# Create a background agent

> Build an agent two ways: pick a template and answer a few questions in chat, or fill in the step-by-step form yourself. Both end up as the same agent, editable in the same place.

- Source: https://spiich.ai/help/background-agents/create-a-background-agent
- Section: Background agents
- Who can do this: Everyone
- Time: 10 minutes
- Last updated: 2026-08-24

## Before you start

- You can set up background agents. Admins can by default; everyone else needs an admin to grant it to them or to their team.
- Your workspace has credits available.
- A CRM is connected if you want the agent to read or write CRM records.

There are two routes into a new agent and they produce the same thing. The guided chat is faster and writes better instructions than most people write cold. The form is better when you already know exactly what you want.

## Route one: a template and a short chat

1. **Open the template library**
   Go to **Background Agents** in the sidebar and click **Create new**. If the workspace has no agents yet, the library is already the page you land on.

2. **Pick a template**
   On the "Pick your agent." screen, click a card. The current templates are **Lead Generation**, **Buyer Discovery**, **Key Account Monitoring**, **Sales to CS Handover**, **Weekly Follow-Up Tracker** and **Custom Agent**. Each card shows what the agent does, its default schedule, three highlights and its credit price.

3. **Answer the questions**
   A chat opens and the builder asks a few sharp questions: which slice of the market, which personas, which weekday and time, what output you want. Type in the box at the bottom and press Enter, or click one of the offered choices. It can search the web to ground its suggestions.

4. **Hand it a list, if you have one**
   Click the paperclip, **Attach a company list**, or drop the file onto the page. Key Account Monitoring can resolve an uploaded account list into CRM organizations and pin the agent to exactly those records.

5. **Open the finished agent**
   When the builder is done it writes the full instructions itself and creates the agent, then shows a card with the name, schedule and record type. Click **View agent**.

6. **Check what it built**
   On the dashboard read the **Agent setup** panel (Schedule, Cost, Run scope, Created) and the **Capabilities** panel, which lists whether Email, Slack and Enrichment are enabled. Click **Configure** to change anything.

> **Note:** **ICP Scoring** and **Pipeline Review** are retired. An old link to either says "New agents can no longer use the template" and offers **Build as Custom**. Agents already built from them keep running and can still be edited.

## Route two: the step-by-step form

The form is a numbered build. A preview sentence at the top restates your configuration in plain English as you go, so you can read back what the agent will actually do before you commit.

1. **Name it**
   Go to **Background Agents** and click **Create new**, then **Custom Agent**. Type a name in **Agent name**, something you will recognise in a list of twenty, for example "Daily deal review".

2. **Choose the trigger type**
   Step 1 asks "What type of agent is this?". Pick **Event-based** to run when a CRM record changes, **Scheduled** to run on a repeating schedule, **Webhook** to run when an external service sends data, or **On-demand** to run it yourself against records you select in a table. Full detail in [Triggers and schedules](/help/background-agents/agent-triggers).

3. **Set the schedule, event or webhook**
   For a scheduled agent, set **Repeats**, the interval, the days and the time. For an event-based agent, pick the record type, whether it fires on **Created** or **Updated**, and which field change starts it. For a webhook agent, generate the webhook URL, set its rate limit and copy the signing secret.

4. **Add the conditions**
   Under "What conditions must be met?" click **Add a rule** and build a field, operator and value test. Set **Match** to `all rules` or `any rule`, and nest a group of rules if the logic needs it. The agent only runs when the rules are true.

5. **Write the instructions**
   Fill in **Objective**, **Process** and **Output**. This is the part that decides whether the agent is useful, so it gets its own section below.

6. **Set the permissions**
   Switch on **Allow email sending**, **Allow Slack messaging**, **Allow phone & email enrichment** or **Proposal mode** as needed, and pick recipients or channels underneath. See [What an agent is allowed to do](/help/background-agents/agent-permissions).

