AI Tools & Actions
The toolkit for building, running and governing agents that act — the agent library, the studio to assemble them, the node catalog they draw on, the evaluations that prove they work and the governance that keeps them safe. Register a function, an API, an MCP server or an action node, and an agent can reach out of the chat and change something in the real world.
From a chatbot to an agent that does
The difference between a chatbot and an agent is action. Tools are how an agent reaches out of the conversation and does something — send the email, update the record, book the slot, call the API.
This is the toolkit that surrounds those actions: a place to build agents, a catalog of nodes to compose, a way to test them before they ship, and the controls that gate and log every move. Each part has its own page — start below.
Explore AI Intelligence Layer →- ✓ Agent library & studio
- ✓ Node catalog to compose
- ✓ Function & tool calling
- ✓ MCP server support
- ✓ Evaluations before you ship
- ✓ Governance, gates & audit
Everything to build and run agents
Each tool has its own page — this is where they connect.
AI Agents Library
Ready-made agents for common jobs, as a starting point you can adapt.
AI Studio
The builder where you assemble, configure and test an agent.
AI Node Catalog
The nodes agents draw on — score, decide, generate, retrieve, act.
AI Evaluations
Test agents against cases before and after they ship.
AI Governance
Guardrails, approval gates and the audit trail over every agent.
Multi-Agent
Compose agents that hand work to each other for larger jobs.
How agents take action
Function calling
The agent calls the right function with the right inputs, on its own.
MCP servers
Connect Model Context Protocol servers so agents use standard tools.
Action nodes
Draggable action nodes let agents send, update, book and execute.
Integrations & APIs
Reach any connected system or REST API from inside an agent.
Approval gates
Require a human sign-off before a high-stakes action runs.
Audited
Every tool call and action is permission-checked and logged.
Connect your systems once and reuse them across agents. Integrations →
Live in four steps
Register tools
Add functions, MCP servers, APIs and action nodes.
Agent decides
It picks the right tool for the step it is on.
Calls the tool
With the right inputs, gated by approval where set.
Acts & logs
The action runs and is written to the audit trail.
How the agent works with AI Tools & Actions
The Flow-Cognition Agent runs a real loop on your data — grounded in your records, governed and logged.
Sense
Reads the task and the tools available to it.
Decide
Decides which tool to call and with what inputs.
Act
Calls it, gated and logged — then continues the loop.
Five-tier model routing · field-level permissions · full audit trail. See the AI layer →
Where teams put it to work
Agents that do
Agents that book, update and send — not just answer — built and tested in the studio.
Workflow automation
Action nodes wire AI decisions into real processes, so a decision becomes a step that runs.
System integration
Agents reach CRM records, cloud telephony, messaging channels and the APIs you connect.
Prove it, then gate it
Evaluate before you ship
Run agents against test cases so you know how they behave before they touch a customer or a record.
Guardrails & gates
Set what an agent may do, require sign-off on high-stakes actions, and keep humans in the loop.
Watch it in production
See what agents did, how tools performed and where to intervene, with the full trail logged.
Questions, answered
A function, API, MCP server or action node the agent can call to do something — send an email, update a record, book a slot — so it acts rather than only answering.
Model Context Protocol — a standard way to expose tools to agents. It is supported here, so MCP servers you connect are available to agents out of the box.
In AI Studio, composing from the node catalog and the agent library, then proving them with AI Evaluations before they ship.
Yes. Any high-stakes action can be gated behind a human sign-off, and every tool call is permission-checked and written to the audit trail. See AI Governance.
Workflows are fixed steps that always run the same way; tools let an agent choose which action to take, and when, within the guardrails you set.
Put the toolkit to work
Start free, or get a guided walkthrough — build an agent, give it tools, test it and ship it under your controls.