Get started with Centrail
Track AI coding spend: create a free account, connect your device, sync agent usage, and understand your first numbers before bringing a team.
From a free account to useful numbers
Centrail is agent economics for engineering teams: a record of captured AI coding usage, estimated cost, and the Git work associated with that usage. Start with your own device before deciding whether to bring a team.
1. Create your free account
Open Centrail signup and create an account. Verify your email, then sign in. Personal use is free; no card is required to begin. Free includes one active device and full history. See current pricing for paid personal and team plans.
2. Connect the device where you use coding agents
You need Node.js 20 or newer and local logs from Claude Code, Codex, or GitHub Copilot CLI. Other Copilot surfaces are not interchangeable with Copilot CLI logs.
Run this in your terminal:
npx centrail@latest connect
Follow the browser pairing instructions and approve the device in your Centrail account. Centrail reads supported local agent logs; it does not require a proxy, gateway, or SDK in your application.
Review which repositories and folders may sync when the CLI asks. You can revisit this with npx centrail@latest setup, list the current selection with npx centrail@latest repos, and exclude a repository with npx centrail@latest exclude <repo>. Device pairing is not permission to disclose every repository to a team; team sharing is a separate choice.
Capture is local-first, not local-only. Usage and Git metadata are sent to your hosted account. Prompts, completions, source code, and diffs are not uploaded as usage data. Repository names, commit identifiers, counts, and other metadata can still be sensitive. Read the privacy policy before connecting a work device.
3. Sync your available usage
After pairing, run:
npx centrail@latest sync
Open your dashboard and review the captured usage. Repeat sync after more agent work to refresh it. Existing local logs determine how much history can be captured; connecting a device cannot recover logs that no longer exist.
To review exactly what was sent, run npx centrail@latest inspect --last. It prints the last payload locally; do not post it publicly without reviewing its metadata.
Optional: sync after each agent turn
First run a manual sync and check its result. Then choose automation explicitly:
- For Claude Code,
npx centrail@latest setup-plugininstalls its Centrail plugin; the current CLI also offers this during connect. - Without the plugin,
npx centrail@latest install-hooksinstalls a Stop hook for Codex and Claude Code. A Stop hook runs when an agent turn ends; it is not an application SDK or a proxy. npx centrail@latest uninstall-hooksremoves the hooks installed by that command. Plugin management is separate.
Hooks can capture session/repository context and trigger sync. They do not guarantee that every session will have a matching commit. Do not install both automation paths for the same agent without reviewing the CLI's existing-plugin detection, and check with your organization before changing agent settings.
How commit attribution gets set up
Use your coding agent in a Git repository and commit the work normally. After you sync, the CLI sends allowed repository identity and commit facts; the current server uses that evidence to match captured usage to commits. You do not need to tag each commit manually or install the GitHub App to begin local Git attribution.
Matching depends on available session, branch, repository, and timing evidence. Missing or excluded context can leave usage unattributed. GitHub Copilot CLI usage capture does not imply the same attribution coverage as Claude Code or Codex.
4. Understand your first numbers
Your first useful result is captured usage you recognize, with an estimated cost and repository context where available. A populated commit chart is not required for setup to have worked.
| Number or view | What it means | What it does not prove |
|---|---|---|
| Tokens | Usage reported in the supported logs that were captured | Complete provider or organization-wide usage |
| Estimated cost | Captured usage priced using model rates | Your invoice, subscription cost, or final bill |
| Cost per Commit / Tokens per Commit | Usage allocated to commits by attribution logic | That an agent produced the commit, or the complete cost of a feature |
| Model Mix | Which models appear in captured usage | Which model produces the best work |
| Unattributed usage | Usage not allocated to a commit | That the work was worthless |
| Git work status | Evidence about work in repository history | Production deployment, customer value, or revenue |
| GreenAI footprint | Modeled energy, water, and CO₂ estimates | Metered data-center emissions |
Choose a date range that includes known agent work. Inspect a repository you recognize and compare its recorded usage with your local activity. Do not divide all account spend by all commits and treat the result as verified feature cost.
5. Bring a team, or request a demo
Once you understand your own numbers, compare the paid Team plan. Team members connect their own devices and choose which repositories to disclose, including whether to share earlier history. Private work outside that selection should remain outside the team view.
The optional GitHub App can supply independent evidence of eligible pushes and merged pull requests. It does not supply token usage, prove agent causation, or grant the team access to members' usage by itself. See how team numbers work.
For a founder demo, contact support@centrail.org and ask for a demo with Brayden. Include the tools your team uses, approximate team size, and the question you want answered. You do not need to send private logs or source code.
If the dashboard is empty or incomplete
- Confirm you paired the same account you are viewing, then run sync again and read its summary.
- Check that the supported agent was used on this device and that its logs still exist locally.
- Widen the dashboard date range. New work and old retained logs may fall outside the selected dates.
- Missing commit attribution is different from missing usage. A session without usable repository context may remain unattributed.
- For a team view, check membership and the repository sharing selection. Personal usage is not automatically team usage.
- If you need help, use support. Share the command, error text, tool/version, and operating system, with private identifiers redacted. Never send credentials, pairing secrets, or raw private logs.
What comes next?
For quick answers about supported tools, team sharing, privacy, and what the numbers mean, read the frequently asked questions. This walkthrough covers the first-use path; the linked reference docs cover the details.
Use the numbers to ask better questions about AI coding spend by developer and repository. They support budget conversations; they are not a forecast or a measure of business value. For how estimates are calculated, read the environmental methodology and team-number definitions.