Archives

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.

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.

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.

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.

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.

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.

A walk through our documented order-support recipe: how an AI employee looks up an order, answers product questions from the shop's own material, where it stops, and where the shop's staff stay in charge.

A demo shows where a product is going; a contract needs what works today. Why we mark every capability with its tense, how to test any vendor's claims the same way, and five questions to ask before you sign.

A system is down, a step errors, a customer sends something the workflow cannot use. Here is what a Wetel AI employee says by default, how to route a failure to a reply you wrote, and where the real reason is recorded.

Credits are a unit for reading your AI usage, and each plan has a monthly spend ceiling checked before work starts. Here is what is metered, how to read it, and what credits do not promise today.

Treat your first AI employee like a new colleague's first week: one narrow job, a written procedure, the right documents, a manager's sign-off where it matters, a rehearsal, a small start and a daily review. A plan, not a guarantee, with what is live in Wetel today and what is not.

A rate-limit error means one operation was called more often in its time window than Wetel allows. Limits are counted per calling address, so every customer behind one backend shares the count. Here is what to do.

A conversation in Wetel opens when your app starts it and stays open until your app or the workflow ends it. One left idle for 30 days is marked expired, and the next message starts a fresh one. Here is what happens at each point.

A walk through one documented deployment: an AI employee that answers students from a library's own catalog, checks rooms and renews loans through the library's own systems, and what the case study does and does not tell us about where people stay in charge.

An AI employee is the wrong tool for some work: fixed steps with exact numbers, irreversible decisions nobody checks, channels or features we have not built, and processes nobody can write down. Here is how to tell, and what Wetel does not offer today.

How a Wetel AI employee answers from your company's own policies and documents, how you can see which passages it used, how you keep the documents current, and what to do when the answer is not in them.

Every AI employee runs on the platform's default model unless someone picks another one, and the default provider tries a second provider automatically if a call fails. Here is who can choose and restrict models, and what you see when something goes wrong.

How to let an AI employee use a tool in another system without writing a plugin: register an MCP server once as a connector, call one named tool from a workflow step, and rely on the checks that sit around every call. What is live, how to test it, and what is not built yet.

Every person in a Wetel workspace is either an owner or a member. Members can build; only owners can decide approvals and hire a role. Here is the full list, how an owner is set, and what is not built yet.

Run your edits as a draft in a test session without publishing them, simulate each approval outcome, then read the run record. What a draft test does, what it does not do, and a short checklist before you publish.

A walkthrough of the two documented ways to build an applicant-screening AI employee in Wetel: a scored first-round interview, and a messaging-app intake that collects a résumé and files it in your applicant tracking system. Plus what happens when a file is missing or wrong, and why a person still makes the hiring decision.

A role template is an AI role with the job description already written: a persona, a conversation flow, knowledge and the connections it needs. Here is what hiring one in Wetel means today, who can do it, what you still set up yourself, and what is not built yet.

How a scheduled workflow lets an AI employee do the Monday report without being asked, the three limits that keep it from spending or failing without end, and how to check on it, since it does not send an alert when it stops.

REA gives AI agents a standardized interface to drive Ghidra, Hopper, and IDA. It can extract your frontend widgets and GraphQL schema in hours. It cannot touch your workflow engine, connector logic, or auth layer. Here's where the real moat lives.

Yes, by workspace. Every request is tied to your workspace through the signed-in session or API key. Here is what that covers, what is not in place yet, and what to ask any vendor.

When demand outstrips capacity, first-come-first-served fails. How emergency triage maps to support queues: tiers, a four-step lifecycle, and where an AI employee fits without becoming an autopilot.

What happens when a customer sends a voice note to a Wetel AI employee, how the language settings work for English, Malay and Chinese speakers, the documented limits, and the honest list of what we have not measured or built yet.

A plain-language round-up of the week's changelog: manager approvals, photos read by URL, speech changes, a per-agent supervision view, roles and model limits, and structured attachments.

An approval request in Wetel does not wait forever and is never approved by default. When its time runs out it expires, and the workflow follows the expired path you wired up. Here is what to put on that path and what the owner sees.

A customer sends a picture of a parcel label, a receipt or a form. Here is how a Wetel AI employee fetches the image, reads it, uses what it read, and leaves a record you can check, plus the limits and the confirmation step we recommend before anything consequential.

Three terms that get used as if they meant the same thing. A plain guide for managers: what each one can do, what a chatbot is missing, and the parts that make an AI employee, with what is live in Wetel today and what is still being built.

Two promises hide behind 'AI employee': one acts alone, the other checks with you. A practical way to decide what an AI employee may do on its own, before you hand over the keys.

A click-through walkthrough of Wetel's SUB_AGENT dispatch node and the certified operation picker — how to build one coordinator agent that hands a conversation off to three specialist agents at once, each pulling real, live data out of a connected partner system, without writing a partner integration by hand.

How Wetel's SUB_AGENT v2 concurrent dispatch and its certified operation catalog combine to let one AI agent reach into multiple partner APIs (OmniChat, Nortia, aisCRM) at once, all three firing concurrently with real data — verified live with real overlapping timestamps and an honest account of what it took to get there.

A complete, reproducible walkthrough — from creating a Lark app in the developer console to a working AI agent that answers real questions about your Lark Base. No code required for most of it, and the one real bug we hit along the way (with the fix) is worth reading even if you're not touching Lark.

Most voice AI platforms call themselves 'interruptible' when they really mean turn-taking. Real barge-in needs voice-activity detection, frame-level audio handling, and a workflow engine that can cancel itself mid-thought — here's what that actually takes to build, and how Wetel does it.

"Just give the agent some tools" hides four distinct layers that get built separately, tested separately, and fail separately. A call-center analogy for the terminology every AI vendor uses, but rarely defines the same way twice.