SproutPadResources
How this works
Seven short explanations of the machinery, each one traceable to the open specification.
Start here
How the product works
What SproutPad does on your behalf, and where it stops and asks.
The agent contract
How an agent reaches SproutPad, what a scoped key can do, and why every entrance returns the same envelope.
Human approval gates
What gets gated, who can decide it, what the approval page must show, and what happens if nobody answers.
Running it day to day
How an agent updates the site, manages DNS, sends and reads mail, and reshapes the project after launch.
Exit paths
What stops billing, how a domain transfers out, what you can export, and which paths still need a human operator.
Under the floor
How the machinery works
The parts that make the guarantees above hold under load, retries, and failure.
Budgets and reservations
Why the money is held before it is spent, what an estimate is not, and how headroom is actually computed.
Durable execution
Tasks, their four states, why restarts are safe, and how multi-step work compensates when a step fails.
The append-only ledger
One entry per mutation, enforced append-only at the storage layer, and why teardown is computed from it rather than hardcoded.
Idempotency and retries
How Idempotency-Key is bound, what replay actually returns, and why adapters carry their own internal keys.
Build against it
The reference, the contract, and the evidence — in the order most people want them.
API reference
Every endpoint, its cost, its retry contract, and the envelope it returns.
The open specification
The normative rules any governor must follow. SproutPad is one implementation of it.
Proof sandbox
Run a fixed fake-money pressure test and watch the limits refuse a request.
Signed conformance
Which named build passed the published checks, and when they expire.
For agents. Everything on these pages has a machine-readable counterpart: llms.txt, openapi.json, and the MCP tool contracts. If a human page says it, an agent can fetch it.