Internal Process Automation with AI

SquadOS can assist employees through natural-language conversations and run background flows. These two journeys use different features:

Need	Correct feature
A person asks a question, requests an action, or needs a conversation	Agent, connected to an inbox and to the required knowledge bases and tools
An event, a contact or conversation change, or a scheduled time must start a sequence	Automations, under Admin → Automations

The product does not automate clicks like an RPA. An agent interprets a conversation and can use the tools linked to it. An Automation follows an explicit flow of triggers, rules, and actions.

Before you build the flow

Define the following in writing:

1. What starts the process — a conversation event, a conversation with no reply, a contact tag, a contact added to a list, an audience, or a schedule.
2. What context exists at entry — some triggers provide a conversation and contact; others start without those objects.
3. What should happen — for example, analyze a conversation, split paths, wait, apply a tag, assign support, or call an external system.
4. When the same contact may enter again — once per conversation, once per contact, or always.
5. How to verify the result — a controlled test, success criteria, and an owner for monitoring failures under Activity.

This definition prevents you from saving a flow that depends on a conversation or contact that the selected trigger does not provide.

Create the Automation

Your role needs Automation write permission to create or change flows.

1. Open Admin → Automations and select New automation.
2. Choose Start from scratch or Follow-up on a stalled conversation. The second starter already contains three spaced attempts, but you can change every step.
3. Rename New automation #N in the top bar.
4. Under Add, choose an item from the Triggers, Actions, or Rules tabs. You can click the item or drag it onto the canvas.
5. Select each step and fill in its settings. The variable picker only shows data available at that point in the path.
6. Open Settings and choose the Re-entry rule. Always creates a new run for every event, including for the same contact; AI steps consume credits on every run.
7. Select Test and inspect the path under Activity.
8. Select Save. SquadOS validates the graph and highlights invalid steps. Saving creates a new version for future runs, but does not turn the Automation on.
9. After saving at least one version, activate the Automation under Settings or with the control in the list.

Saving and activating are separate actions

While the Automation is off, its trigger creates no runs. A run that already started stays on the version it entered with; a new version applies only to future runs.

Available triggers

Trigger	When to use it
Conversation event	When a conversation changes state
No reply on conversation	When your team spoke last and nobody replied after the configured interval
Contact tag	When a tag is added or removed
Contact added to list	When a contact joins a list via the API, an import, or a manual action
Audience	To start a segment on the scheduled date
Schedule	To run once or repeatedly at the configured time

The old Webhook trigger was discontinued. To start a flow from an external system, add the contact to a list through the Lists API and use Contact added to list.

Actions, rules, and AI

The editor provides Condition, Wait, and End to control the path. Current actions can send a message or email, apply a tag to the contact or conversation, pick a contact, get the contact’s conversation, assign support, hand off to another AI agent, resolve the conversation, save a value to the contact, and call an external webhook.

Analyze with AI classifies and extracts data from the conversation. In the current version, this step does not run agent tools or search knowledge bases. If the flow must send data to another system, use Call webhook; if a person needs a conversation in which the agent searches knowledge or runs tools, use the Agent journey.

See the complete contracts under Triggers, Actions, Rules, and Variables.

Test without confusing simulation with real effects

A test can run the Current canvas, even if it has never been saved, or the Live version, when one exists. If the flow has more than one trigger, choose which one to simulate and provide the conversation, contact, or data required for that entry.

• sending actions are simulated;
• AI steps run for real and consume credits;
• Get the contact’s conversation may open a real conversation, although it sends no message;
• a step that requires a conversation or contact fails when the test has no such target.

When the test ends, open Activity. Each run shows its status, cost, input, output, and the path taken on the version it used. Read Runs and tests to interpret each state.

Two safe examples

Triage after a conversation

Use Conversation event → Analyze with AI → Condition. In each branch, save the result to the contact, apply a tag, or assign support. Test cases that should take different paths before saving and activating the flow.

Entry sent by an internal system

Create a dedicated list, use Contact added to list as the trigger, and register the contact through the API. Then add the required actions. Sending a message requires a conversation: configure Get the contact’s conversation with the correct inbox and account for the fact that this step may create a new conversation.

Agent for internal support

When the process starts with a human question or request, create an Agent, link the required knowledge bases and tools, and connect an inbox. An agent can search internal policies, answer questions, or run an HTTP/Composio tool during the conversation; this does not turn those tools into steps in the Automations editor.

• Tools and Skills
• Custom HTTP Tools
• Inboxes
• Automations overview

Start with a short flow, test expected and unexpected inputs, save, activate, and monitor Activity before extending the sequence.