how much you're spending in real-time. Your provider dashboard updates with a delay, and doesn't break down by session, tool, or project.

agentcost gives you:

Real-time cost tracking as requests happen
Per-model, per-provider, per-session breakdowns
Budget alerts ("you've burned $10 today")
Historical trends (daily/hourly)
Zero config — works as a drop-in proxy

Features

Transparent proxy — intercepts and forwards API calls, logs usage
Multi-provider — OpenAI, Anthropic, Google AI, DeepSeek, any OpenAI-compatible API
Terminal dashboard — blessed-contrib TUI with charts, tables, live log
Web dashboard — browser UI with WebSocket real-time updates
Accurate pricing — built-in price table for all major models (June 2026)
Budget alerts — set a daily budget, get warned when you exceed it
Session tracking — group costs by session/conversation
SQLite storage — lightweight, no external database needed
Offline history — query past usage anytime
Zero dependencies on external services — everything runs locally

Quick Start

# Install
npx agentcost

# Or clone and run
git clone https://github.com/betty02/agentcost.git
cd agentcost
npm install
npm start

This starts:

Proxy on http://localhost:4000
Web dashboard on http://localhost:4001
Terminal dashboard in your current terminal

Setup with Your AI Tools

Point your AI tool at the proxy instead of the real API:

Claude Code / Anthropic SDK

export ANTHROPIC_BASE_URL=http://localhost:4000/anthropic
# Then use Claude Code normally

OpenAI SDK / GPT-based tools

export OPENAI_BASE_URL=http://localhost:4000/v1
# Or in your code:
# client = OpenAI(base_url="http://localhost:4000/v1")

Cursor

In Cursor settings → Models → OpenAI API Base URL:

http://localhost:4000/v1

Aider

aider --openai-api-base http://localhost:4000/v1

Continue (VS Code)

In .continue/config.json:

{
  "models": [{
    "apiBase": "http://localhost:4000/v1",
    "provider": "openai",
    "model": "gpt-4o"
  }]
}

Google AI / Gemini

export GOOGLE_AI_BASE_URL=http://localhost:4000/google

Any OpenAI-compatible tool

Just change the base URL to http://localhost:4000/v1. The proxy auto-detects the provider from the request path and headers.

Web Dashboard

Open http://localhost:4001 in your browser for a live dashboard showing:

Today's spend with request count and token usage
Hourly bar chart — see when your agents are most active
Model breakdown — which models are costing you most
30-day trend — daily spend over time
Live request log — every API call as it happens

The dashboard updates in real-time via WebSocket — no polling, no refresh needed.

Terminal Dashboard

If you prefer staying in the terminal, the built-in TUI shows the same data:

# Run TUI only (if proxy is already running separately)
npx agentcost --tui-only

# Run everything together
npx agentcost

Press q to quit the TUI.

Configuration

Config is stored at ~/.agentcost/config.json. Created automatically on first run.

{
  "proxyPort": 4000,
  "webPort": 4001,
  "dailyBudget": 10.00,
  "alertThreshold": 5.00,
  "pricing": {
    "gpt-4o": { "input": 2.50, "output": 10.00 },
    "claude-sonnet-4-20250514": { "input": 3.00, "output": 15.00 }
  }
}

Options

Key	Default	Description
`proxyPort`	4000	Port for the proxy server
`webPort`	4001	Port for the web dashboard
`dailyBudget`	null	Daily budget in USD (null = no limit)
`alertThreshold`	5.00	Warn in TUI when daily spend exceeds this
`pricing`	(built-in)	Override pricing per 1M tokens for any model

Custom Pricing

If a new model comes out or prices change, add it to config:

{
  "pricing": {
    "my-custom-model": { "input": 1.00, "output": 5.00 }
  }
}

Prices are per 1 million tokens.

CLI Options

Usage: agentcost [options]

Options:
  --proxy-only       Run only the proxy server
  --tui-only         Run only the terminal dashboard
  --web-only         Run only the web dashboard
  --all              Run proxy + TUI + web (default)
  -p, --port <port>  Proxy port (default: 4000)
  -w, --web-port     Web dashboard port (default: 4001)
  --no-tui           Run proxy + web without TUI
  -V, --version      Output version number
  -h, --help         Display help

