An agent your team addresses in Slack.
Your team already works in Slack, so that is where the agent answers. A person mentions it by name in a channel, it works in the thread that opens, and anything that leaves your company waits there for a named person to approve it.
How a person addresses it
The agent has a name in your workspace, the way a colleague does. Nobody installs anything or learns a new tool: they ask in the channel where the work already happens.
- Mention it by name
A person writes the agent’s name in a channel it has been added to, followed by what they want done. Slack delivers that mention to your agent, and the agent starts work on it.
- The thread is the session
The thread under that message is one session. The agent answers in that thread and no other, and a second mention in the same thread continues the same session rather than starting a new one.
- A slash command, where you prefer one
A team that would rather type a command than a mention can have one. Slack sends the command to the agent, which acknowledges it at once and starts the work.
What the approval gate looks like
An approval gate is where the agent stops and waits for a named person. In a channel, it is a message with a button under it.
- The agent states what it wants to do
Before an outward action, the agent posts in the thread what it is about to do and for whom, with the draft or the figure attached, and a button to approve it. Then it stops.
- A named person presses the button
Slack delivers the press to your agent. The rule that only a named person’s press counts is ours to write, because Flue delivers the press and leaves the policy to us.
- The agent cannot choose where it posts
The channel and the thread it answers in, and the token it answers with, are bound in code we write. The model chooses the words of a message and nothing else: not a workspace, not a channel, not a credential.
An outward action is anything that reaches someone outside your company: a sent email, a published page, a posted message, a payment. Only a named person at your company approves one, and the press in Slack is that approval.
What you provide
Three things, all of them yours. We set the agent up inside them with you, so the account, the keys and the code are yours from the first day.
- A Slack workspace
The one your team already works in, and the channels the agent is to answer in.
- A Slack app in that workspace
The agent holds two values from it. The first is the app’s signing secret: the agent verifies the signature on every request before it reads a word of it. The second is the bot token the agent answers with. Both stay in your project.
- Somewhere for it to run
Your own Cloudflare account, or Node.js. Flue’s Slack channel runs on both.
If your team is in Microsoft Teams
The agent reads the same way to your team: a person mentions it in a channel, and it answers where it was asked. What changes is what you provide, and it changes on your side of it.
- Slack holds two values
The app’s signing secret and its bot token, both issued in your workspace.
- Teams holds three, and one registration
An app ID, a tenant ID and an app password, and an Azure Bot registration whose messaging endpoint points at your agent before Teams sends it anything. The registration is the step Slack does not have.
Teams needs one step that Slack does not: an Azure Bot registration and an app password stored in your project. We do that work and quote it as its own line.
Read what Flue verifies for TeamsWhat it costs
An agent in Slack is shaped around the work your team asks it for, so it is custom work, and the custom door prices it.
The door states the floor to build, what the month buys, and what a custom agent needs from you before we start. What custom AI agent development costs