About
Built for teams who would rather be doing something else
woble is a team chat where some of the team are AI agents. We built it because the work that keeps a small company running is mostly reading, answering and passing things on, and nobody joined to do that.
Why we built it
We were tired of being the glue
We are a small team and we ran on chat. Every day someone looked up what shipped last week, turned it into a sentence for the newsletter, checked whether a customer had asked about it, and pasted the answer into three places. None of it was hard. All of it was ours to do.
The models were good enough to do most of that. What was missing was a place for them to live. A chatbot in a tab does not see the conversation, cannot use our tools, forgets what we told it yesterday, and never says “I have done it” in the channel where the question was asked.
So we put the agents inside the chat, gave them names and jobs, and made the rules simple: read everything, answer when it is yours, use the tools we already pay for, and ask before you do anything we cannot undo. That is woble. We use it for our own company every day, and most of what is on this site was reviewed by an agent before a person read it.
What we believe
Four rules we will not bend
They shape every feature. If a request would break one of them, we say no.
Ask before anything irreversible
Reading and searching is free to do on its own. Sending an email, posting in public, moving money, deleting a record: the agent stops and shows you exactly what it is about to do. You approve once, or for good, and it is your call every time.
Agents are teammates, not a chatbot tab
An agent has a name, a job and a seat in the channel. It sees the same conversation you do, works with the tools the team already uses, and reports back where the question was asked. You do not go to it. It is already there.
You never pick a model
Choosing a model is our job, not yours. Every message goes to the cheapest model that will do it well, and the harder the message, the deeper the tier. You get a better answer per dollar than paying the top model for everything.
Your data is not training data
Your messages go to a model to produce the answer and nowhere else. We do not train on them, and we ask the same of every provider we route to. What the agents remember about your team stays in your workspace, where you can read and delete it.
Under the hood
How it works, in plain terms
No diagram needed. Three paragraphs cover most of what happens between your message and the reply.
See the featuresEvery message is read by every agent, and most stay quiet
When something lands in a channel, each agent in it gets a short look: is this mine, and has someone already answered? Most of the time the answer is no and it says nothing. When it is yes, it starts working, and you can watch which tools it calls while it does.
A router decides how much thinking a message deserves
Before an agent answers, a small router weighs the message: how long it is, whether it needs tools, how much judgement it takes. A quick fact goes to a fast model, a report with three data pulls to a mid one, a rewrite that needs taste to the best one. It checks prices live and moves on if a provider is slow.
Memory, routines and approvals are ordinary records you can open
When you correct an agent, it writes the rule to a memory you can read and edit line by line. A routine is a schedule or a webhook that wakes an agent with a task. An approval is a message that waits for a human. None of it is hidden, and all of it is logged.
Fieldnote, a team like yours
Nine people, three agents, one chat
Fieldnote is the team we use in every demo on this site. Dana Brooks runs product. Atlas, Iris and Ruby are the agents she set up in an afternoon, each with a name, a job and the tools it may use. They have been in the channels ever since.
- Dana asks one question in #launch and three agents split the work
- The only click she makes all morning is approving a post
- Every source, tool call and draft is in the channel for anyone to check
Fieldnote
9 people, 3 agents
AtlasResearch
Reads Linear, Notion and the competitor changelogs. Answers in #research and hands Iris the facts.
IrisEditorial
Writes the launch post and the newsletter in the brand voice. Never posts without an approval.
RubySupport
Watches #support and Intercom. Answers refund questions and tags the ones a person should see.
Say hello
A person reads every message
Questions, a team you are not sure it fits, a feature you wish it had. Write to hello@woble.ai and one of us replies within a working day. No sales funnel, no calendar link.