How It Works

┌─────────────┐     ┌──────────────────┐     ┌──────────────────┐
│  Your AI    │────▶│   agentcost      │────▶│  OpenAI /        │
│  Tool       │◀────│   proxy (:4000)  │◀────│  Anthropic /     │
│  (Cursor,   │     │                  │     │  Google API      │
│   Claude,   │     │  ┌────────────┐  │     └──────────────────┘
│   Aider)    │     │  │ Log to DB  │  │
└─────────────┘     │  └────────────┘  │
                    │        │         │
                    └────────┼─────────┘
                             │
                    ┌────────▼─────────┐
                    │  SQLite DB       │
                    │  (~/.agentcost/  │
                    │   usage.db)      │
                    └────────┬─────────┘
                             │
              ┌──────────────┼──────────────┐
              │              │              │
     ┌────────▼───┐  ┌──────▼─────┐  ┌────▼──────┐
     │  TUI       │  │  Web       │  │  JSON API │
     │  Dashboard │  │  Dashboard │  │  /api/*   │
     │  (terminal)│  │  (:4001)   │  │           │
     └────────────┘  └────────────┘  └───────────┘

You point your AI tool's base URL at the agentcost proxy
The proxy forwards requests to the real API transparently
When the response comes back, it extracts token usage from the response body
It calculates cost using the built-in pricing table
Logs everything to a local SQLite database
The TUI and web dashboards query the DB and show real-time stats

Your API keys stay on your machine. The proxy just passes headers through — it doesn't store or inspect your API key.

API

The web server exposes a JSON API you can query programmatically:

Endpoint	Description
`GET /api/stats`	All-time totals
`GET /api/today`	Today's spend, tokens, request count
`GET /api/hourly`	Hourly breakdown for today
`GET /api/models`	Top models by cost today
`GET /api/providers`	Cost by provider today
`GET /api/recent?limit=50`	Recent requests
`GET /api/daily?days=30`	Daily totals over N days
`GET /api/sessions`	Session summaries

Supported Providers & Models

Pricing included for:

Provider	Models
OpenAI	gpt-4o, gpt-4o-mini, o1, o1-mini, o3, o3-mini, gpt-4-turbo, gpt-3.5-turbo
Anthropic	claude-opus-4, claude-sonnet-4, claude-3.5-sonnet, claude-3.5-haiku, claude-3-opus, claude-3-haiku
Google	gemini-2.5-pro, gemini-2.5-flash, gemini-2.0-flash, gemini-1.5-pro
DeepSeek	deepseek-chat, deepseek-coder

Unknown models default to a mid-range estimate. Add custom pricing in config for accuracy.

Data Storage

All data is stored locally in ~/.agentcost/usage.db (SQLite). No data leaves your machine.

# View raw data
sqlite3 ~/.agentcost/usage.db "SELECT * FROM requests ORDER BY id DESC LIMIT 10"

# Export to CSV
sqlite3 -header -csv ~/.agentcost/usage.db "SELECT * FROM requests" > export.csv

# Reset all data
rm ~/.agentcost/usage.db

Troubleshooting

Problem	Solution
Tool ignores base URL	Some tools only respect env vars on fresh start — restart the tool
SSL/TLS errors	The proxy speaks HTTP. Tools expecting HTTPS need `HTTPS=false` or similar
Port in use	Change port: `agentcost -p 5000 -w 5001`
No usage in response	Some streaming responses don't include usage — we log the request but cost shows $0
Wrong cost shown	Update pricing in `~/.agentcost/config.json`
TUI looks broken	Ensure terminal is at least 120×30 chars. Try `--no-tui` and use web dashboard

Contributing

PRs welcome! Ideas:

Docker container version
Grafana/Prometheus export
Per-project tracking (based on working directory)
Cost predictions ("at this rate you'll spend $X this month")
Slack/Discord webhook alerts
Browser extension for web-based AI tools
Support for streaming response token counting
Multi-user support with separate API keys

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
src		src
web/public		web/public
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
package.json		package.json

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

💸 agentcost

Why?

Features

Quick Start

Setup with Your AI Tools

Claude Code / Anthropic SDK

OpenAI SDK / GPT-based tools

Cursor

Aider

Continue (VS Code)

Google AI / Gemini

Any OpenAI-compatible tool

Web Dashboard

Terminal Dashboard

Configuration

Options

Custom Pricing

CLI Options

How It Works

API

Supported Providers & Models

Pricing included for:

Data Storage

Troubleshooting

Contributing

License

About

Releases

Packages

Contributors

Languages

Folders and files

Latest commit

History

Repository files navigation

💸 agentcost

Why?

Features

Quick Start

Setup with Your AI Tools

Claude Code / Anthropic SDK

OpenAI SDK / GPT-based tools

Cursor

Aider

Continue (VS Code)

Google AI / Gemini

Any OpenAI-compatible tool

Web Dashboard

Terminal Dashboard

Configuration

Options

Custom Pricing

CLI Options

How It Works

API

Supported Providers & Models

Pricing included for:

Data Storage

Troubleshooting

Contributing

License

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages