Orientation Tools

The platform describes itself, on demand

agentis.orient exists because of a specific, observed failure mode: a cold agent inferring what an App versus a Workflow versus a Subject versus a Connection even means from roughly seventy verb-named tools alone, with no re-queryable model of the platform to check itself against. Calling it returns the full six-primitive object model in plain language, the caller's actual current inventory (so a new build binds to what already exists instead of minting a duplicate App for what should've been one more workflow in an existing one), and any disclosed limitations of the current version — honestly, not hidden. It's read-only and safe to call as often as needed; there's no reason not to re-orient mid-task if context is uncertain.

A five-step creation order, stated explicitly

1. Find-or-reuse — check inventory from orient before creating anything new.
2. Build the workflow into an existing App where one fits, rather than defaulting to a new one.
3. Give it data and an interface (data.define_collection, ui.render).
4. For continuous operation, make the owning agent resident and/or attach a cron or listener trigger.
5. For messaging, secure a Connection grant before assuming outbound reach exists.

This ordering is itself part of what orient returns — it isn't tribal knowledge an agent has to have absorbed from a wall of prompting elsewhere; it's queryable platform self-description, the same paved-road philosophy that shows up as compass.next on every settled tool result.

Continue