Cost control for AI coding agents.In view and in budget.

Passcard reads token usage from the agents on every device, attributes each dollar to the person, project and client behind it, and stops supported actions when a budget is reached.

  • Free plan, no card required
  • No proxy in the request path
  • SSO and SCIM on Business

Works with the agents your engineers already use

Local collectors and budget hooks for eight coding agents.

Claude Code
Codex CLI
Cursor
Gemini CLI
Grok CLI
Copilot CLI
OpenCode
pi

Why Passcard

From one invoice total to clear answers.

Coding agents run on every laptop. Without a shared view, their cost shows up as a single number at the end of the month.

Before

Without Passcard

  1. Spend stays in local logs

    Each agent writes token counts on the machine it runs on. Nobody sees them in one place.

  2. Loops run unnoticed

    An agent can retry the same task for hours. Nobody gets a warning while the spend climbs.

  3. One total on the invoice

    The provider bill cannot say which person, project or client caused the cost.

  4. Keys pasted into prompts

    API keys end up in agent chats, and nobody notices until they leak.

Finance sees one total. Engineering sees nothing until it is spent.

After

With Passcard

  1. One view across devices

    The desktop app reads those logs on every paired device and syncs usage to one workspace.

  2. Budgets stop the next action

    Installed hooks check supported actions against personal, project and team limits and warn or block.

  3. Every dollar attributed

    Costs land on the person, project and client behind them, with session details to back them up.

  4. Exposures caught on the device

    Passcard flags likely keys locally. Only a masked finding leaves the machine.

Every dollar has an owner, and budgets act before the next action.

Capabilities

Visibility, attribution and control. In one workspace.

For engineering teams that run coding agents every day, and for freelancers who bill that work to clients.

Usage across every agent

Token usage from eight local coding agents, priced at published API rates and updated as logs change.

Attribution to projects and clients

Repository remotes connect sessions across clones and devices. Projects roll up to clients.

Budgets with local enforcement

Daily, weekly, monthly or total limits per workspace, team, person or project. Hooks warn or block before supported actions.

Reporting by person and team

Compare usage across people and teams, with session details for model, timing and estimates.

Secret detection on the device

Likely API keys in agent chats are detected locally. Only masked findings are uploaded.

Governance for larger organizations

Single sign-on, SCIM provisioning, audit export and a read API on Business.

Ready for your security review.

Passcard tracks cost from usage metadata. Your security team can see exactly what leaves a machine and who can access it.

Data on the device

Metadata, not content.

Usage tracking does not upload prompt text, tool argument payloads, file contents, absolute file paths or environment variable values.

  • Token counts, models, agents and timestamps
  • Repository and branch names, relative file paths
  • Secrets detected locally, uploaded masked
  • A revocable token for every paired device
Rollout and enforcement

Fits how you run engineering.

Pair each machine with a member, map repositories to projects, and decide how budgets behave when data is missing.

  • Desktop apps for macOS, Windows and Linux
  • Hook install snippets for central rollout on Team and Business
  • Fail-open or fail-closed budget policy
  • Cached budgets keep working offline
Security and governance

Controls your administrators expect.

Roles separate administration, billing and everyday use. Business adds identity provisioning and audit access for larger organizations.

  • Single sign-on
  • SCIM provisioning
  • Audit export
  • Read API
  • Admin, member and billing roles
  • Device token revocation

Good to know

Questions from engineering and finance.

How usage is measured, which metadata is uploaded, and what budgets can and cannot stop.

Usage and session metadata: token counts, model and agent names, repo and branch names, timing, identifiers, tool counts, relative filenames and device details. Prompt text, tool argument payloads, file contents and absolute paths are excluded. The optional Ticket Beta adds Git evidence and selected provider metadata.

Free to start

Put your AI spend in focus.

Start free with one device and three projects. No card required.