AI cost visibility for engineering leaders

You don't know where your AI spend goes.
Costpilot does.

Costpilot pulls usage and cost from every AI tool your team uses and pins each dollar to the issue that caused it. So you always know what a feature really costs, not just what a model did.

See the demoGet startedTalk to us

Cost by epic

$5,897 in scope · last 30 days

  • AI assistant in product$1,31622%
  • Self-serve onboarding$904.615%
  • Platform hardening$805.614%
  • Data platform migration$789.413%
  • Search & discovery v2$717.112%
  • Mobile app 2.0$390.86.6%
  • SOC 2 & enterprise compliance$355.86.0%

Live demo - Try Costpilot right here. Real data, no signup.

This is the actual Costpilot dashboard running on sample data. Click through epics, issues, developers and models, switch the date range and see how AI spend rolls up per feature.

demo.getcostpilot.io
Open full screen

Sample data from a fictional team. Your own deployment shows your telemetry, self-hosted in one click.

Why the invoice isn't enough

AI costs are climbing. The invoice won't tell you why.

Today, visibility ends at the provider's total. Costpilot goes one step further and ties every token you burn to the issue that burned it. Stop asking what a model costs. Start answering what a feature costs.

  • Consolidate. Every provider, every model, one dashboard. No more stitching invoices together just to see your total AI spend.
  • Attribute. Every dollar carries the name of its feature, instead of vanishing as an anonymous line in a usage log.
  • Report. A per-feature cost view leadership can actually act on. Not a token counter.

Same spend. Re-sorted by the value it created.

Cost by issue

$5,897 in scope · last 30 days

  • Auto-summaries for reports$490.18.3%
  • Observability: OpenTelemetry rollout$488.48.3%
  • Eval harness & guardrails$456.67.7%
  • Invite teammates flow$424.77.2%
  • dbt model consolidation$403.76.9%
  • App Store / Play release automation$385.56.5%
  • Conversational support assistant$369.56.3%

Roadmap - What works today, what we are building, where you can jump in.

Costpilot is built in the open on GitHub. The roadmap lives in the issues; this is the short version.

Live

Cost per feature, four coding agents

AI spend rolled up per epic, issue, developer and model. Claude Code, Gemini CLI, Codex CLI and Cursor feed in through one hook. Self-hosted as a single service with SQLite.

See the core repo
In progress

Spot savings per feature

A large share of tokens goes to models where a smaller one delivers the same result. Costpilot will show, per feature, where a switch saves money without costing quality.

Follow on GitHub
Open for contributions

Cost estimates before you build

Effort gets estimated before a feature is built, AI cost does not. Usage data from similar issues can give a range up front. Want to build this with us? The issue is open.

Pick up the issue
Suggest a featureAll open issues

Get started - Self-hosted in minutes. One service, your data.

Costpilot runs as a single service with SQLite on a volume. Deploy it on Railway, Docker or Fly.io, install the agent on each developer machine, and every AI coding session is attributed to its issue automatically.

1

Run the Costpilot server

Clone the core repo and start it locally with a seeded demo dataset, or deploy it with one click on Railway. The API serves the dashboard itself.

git clone https://github.com/getcostpilot/core
cd core && npm install
npm run seed && npm run dev
getcostpilot/core on GitHub
2

Install the agent

One command on each developer machine turns on telemetry export for Claude Code, Gemini CLI, Codex CLI and Cursor, and installs the hook that maps every session to the issue on your git branch.

npm install -g git+https://github.com/getcostpilot/agent.git
costpilot-agent init --endpoint https://costpilot.example.com
getcostpilot/agent on GitHub
3

Work on a branch

Check out a branch that carries an issue id, such as feature/PROJ-123-login, and use your coding agent as usual. Cost for every prompt shows up under PROJ-123, per epic, developer and model.

Maintainers - Built by people who wanted to know what a feature costs.

Costpilot came out of a simple gap: developer time is estimated and reported, AI cost is not. We maintain the repos, review pull requests and keep the demo running. Contributions are welcome.

Patrick Roos

Patrick Roos

@bitsmuggler

Tech lead and software architect. Builds the server, the dashboard and the agent, and makes sure cost per feature stays a number leadership can act on.

Florian Bosshart

Florian Bosshart

Enterprise architect and product owner. Brings the business-case view: what a feature is worth, what it costs, and how that becomes a team's priorities.

Get involved

Costpilot grows with the people who use it.

Contact - Talk to us

Want a demo with your own numbers, or just have a question? Tell us a bit about your setup and we'll get back to you.