7. **Add the guardrails**
   Optionally turn on **Run once per record** and choose a cadence, and set a **Daily run limit** so a misfiring trigger cannot spend the month's credits in an afternoon.

8. **Create it**
   Click **Create Agent**. The button stays disabled with "Complete all steps to create the agent" until every required step is filled. Back on the list, read the preview sentence on the row and check the **Next run** time is what you expected.

## Writing instructions the agent can follow

Vague instructions are the most common reason an agent produces something you did not want. Three rules help.

**Put the constraints in Process, not Objective.** Objective is one sentence about the goal. Everything qualifying it belongs in Process: "Only include people in Sweden. Start by searching LinkedIn, then the company website. Skip anyone already in the CRM."

**Describe the artifact in Output.** Say what should exist when the run ends. "Add each person to the CRM with their LinkedIn, email, and a short note" is checkable. "Give me good leads" is not.

**Name your own fields.** If your CRM has a Source field or an ICP fit rating you want set, say so by name. The agent works with your object types, so it can only fill a field you point it at.

For worked instruction sets you can copy, including a full lead generation and buyer discovery sequence, read the [Background Agents setup guide](/articles/background-agents-setup-guide).

## Editing an agent later

- **Configure** on the dashboard, or the pencil icon on a row, reopens the same form. **Save Changes** stays disabled until something changes, and a "You have unsaved changes" warning appears if you try to leave.
- The speech-bubble icon opens **Edit via chat**, where you describe the change in words instead. If a builder conversation already exists it reads **Resume chat**.
- The copy icon makes your own copy of any agent you can see, running as you. The copy is named "(copy)" and starts paused.
- The trash icon deletes it.

> **Careful:** Deleting an agent permanently removes it and stops its automated workflow. This cannot be undone.

## Limits

- A non-admin can only create agents that run as themselves. Pointing an agent at another member is admin-only.
- A fixed list of records attached to a scheduled agent is capped at 500 records, cannot contain duplicates, needs a record type, and cannot be combined with filters.

## If something goes wrong

**Create Agent stays greyed out.** Something required is still empty. The hint reads "Complete all steps to create the agent"; scroll back and look for a step with no answer, most often a missing condition ("Add at least one condition") or an empty instruction box.

**The builder chat says "Agent unavailable".** Send another message. The conversation is preserved, and nothing was created.

**The buttons that create or edit an agent are missing.** You do not have the grant. You can still read agents, review runs and approve staged changes. Ask an admin to switch **Agents** to `Allowed` for you in [`Settings > User Management`](https://app.spiich.ai/settings/user-management).

## Frequently asked questions

### Can I start from a template and then change everything?

Yes. A template only seeds the conversation, the default schedule and the price. Once the agent exists, **Configure** opens the same form a custom agent uses and every field is editable, including the trigger type.

### What happens to the answers I gave the builder?

They are kept on the agent dashboard under **Setup answers**, with a **View setup conversation** link back to the chat. That is the fastest way to remember why an agent is scoped the way it is six weeks later.

### Do I need a CRM connected to create an agent?

Not for every agent, but for most useful ones. Event-based triggers need a CRM because they fire on CRM record changes, and any agent that writes records needs somewhere to write them. A prompt-only scheduled agent that just researches and emails you can run without one.

### Can I ask Spiich to build the agent from Chat?

Yes, if you are allowed to set up agents. Describe what you want in normal Chat and the assistant can create or update the agent for you. It can also list your agents and their runs for anyone, regardless of the grant.

## Related

- [What background agents are](https://spiich.ai/help/background-agents/background-agents-overview)
- [Triggers and schedules](https://spiich.ai/help/background-agents/agent-triggers)
- [What an agent is allowed to do](https://spiich.ai/help/background-agents/agent-permissions)
- [Read and manage agent runs](https://spiich.ai/help/background-agents/monitor-agent-runs)
- [Background Agents: Setup Guide](https://spiich.ai/articles/background-agents-setup-guide)
- [Background Agents](https://spiich.ai/product/background-agents)