Blog
Analogies and real architecture behind building AI agents with Wetel.
-
Procedures you can read: why an AI employee should follow a graph, not one long prompt
An AI employee's way of working should be something your team can read, test and change on purpose: a set of steps with clear branches, not one long instruction you hope the model keeps to. Here is how that works in Wetel, and what it does not fix.
-
Designing approval steps people actually read
An approval step only protects you if the manager reads it. How to write the title, summary, editable fields, deadline and replies of a Wetel approval request so a person can decide quickly and well, instead of clicking through.
-
How do people reach an AI employee?
Today there are four ways: a 3D avatar that speaks and listens, a text chat you build from a working reference, your own app through the API, and a messaging app that is connected directly. Phone calls are not built yet.
-
Reading a file a customer sends in chat: how an AI employee receives it, checks it and uses it
A customer sends a PDF, a photo or a document in the middle of a conversation. Here is how a Wetel AI employee receives the file, checks its type before touching it, opens it, hands it on, and what happens when a file is missing, unsupported or too large.
-
Which voice should my AI employee use for which language?
A speaking voice belongs to one language and one quality tier, and the language an AI employee listens in is a separate setting. How to pick both, what the tiers trade off, and what we have not measured.
-
What is a run record, and how do I read one?
A run record is the saved account of one workflow run: each step the AI employee took, what it produced, and where it failed. Here is how to read one with the documented query, and how to use it to check a reply against what really happened.