Tools are your real systems
An agent's tools are the same 3,585 integration nodes your flows use. No bespoke plumbing per model.
AGENTS
An agent your team messages in Teams, Slack or email
An agent is just a flow with judgement. It reasons about what to do next, then acts through the nodes you gave it. Give it real tools and it does real work, and it can only ever use the tools you gave it.
Sales
Customer Success
Finance
Product
Social
TeamsOperations
editor.flomation.app/executions/8f2a41
Purchase Deskawaiting approval
REASONING
Four lines below reorder point. Two suppliers approved, two not. Raising POs for the approved pair only.TOOL CALLS
stock.levels210mssales.velocity340mssuppliers.lookup120mspurchase_order.createapprovalTwo POs over £2,000 need a human tick before they go out.
TOOLS IN THIS FLOW
XeroShopifyGmailPostgres
Four nodes, so four tools. There is no fifth thing it can reach for.
CHANNELS
Works where your team already works
Nobody has to open Flomation to use an agent. Channel access is configured in the flow like everything else, so one agent can answer in Slack for ops and by email for suppliers without building it twice.
- Microsoft TeamsOne to one, in a group chat, or in a channel.
- SlackDirect messages, mentions, and threaded replies.
- EmailIt reads the thread it was sent, and replies in it.
- TelegramTyped messages or voice notes, answered in kind.
- SMS and voiceA Twilio number, for people who do not use chat.
- Web and MessengerEmbedded on your site, or on a Facebook page.
CONTROL
It can only do what you have given it
An agent's tools are nodes in its flow. If Xero is not in the flow, the agent cannot touch Xero, however anyone asks it to. There is no rogue action because there is no action outside the flow.
The flow is the whole action space
Add a node and the agent gains a tool. Take it out and the tool is gone. Nothing is granted by prompt, so nothing can be talked out of it.
A person signs off anything expensive
Put an approval gate in the flow and the agent stops there, mid run, until someone ticks it. It waits rather than guesses.
Every run lands in the Execution view
Reasoning, tool calls, inputs and outputs, step by step and replayable. The same view you already use for your flows.
And the rest of what you would expect
Your model, your keys
Bring your own provider, or run a local model on your own runners for sensitive work.
Hand off to a flow
When the judgement's done, deterministic work belongs in a flow. Agents call them directly.
GOOD FIRST AGENTS
Five that earn their keep in the first week
- SalesSlack
- A new enquiry lands, and it files the lead in the CRM with a summary attached and posts the gist to the channel, so the first reply is not the bottleneck.
- Customer SuccessEmail
- It answers supplier and customer email from your own order data, in the thread it was sent, and hands over the moment it is out of its depth.
- FinanceTeams
- Ask it which invoices are overdue with Acme. It checks Xero, drafts the chaser in the right tone for that account, and waits for you to approve it.
- ProductSlack
- It turns support threads into tagged feedback as they arrive, then tells the people who asked for something on the day it ships.
- SocialTeams
- Ask it for this week in public. It drafts posts from what actually shipped and waits for a human to say yes before any of it goes out.
Give it the tools, put it in the channel, let the team ask
Wire an agent to the systems it needs, drop it into Teams, Slack or email, and people can simply ask it for things.
Every run is in the Execution view next to the reasoning that produced it, so widening its remit is a decision you make with the evidence in front of you.