> ## Documentation Index
> Fetch the complete documentation index at: https://docs.meetsquad.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Agents
> A standing job that runs on a schedule or an event, works your brief, and produces signals, knowledge, or a written report.
An agent is a standing job. It runs on a schedule or when something happens, works the brief you wrote for it, and produces something you can use. Set it up once and it keeps going without you.
Open **Agents** in the sidebar. Two tabs sit at the top. **Agents** holds everything running in your workspace. **By Squad** is a gallery of ready-made agents you can start from, covered in [Agent templates](/agent-templates). **New agent** in the top right opens the builder.
Five ready-made agents written by Squad. Create one in a click, or open it prefilled and change anything first.
What makes an agent run: a schedule, a workspace event, or an event from a connected integration.
One list that decides what an agent can reach and what it can do there.
Every time an agent ran, what it changed, and what it produced.
## The agents list
The list opens as a grid of cards, one per agent, with a toggle in the toolbar to switch to compact rows. The toolbar also carries search, filters, and sort. Filter by **Status** (Active, Paused, Error) or by **Cadence** (Hourly, Daily, Weekly, Monthly, Quarterly, Annually). Your choice of view, filters, and sort stays in the URL, so a filtered list is shareable.
Each card and row shows the agent's name with a second line naming the page it watches or the opening of its brief, then its cadence, when it last ran, and its status. The overflow menu holds **Run now**, **Pause** or **Resume**, and **Delete**. Right-clicking a row opens the same menu. Click anywhere else to open the agent.
## Statuses
An agent is in one of three states.
Scheduled and event triggers fire normally.
Nothing fires. The agent keeps its configuration and history. Hit **Resume** to bring it back.
The last run broke badly enough to stop the agent. It does not recover on its own, which keeps a dead source from being retried forever. Open the agent, expand the most recent run, read the error, fix the cause, then **Resume**.
## Building an agent
**New agent** opens one scrolling form. Fill it top to bottom and click **Create agent**.
The name (unique in your workspace) and an optional description. The description is a note for your future self and is never fed to the agent.
Decide when the agent runs. A schedule, a workspace event, an integration event, or a mix. See [Triggers](/agent-triggers) for every option.
The standing brief the agent reads on every run, written in plain English. Say what to look at, what counts as worth flagging, and what a good result looks like. The agent decides each run what to reach for within the access you gave it.
What the agent may touch: your knowledge base, insights, goals, the open web, and any connections you add.
Levels are enforced, not suggested. See [Tools and access](/agent-tools-access).
What the agent produces.
## The three outputs
Three independent switches. Turn on any combination.
Extracts product signals from what the agent read this run and pushes them into the pipeline, where they cluster into [insights](/insights) alongside everything else.
Lets the agent update or create documents in your [knowledge base](/knowledge-documents) from what it found.
Produces a written brief answering a question you set. Turning it on opens the report settings underneath it.
At least one output must be on or the form blocks the submit and tells you so.
Signals and knowledge route themselves. A report is the only output with delivery options.
## Report settings
The **Report question** is what every report from this agent answers. Squad drafts one from your instructions the first time you turn reports on, and redrafts it when you edit the instructions and click away. **Regenerate** rewrites it on demand. Edit it freely.
**Report title** is optional and accepts `{{date}}` and `{{agent.name}}` as placeholders.
**Report when nothing changed** is off by default, so a quiet run produces no report. Turn it on for standing digests that should ship every period regardless.
Every report is saved to the system **Agent Reports** folder, so it is always editable and searchable even with no delivery configured. **Delivery** adds copies elsewhere: a **Slack** channel (available once Slack is connected) or **Email** to selected org members. Knowledge and insights are not delivery channels. The Gain knowledge and Discover insights outputs already cover that ground.
## The agent detail page
Opening an agent gives you the whole thing on one page.
The header carries the name, description, and the lifecycle controls: **Run now**, **Edit**, **Pause** or **Resume**, and an overflow menu with **Delete**. Deleting stops future runs and removes the agent. Reports it already wrote stay in your knowledge base.
Below that, every part of the configuration is editable in place: **Instructions**, **Triggers**, **Tools and access**, **Report** (shown only when reports are on), and **Run History**.
The sidebar holds an **Info** panel with **Status**, **Next run**, **Last run**, and **Created**, a **Latest Report** card linking to the newest report, and an **Output** panel summarising what this agent produces.
## Run now
**Run now** needs an active, idle agent. On the detail page the button stays visible and goes disabled with a tooltip saying why: the agent is paused (resume it first), it has errored (review the latest run first), or a run is already in progress.
One run at a time per agent is a hard rule. In the list menus the item behaves the same way. It appears for active agents and greys out while a run is in flight.
See [Run history](/agent-runs) for what a run records and how to read it.
## Recipes
Point the instructions at the pricing page, add a weekly schedule, add the page under Web pages in Tools and access, and turn on **Generate report**. Accept the drafted question and add a Slack delivery to your competitive channel.
Describe what to capture, add a daily schedule, give the agent the sources it needs, and turn on **Gain knowledge** only. No report, no delivery.
Add a workspace trigger for **Insight created** with a minimum score, turn on **Generate report**, and email the result to whoever needs to act. No schedule required.
Add a monthly schedule, turn on **Generate report** with **Report when nothing changed** on so the digest always ships, and add an Email delivery.