Reference

What the tools cover

The connection offers purpose-built tools for the work that comes up constantly — reading a collection's shape, finding entries, writing them with validation, checking booking readiness, looking up automation options, inspecting a run, and installing a blueprint — plus step-by-step recipes for multi-step tasks. Prefer them over improvising: they validate input and explain failures.

Requirements: a connection to a business, with a role permitting the work — see how an agent connects.

An agent could do everything through raw calls against the underlying interface. It should not, and the reason is practical rather than stylistic: the purpose-built tools validate what you send, refuse it with an explanation rather than a shrug, and hand back exactly what you need for the next step. They are the difference between an agent that works reliably and one that guesses.

What the tools cover

  • Understanding a collection. Ask what collections exist, and what fields one has — including the exact form each field expects when you write to it. This is the tool to use before writing anything, because it removes the guesswork that causes most failed writes.
  • Finding entries. Search a collection and get back a compact list with the identifiers you need to act on — rather than pulling whole records and sifting through them.
  • Creating and updating entries. Writes go through the same validation the interface uses, and a rejection names the field and what it expected. Fix that field and retry, rather than re-sending the same thing.
  • Checking booking readiness. A snapshot of the location, department, hours, and settings chain, with a checklist of what is still missing — far faster than reasoning about why availability is empty.
  • Looking up automation options. What automations can be triggered by and made to do, including the pitfalls that cause silent failures. Read this before composing an automation, not after it hangs.
  • Inspecting a run. Step by step, what fired, what each step did, and where it stopped — the answer to "why did nothing happen".
  • Installing a blueprint. The staged flow: what it will create, what conflicts with what you have, then the install itself.

Recipes come first

For anything spanning several steps, there are recipes: short sequences for common tasks like creating a collection, adding a field, setting up stages and pipelines, making something bookable, inviting a teammate, connecting a mailbox, importing media, and building a report.

The habit worth forming: if a task takes more than one step, look for the recipe before improvising. A recipe encodes both the order and the traps — what must exist before what, which step quietly depends on another. Rediscovering that by trial and error is slower and leaves half-finished state behind.

Beyond the tools

The curated set covers common ground, not every corner. The wider surface of the product remains reachable for the long tail, so an unusual need is not a dead end. The order of preference is simple: a recipe if one exists, then a purpose-built tool, then the wider surface.

Working habits that avoid trouble

  • Read the shape before writing. Ask what a field expects rather than assuming — the single biggest source of rejected writes.
  • Never write computed fields. Rollups and lookups are calculated and refuse writes by design.
  • Narrow large responses — ask for fewer records or fewer fields — rather than repeating a request that came back truncated.
  • Confirm which business you are in before concluding something is missing.
  • Fix only what a rejection names, then retry once, rather than changing several things at random.

What the connection can do overall is in working with Sonta programmatically; connecting and scoping in how an agent connects. Understanding what you are acting on starts with the data model.

Frequently asked questions

My writes keep being rejected.
Ask for the collection's field shapes before writing — a rejection names the field and the form it expected, so fix that one field and retry rather than re-sending.
Should I use the tools or call the interface directly?
Prefer a recipe when one exists, then a purpose-built tool, then the wider interface for the long tail.
A response came back truncated.
Ask for fewer records or fewer fields. Repeating the same request returns the same truncated result.