🚀Put an agent to work

Using agents

Build an agent once and use it everywhere. The same saved agent — with its model, prompt, knowledge, and tools — can answer in chat, run as a step in a workflow, respond over the API, or collect input as a form.

Four ways to run an agent

One configuration, four entry points. Pick whichever fits how people need to reach the agent:

🤖One agentBuilt once
💬In chat#-mention it
⚙️In a workflowAs an agent node
🔌Over the APICalled programmatically
📝As a formCollect structured input
A single saved agent is reachable from chat, workflows, the API, and forms.

Mention it in chat

In any chat, type # and pick the agent to attach it to the conversation, or open a chat directly with the agent. Its conversation starters appear as one-click prompts, and from there it behaves like any teammate — reasoning, calling its tools, and citing the knowledge it retrieved. Write actions are gated for your approval, as described in Tools & actions.

Drop it into a workflow

Add an agent node to a workflow and select your saved agent. The node brings the agent's own model, system prompt, knowledge, and tools with it — you don't reconfigure any of that in the graph. Pair the agent with a structured-output schema and downstream nodes can parse its result by field. Remember that, in a workflow, the agent's write actions run without a confirmation prompt — the wiring is the consent. Workflows.

Call it over the API

Invoke an agent programmatically with an API key and receive its response — free text, or schema-shaped JSON if you gave it a structured-output schema. This lets you embed an agent in your own product or scripts. API keys are issued under Settings, with usage tracking and an optional expiry. Settings for managing keys.

Run it as a form

If you defined form fields in step 5 of the builder, the agent can be run as a fillable form. Teammates enter the typed inputs you specified, those values are passed to the agent as structured input, and the agent runs against them — a clean way to let non-technical users trigger an agent without writing a prompt.

Templates, Try, versions & insights

Around the agent itself, Your Office AI gives you tooling to start fast, test safely, and keep an agent healthy over time:

ToolWhat it does
TemplatesStart from a pre-built agent and adapt it instead of beginning from a blank persona — a fast way to stand up common roles.
Try panelChat with the agent live inside the builder to test its tone, tools, and knowledge before you save or share it.
Version historyEvery saved change is tracked, so you can review past versions and roll back if a change regresses the agent's behaviour.
Usage insightsSee how often and how the agent is being used, so you can spot which agents earn their keep and which need tuning.
ℹ️
Share it with the right people

Before colleagues can use an agent, share it at the right scope — Private, Specific people & Agents, Workspaces, or Organisation, each with view or edit. See Members & Roles for sharing details.