foldrun.

Pricing

Pay for what runs. Not for who watches.

A flat base per account, unlimited seats, and three meters that count what your agents actually used. Free wherever you run it yourself.

Local

Free

forever, on your machine

The open-source CLI against your own model key. Every part of the format, every check, every run — no account and no server.

  • Agents, flows, tools, skills, evals
  • foldrun check, run, eval
  • Claude Code subagents deploy as-is
  • Apache-2.0
npm install -g foldrun

Cloud

Early access

a flat monthly base, plus what runs

The same folder, running without you: schedules, sandboxes, gates, vaulted secrets, budgets, a team. Unlimited seats — agents work, people supervise, and we do not tax the supervisor.

  • One base fee per account, seats unlimited
  • Usage metered: tokens, compute, storage
  • Bring your own key: tokens cost you nothing
  • Git deploys, previews, approvals, notifications
Request early access

Self-hosted

Free

your hardware, your data

The whole platform on one box with docker compose, or on your cluster with the published manifest. gVisor isolation, the same dashboard, the same everything.

  • One container, or a Kubernetes manifest
  • Your object store, your Postgres
  • Metering optional — the caps work without it
  • Nightly backups you can restore
Self-hosting guide

What is metered

Tokens

Model usage at cost, times a margin. Bring your own key and the margin does not apply — a BYOK step is billed nothing for tokens.

Compute

Sandbox seconds: what a step reserved (cores and memory, while it ran) and what it actually burned. No per-step fee — steps are sandbox-seconds, and a step fee would charge compute twice.

Storage and network

GB-months in the file store, accrued daily; GB the sandbox moved, both directions.

Every run records what it cost, per step. The Usage page shows spend by workspace and by meter. How budgets and billing work →

Questions

Why no per-seat pricing?
Because agents do the work and people supervise it. A fee per supervisor taxes exactly the thing that makes autonomous work safe. Invite the whole team.
What does a typical desk cost to run?
The reference desks — a weekly SEO measurement, a daily content radar, a nightly lead pipeline — run between fifty cents and eight dollars a run in model and compute. Every run shows what it cost, and three levels of budget cap it.
Can I set a hard limit?
Yes: per run, per workspace per month, per account per month. A run that crosses its cap stops between steps and the notification says why.
Is my data used to train anything?
No. Your workspaces are files in your account; your model calls go to the provider you chose with the key you chose. We run infrastructure, not models.
Can I move between Local, Cloud and Self-hosted?
The workspace is the same folder in all three. foldrun deploy or a git push moves it; nothing is converted.