Simple, transparent pricing
Free to start, $19 per seat
Every plan includes hosted agents, toolkits, skills, channels, schedules, the gateway for your own agents, access policies and the audit log. Tool calls are free. You pay for sandbox time and model tokens, and Free includes $3 of that every 30 days. Pro includes $10 per seat each month.
What each plan includes
Free renews $3 of usage every 30 days. Pro adds $10 for every billed seat on each paid invoice. The balance belongs to the workspace rather than to a person, so one teammate's busy week does not strand anyone else.
In every plan
- Hosted agents and every marketplace tool
- Per-user policies on every action
- Complete request and response audit logs
Free
Run real agents on the complete gateway, free every month.
$0
forever
- $3 of usage every month
- Self-host with Docker Compose
- Up to 2 workspace seats
Pro
Most teamsFor teams running agents in production.
$19
per seat, per month
- $10 of usage per seat each month
- Self-host with Docker Compose
Enterprise
Managed hosting, SSO, extended retention and an SLA.
Custom
annual agreement
- Usage allowance to match your volume
- Managed hosting, SSO, extended retention, and an SLA
Seats follow your active member count. Adding or removing a teammate adjusts the subscription quantity and prorates the difference.
What you actually pay for
Tool calls are free. Money goes on two things: the time your code holds a sandbox, and the tokens your models read and write. That is the entire rate card.
$0every tool call
Tool calls
Calling a tool costs nothing. Not on the direct path, not at volume, not on any plan. Searching the catalog, listing toolkits, reading a usage guide and building a toolkit from a specification are free in the same way.
$0.08per hour, 1 vCPU and 2 GB
Sandbox time
What your code costs while it is running. CLIs and stdio MCP servers run in your project sandbox, and so does any action a connector marks as sandboxed. Billed per GB-second of memory actually held, which is at parity with E2B and Daytona. Time a call spends queued is never billed — our queue is our problem.
2xOpenRouter's published price
Model tokens
What the models read and write when you use the managed provider. Computed from OpenRouter's live price rather than a rate card of ours, so it follows their changes the day they make them and there is no second copy of every price to go stale.
$0with your own key
Your own API keys
Bring your own provider key and the tokens are still counted, still attributed and still drawn in your usage charts. They are charged nothing. You are already paying the provider; we are not going to charge you for watching.
The install dialog states whether a tool runs in a sandbox before you install it, and every call records its path in the audit log. A direct call never touches a sandbox, so it bills nothing at all.
Add money at any time
A top-up is a one-time purchase at face value. $20 adds $20 of usage — no bonus, no rate, no arithmetic. It lands in the same workspace balance as your monthly allowance and does not expire, so a quiet month is not a loss and a spike does not need a plan change.
Top-ups work on every plan, including Free.
Questions about usage and seats
What does usage cost?
Two things spend money. Sandbox time is $0.08 an hour for a 1 vCPU, 2 GB sandbox, billed per GB-second of memory actually held, and time spent queued is never billed. Model tokens through the managed provider cost twice what OpenRouter publishes, computed from their live price. Tool calls themselves are free, and so is everything around them: searching the catalog, listing toolkits, reading a usage guide, building a toolkit. Calls made with your own provider key are metered and drawn in your charts, and charged nothing.
How do I know whether a tool runs direct or sandboxed?
The install dialog tells you before you install it, and the audit log records the execution path for every call that has already happened. It is the only thing that decides whether a call bills anything at all: a direct call is answered from Toolspoke's own process and never touches a sandbox, so it is free outright. You never have to find out from an invoice.
Do agents pay to browse the catalog?
No, and they do not pay to call a tool either. Searching for a tool, listing toolkits, reading a usage guide and making the call are all free. What you pay for is the sandbox time a call holds and the model tokens it reads and writes.
Is the balance per user?
No. Your workspace has one balance, so teammates and their agents spend from the same money. The seat count follows your active member count and decides how much included usage each paid invoice adds, not who may spend it.
Does the free allowance run out for good?
No. Free is topped back up to $3 of usage every 30 days, so a heavy week costs you the rest of that month and nothing more. It tops up to that figure rather than adding to it, which means an idle workspace does not quietly accumulate a year of allowance — and money you paid in is never reduced to make room for a free one.
Is a bigger top-up cheaper per dollar?
No, and there is no longer a rate to compare. Every top-up is face value: $20 adds $20 of usage, $100 adds $100. No volume bonus, no best-rate rung. A seat is still the better deal, for an honest reason rather than an arithmetic one: $19 buys a seat, the features and $10 of included usage every month, and a top-up buys none of that.
Why one paid plan instead of two?
Because the second one was sorting people by how much they call rather than by what they need, and every plan already includes the whole product. Pro is $19 per seat and includes $10 of usage a month for each one — about 250 GB-hours of sandbox, or millions of tokens of a cheap model, on top of tool calls that cost nothing. Past that you add money at face value, and Enterprise exists for the things that are genuinely different: managed hosting, SSO, extended retention and an SLA.
What happens when we add a teammate?
The Stripe subscription quantity follows your active member count, with prorations. The next paid invoice adds the plan's included usage for each billed seat, keyed to the invoice so a redelivered webhook cannot add it twice.
Can two people use different credentials for the same tool?
Not on the same install. An install holds exactly one credential, and that is deliberate: which of two stored values a call used should never depend on who owned the agent key. When you genuinely need two, give each its own project. Projects are already the line agent keys, policies and the sandbox are drawn on, and they cost nothing extra.
Can we run this without Stripe?
Yes. Toolspoke runs with billing entirely unconfigured, which is the normal shape of an internal deployment. Every workspace starts on Free and nothing asks for a card.
Running this inside your own infrastructure? Stripe is optional. Toolspoke starts with billing unconfigured, every workspace on Free, and no card prompt anywhere.
Self-hosting guideStart with the free plan
Tool calls cost nothing, and $3 of sandbox time and model tokens is renewed every 30 days. No card required, and nothing held back for a higher tier.