What the AI does
Every 30 minutes, when there are new messages, TXM reads them together with the conversation around them. Then it:
- finds requests and promises (“can you send the report by Thursday?”, “I’ll fix it today”) and turns each into a task with an owner, the person who asked, a due date and a project;
- updates tasks when people say they started, are waiting for something, finished, or found the bug again;
- links tasks that depend on each other, so a blocked task shows what it is waiting for;
- keeps the evidence: every task links to the messages it came from and the messages that changed it.
You do not type commands or fill in forms. People keep writing the way they already do.
What it looks like
| Someone writes in the group | What TXM records |
|---|---|
| “@dawit can you push the onboarding flow before Friday’s demo?” | A task for Dawit, due Friday, asked by the person who wrote it |
| “on it” (as a reply) | Dawit owns the task and it is in progress |
| “waiting for the API key from the client” | The task is blocked, and the note says what it waits for |
| “done, check v1.0.452” (as a reply) | The task is closed, with the reply as the reason |
| “it broke again” | The task is reopened |
| A screenshot of an error and “can you look?” | A task named after the error in the screenshot |
How it decides
The AI reads a message the way a teammate would, not by keywords:
- Replies are about what they reply to. A short “done” closes the task the replied message belongs to.
- Who speaks matters. “I did it” from the person doing the work closes the task. “Is it done?” from the person who asked is a question, not a new task.
- “ok” is not done. “ok”, “noted” and 👍 mean someone saw the request. The task stays open.
- Pictures count. A screenshot of the working screen is proof; a screenshot of the same error means it is still broken.
- People win. If someone changes a task on the board, TXM keeps that decision.
You stay in control
The AI can be wrong. When it is, fix the task on the board with one tap: change the owner, the status or the due date.
Every task shows the messages behind it, so you can always see why TXM did what it did. TXM never posts in your group
on its own. It answers /tasks, /mytasks, /captured and /board, and its replies delete themselves after 30
minutes.
AI task manager or command bot?
| Command bots | TXM (AI) | |
|---|---|---|
| Creating a task | Type a command with the right format | Nothing: it reads the conversation |
| Closing a task | Someone remembers to close it | Someone says it is done |
| Mixed languages | Usually English commands only | 50+ languages, including mixed messages |
| Screenshots | Ignored | Read as part of the task |
| Keeping it current | Depends on people | Updates itself every 30 minutes |
Languages
TXM works in 50+ languages, including Amharic in its own script or typed in Latin letters, and messages that mix two languages in one sentence.
Questions
Does the AI post in our group?
No. TXM never posts on its own. It only answers commands, and its replies delete themselves after 30 minutes.
Can the AI make mistakes?
Yes. Fix any task on the board with a tap. Your change always wins, and every task shows the messages it came from.
Do we need to write in a special format?
No. Write the way you already do. TXM reads normal conversation, replies and screenshots.
How quickly do tasks appear?
Within 30 minutes of the message, when there are new messages.
Who can see the tasks?
Only members of the group. The board checks membership every time it opens.
Is it free?
Most features are free.