Three lines on the bill, and you control every one of them. No execution counter, no machine count.
Two tiers for the hub, one line for cloud agents.
Concrete allowances, then a unit price, against a cap you control.
| Usage line | Price | Included | Note |
|---|---|---|---|
| Advanced thinking (overflow) | metered | 500 flows / month | Against a budget you set. Alerts at 75, 90, 100%. Stop-at-limit. Every request on your statement with its reason. |
| Cloud agents | $5 per trained agent / month | — | Trained agents we host 24/7. Your own agents on your own machines are free and unlimited. |
| Capacity routing | included in Team | every machine in the org | A laptop sends the job to the machine with headroom; it runs there on hardware you own and the receipt comes back. No node count. |
| Storage on the hub | $0.25 per GB-month | 5 GB | The hub holds receipts (~2 KB), workflow definitions, approvals and index entries. Your data stays on your machines. A 20-person team uses about 0.4 GB in three years. |
| The switch | always free | every tier | Stops every agent on every machine. Works past a budget limit. |
Set it once. The router stays under it. Local work never pauses.
Every model at once. Drag how much of the work your machine answers; every row moves with it.
$ per 1,000,000 tokens of work. Blue: every request to that model at its list price. Pink: RailCall, your share answered on your machine at $0 and the rest sent to a model of the same class, metered. 90% is the design target, not a promise; drag to see your own mix.
Twenty people, ten machines, twenty agents on their own hardware, two we host, 20,000 AI-touching steps. Drag it to your own team.
Everything local is $0 and never counted. Overflow is metered against a budget you set, priced here at the top model’s class (Fable / Astra) so it matches the competitors’ AI line; rounded to $5. 500 overflow flows a month are included.
Cursor is a code editor and coding agent with an included usage pool. RailCall is a governed local runtime and team hub, not an editor. This compares what each charges per human and per token; if the team keeps Cursor, the two bills stack.
Cursor Teams is $40 a standard seat (cursor.com/pricing, 2026-10-07): $800 for 20 people before any model usage. The same 20,000 AI steps at the top model’s list are $1,400, plus about $15 of Cursor’s token rate; Cursor bundles a usage pool of unpublished size into the seat, so its line is a ceiling, marked ≤. Even if that pool absorbed the entire seat price in usage, the bill would still be about $1,415. Claude Team is $20 a seat billed annually; the same 20,000 automation steps run through the Claude API, outside the seats, at list. Inside and beyond the pool, third-party model usage is billed at the provider’s API list price plus Cursor’s $0.25-per-million-token rate on Teams (cursor.com/docs/account/pricing, 2026-10-07). RailCall: 20 × $5 + 2 hosted agents + the overflow line from this team’s statement, $210, for $320 all-in. 90% local is the design target, not a promise: at 80% local the overflow line is $490 and the bill about $600, still a fraction of either. Access to Cursor’s cloud agents is included in its seat and their model usage draws on the same pool; ours are a separate $5 line.
A human with a RailHub login who can build, approve or administer. Machines, agents, stations and cloud agents are never seats and are unlimited at every tier.
No. Everything on your machines runs without one. You log in only for the hub, cloud agents and the overflow budget.
The ~10% of requests your machine can't answer at the quality you need. They go to a bigger model and are metered against your budget; each one shows on your statement with the reason it left.
Overflow pauses if you chose stop-at-limit. Local flows keep running. The switch keeps working. You raise the budget and it resumes immediately.
Receipts, workflow definitions, approvals, index entries. Never your data, never your files. Heavy data stays on your machines.
Yes. The local product has no dependency on the hub. The hub is for teams that want one ledger, one library and approvals across machines.
No. Every hub gets the audit features: evidence packs, outside-auditor anchoring, cross-machine chains. SSO and SCIM are roadmap, and we do not hold a SOC 2 report. The trust page states exactly what is in place.
No. It is a usage fee; you control it with the budget and see every request on your statement.
Cursor Teams is $40 a standard seat with an included usage pool of unpublished size: $800 for the 20-person team on this page before model usage. The same 20,000 AI steps at the top model’s list add $1,400 plus about $15 of Cursor’s token rate, so Cursor’s line is a ceiling, at most $2,215. RailCall’s bill for the same people is $320, of which $210 is metered overflow. RailCall is not an editor, so if you keep Cursor the bills stack. 90% local is the design target, not a promise; at 80% local the RailCall bill is about $600.