Free to try.
Paid when it needs to persist.
Every plan runs a real Postgres database per instance. That is why the free tier has a clock on it: an idle database that nobody is paying for gets spun down, always, without exception.
Free
One system, 24 hours, no card.
- One mock system of your choice
- MCP server plus REST API for that system
- Company knowledge graph, read only
- Data explorer and generated schema docs
- Instance is spun down 24 hours after creation
24 hour lifetime · 1 system · 1 instance
Your current plan$49.99 / month
Keep the data. Five systems.
- Up to five mock systems per instance
- Data stays live as long as the subscription is current
- Three concurrent instances
- Knowledge graph refresh on demand
- Generated OpenAPI documentation per instance
Runs while the subscription is current · 5 systems · 3 instances
$199 / month
Data that keeps moving, so your agents have something to react to.
- Everything in Persistent
- Fresh records injected every minute, hour, day, week or month
- Scenario engine: support spikes, quarter-end audits, outages, launches
- Existing records mutate over time, so history is not frozen
- Unlimited systems per instance
- Scheduled agent runs have something new to work on every cycle
Runs while the subscription is current · 40 systems · 10 instances
From $999 / month
Many people, one shared world. Priced per engagement.
- Everything in Live
- Multiple named users on the same instances
- Shared MCP endpoints across a cohort or a leadership team
- Org-level roles, audit log and SSO on request
- Custom systems, custom volumes, custom scenarios
- All pricing quoted per engagement
Runs while the subscription is current · unlimited systems · 100 instances
Request a quoteOne month of Persistent, no charge
Start the Persistent plan and the first month is free. Card details are taken at the start so the subscription continues without an interruption, and Stripe cancels cleanly from the billing portal at any point before it converts. The trial is offered once per customer.
The database is destroyed, not just hidden
When a free session hits 24 hours, or a subscription lapses and its grace period ends, the Postgres service is deleted. The MCP servers reply that the session has expired rather than timing out. The schema and the knowledge graph survive, so re-subscribing rebuilds the same world at the same URLs.
What Live actually does
A worker checks every minute for instances that are due. When one is, it picks a scenario weighted for that company’s systems and writes it into the data: a support spike traced to one product fault, a quarter-end close with accruals and reclasses, a runaway query that burns a month of warehouse credits overnight, a social post that goes badly.
It also advances what is already there. Tickets get solved, deals change stage, invoices age. That is what makes an autonomous agent worth running on a schedule against this: there is genuinely something new each time it wakes up.
Cadence is per instance and you choose it: every minute, hour, day, week or month.
Questions people actually ask
Does any of this touch our real systems?
No. There is no path from this product to a real vendor. Nothing authenticates against your Salesforce, and no request leaves this platform for one. Every business record is generated.
Is the company research a privacy problem?
The crawler reads only what is publicly published, honours robots.txt, caps how much it fetches and identifies itself. The knowledge graph contains nothing you could not read yourself on the company website.
Can several people at the same company share one instance?
On the Enterprise plan, yes: named users share the same MCP endpoints and the same world, which is what makes a cohort exercise work. On the other plans each person gets their own instance, though the detected systems are shared so nobody repeats the research.
What stops a database running forever by accident?
Four independent checks, every minute: the hard TTL, the subscription entitlement, an idle timer, and a global ceiling on concurrent instances. Any one of them is enough to spin an instance down, and every teardown is logged.
Can we self-host it?
The whole thing is one Next.js service, one worker and Postgres. Enterprise engagements can run it inside your own infrastructure. Get in touch.