A standalone AI Agent is an automation that thinks. Instead of following a fixed flow, you describe what you want in plain language, choose the events that should trigger it and the actions it's allowed to take, and the agent works out the rest on its own β reading the event, deciding which steps to run, and calling your integrations to get the job done. Manage them at https://v2.egrow.com/ai-agents/standalone.
Note: this is different from the inbox AI Agent. The inbox agent chats with your customers in conversations. A standalone agent runs behind the scenes: an event fires, the agent runs once, does its work, and stops β like an automation, but powered by AI reasoning instead of fixed nodes.
When to use a standalone agent
Reach for one whenever a task needs judgment that a rigid automation can't express β for example:
- "When an order is confirmed, ship it via the cheapest carrier that covers the city and notify the customer on WhatsApp."
- "When a new WhatsApp message arrives, read it, tag the conversation, and if it's a complaint, alert the team."
- "When an order reaches Delivered, add a follow-up note and record the payout."
You write the intent once; the agent handles the variations.
Before you begin
- At least one connected integration the agent will act through (WhatsApp, a carrier, Shopify, Google Sheets, etc.). Connect them under https://v2.egrow.com/integrations.
- An idea of the event that should start the agent (a new message, a stage change, an incoming webhook).
- Access to the AI Agents page: https://v2.egrow.com/ai-agents/standalone.
The agents list
The standalone page lists your agents in a table showing each agent's name, its AI model, the integrations it can use, its webhook triggers, and an enabled switch. Toggle an agent off to pause it instantly; click its history icon to see past runs, or the trash icon to delete it. Click Add Agent (top right) to open the builder β the same panel opens when you click an existing agent to edit it.
Step 1: Name it and pick a model
- Agent Name (required) β a unique name, e.g.
Ship & Notify. - AI Model (required) β the model that powers the agent's reasoning. Each shows its provider; pick one and you can change it later.
Step 2: Write the Instructions
The Instructions are the heart of a standalone agent β a plain-language description of what it should do when it runs. Be specific about the goal and any rules, for example:
"When an order is confirmed, ship it via Ameex and notify the customer on WhatsApp. Never ship to a blacklisted number. If the city isn't covered, add a note and leave the order for the team."
The agent reads the triggering event plus these instructions, then decides which of its granted actions to run.
Enhance with AI
Not sure how to phrase it? Click Enhance with AI. eGrow rewrites your draft into a clearer, more complete prompt. You can add a short note telling it what to change (for example, "keep the phone number visible, don't add a tone section"), then review the proposal and either Use prompt, regenerate, or discard it. Your original is never overwritten until you accept.
Step 3: Grant Integrations Access
Open Integrations Access and tick the connected platforms this agent may interact with β the WhatsApp number, the carrier, the store, and so on. The agent can only touch integrations you grant here. When you've written instructions, eGrow recommends the connections it thinks the task needs and marks them with a green Recommended badge, so you can grant the right ones in a click.
Step 4: Choose Webhook Triggers
Open Webhook Triggers and select the events that start the agent β for example New WhatsApp message received, or an order-stage change. Each time a selected event fires, the agent runs once against that event's data. As with integrations, relevant triggers are surfaced with a Recommended badge based on your instructions. Search to find a specific event if you have many.
Step 5: Grant Modules (the agent's allowed actions)
Open Granted Modules to choose exactly which actions the agent may perform β send a WhatsApp message, create a parcel, add an order note, append a row to a sheet, and so on. This is your safety boundary: only the modules you grant are exposed to the AI, so an agent can never take an action you didn't allow.
Important: modules come from the integrations you granted in Step 3 β pick your integrations first, then the module list fills in. Recommended modules for your instructions are badged, but you stay in control of the final list. Grant the fewest modules the task needs; fewer actions means more predictable behavior.
Step 6: Save and enable
Click the primary save button at the top of the panel. Back in the list, use the enabled switch to turn the agent on. Once enabled, it listens for its webhook triggers and runs automatically whenever one fires.
Reviewing runs (history)
Every time an agent runs, that run is recorded. Open an agent's history from the list to review past runs β what triggered them and what the agent did β so you can confirm it's behaving correctly and refine the instructions, integrations, or modules if needed.
Tip: start with a narrow agent β one trigger, a couple of modules, a tightly written instruction β enable it, watch a few runs in the history, then widen its scope once you trust it.
Standalone vs. inbox agent β which one?
- Use the inbox agent when you want AI to hold a conversation with customers β answering, selling, taking orders in chat.
- Use a standalone agent when you want AI to react to an event and perform actions across your integrations, without a conversation β the smart alternative to a fixed automation.
What's next
Compare with rule-based flows in How eGrow Automations Work, connect more tools under Integrations, and give your customer-facing assistant a home in the inbox AI Agent.