Chat Dispatch

Message Forge, Work Happens — Text, Voice, and Context

Send a message to a bot and a workflow starts. Speak instead of typing and it transcribes first. Reply to something and it knows what you're replying to. The guess is always shown with a confirm button before anything runs.

A message is a valid way to start work

No dashboard, no form. Describe what you need in the chat you already have open and Forge resolves it to one of your organisation's workflows.

It always asks first

The classifier's guess never fires by itself. You get its interpretation with a confirm button; ambiguous requests offer a picker instead of choosing for you.

Voice notes work

Send audio instead of typing. It transcribes, then follows exactly the same confirm-before-running path as text.

It remembers the conversation

Follow-ups land in context. Reply to a message and Forge resolves what you replied to; ask "and on Monday?" and it still knows what you were talking about.

The job

The distance between “someone should look at that” and “someone is looking at that” is usually a context switch — open a tool, find the right place, fill something in. That distance is why a lot of small work never starts.

Forge shortens it to a message. The chat you already have open becomes the place work begins.

What happens when you send a message

It works out what you meant. A classifier maps your message onto the workflows your organisation actually has. It is not choosing from a generic menu — it is choosing from your catalogue.

It shows you its guess. This is the part that makes the feature usable rather than alarming. The interpretation comes back with a confirm button; nothing dispatches until you press it. When a request is genuinely ambiguous, you get a picker rather than a confident wrong answer.

You can be explicit when you want to be. Naming the workflow directly skips the guessing entirely.

Then it reports back. The “working on it” message is edited in place when the run finishes, so the chat reflects reality instead of accumulating stale notices.

Voice, because typing is not always available

Send a voice note instead. It is transcribed and then follows the identical path — same classification, same confirm gate, same reporting. The reason to mention this at all is that it is where the async promise gets real: describing a task out loud while walking, and finding the work done later, is a different relationship with a tool than sitting in front of one.

Context, so a follow-up is a follow-up

Two things make a chat feel like a conversation rather than a series of unrelated commands:

  • Replies resolve. Reply to a specific message and Forge reads what you replied to, so “do this one too” has a referent.
  • History carries. Ask a question, then ask a follow-up that only makes sense given the first — “and on Monday?” — and it still knows the subject. Conversation history is stored per chat, encrypted at rest, and bounded by a retention window.

Boundaries

  • Confirm-before-run is not optional. A guess never dispatches on its own.
  • Only your organisation’s workflows are reachable from your chat.
  • An unbound chat does nothing. A chat that has not been connected to an organisation is ignored rather than guessed at.

Under the hood

Ready to ship faster?

Forge runs the full cycle autonomously — from labeled issue to reviewable PR.