Tajo agents

In short: these are the nine Tajo agents that sync work runs through. One builds a sync, one runs it, seven work on the contacts and orders it lands. None of them can publish a sync or send a message without your approval.

Each card below has the same five parts: the job in one line, what the agent works from, what it never does, what waits for you, and a first request that works today. For how the agents a sync needs fit together, read Tajo 101.

Brevo Integration Architect

Its job: turns a source system into a sync rule that writes to Brevo or HubSpot.

• Works from: a few real records it samples from your connected source, and the catalogue of Brevo endpoints.
• Never does: writes raw API calls for production data. Every sync it builds is a declared rule that maps fields to a typed action, such as upserting a contact or tracking an event.
• Needs your approval: publishing the rule. It validates the field map, dry-runs the rule without writing anything, saves a draft, then asks. The request expires after 30 minutes, and it activates only with the approval you gave.
• Getting started: “Build a sync from Shopify customers to Brevo contacts. Show me the field map and a dry run before you save the draft.”

Sync Operator

Its job: runs your active sync rules and reports what was pulled, loaded and refused.

• Works from: the rules you already approved.
• Never does: changes a rule, or runs one you have not published.
• Needs your approval: nothing new. The rule was approved when it went live. Tajo also runs each active contact sync on its own, at most every 15 minutes.
• Getting started: “Run the Shopify customers sync now and tell me how many records loaded and how many failed.”

Segment Builder

Its job: turns a description of who you want into a contact list.

• Works from: synced contacts and each contact’s consent per channel.
• Never does: sends anything. A list on its own reaches nobody.
• Needs your approval: nothing. The send that uses the list is where approval happens.
• Getting started: “Make a list of contacts who opted in to email.”

Compliance Reviewer

Its job: checks a draft before it goes out: consent, unsubscribe link, sender domain, content red flags.

• Works from: the draft and the channel it is meant for.
• Never does: approves a send for you. When a check fails, it lists what failed and stops.
• Needs your approval: a send that still has to go out despite a warning goes to you as an approval request.
• Getting started: “Check this draft for an unsubscribe link and consent problems before I approve it.”

Engagement Designer

Its job: drafts email, SMS and WhatsApp messages and reusable templates.

• Works from: your templates and campaign drafts.
• Never does: sends. Scheduling a send is not one of its tools, so every draft waits for review.
• Needs your approval: anything risky it flags, such as a large audience or a promotional claim.
• Getting started: “Draft a welcome email for contacts who opted in to email this month.”

Performance Analyst

Its job: reads open, click and unsubscribe rates, flow completion and segment health over up to 90 days, and writes recommendations that name the segment, flow or campaign.

• Works from: your workspace’s engagement results.
• Never does: sends, or edits a flow or campaign itself. It writes a fix up as a campaign draft or an approval request.
• Needs your approval: any send made from its drafts.
• Getting started: “Look at the last 30 days of email results and tell me what to change.”

Orchestrator

Its job: takes a free-form request and chains the other tools to finish it in one run.

• Works from: everything above.
• Never does: skips a gate. It and Win-back are the only two that can schedule a send, and that tool always stops for your approval.
• Needs your approval: every send, and any send to 50 or more contacts regardless of who asked. Marketing sends also need an opt-in on the channel they use: an email opt-in never covers SMS.
• Getting started: on the home screen, “Every Monday, review last week’s email results and tell me what to change.” A request with a schedule in it becomes a routine.

Win-back

Its job: finds customers who ordered before but not lately, sizes that audience and drafts a win-back campaign for it.

• Works from: order events synced from your store (order.placed) and your contacts’ consent per channel.
• Never does: creates a discount code; it has no tool for one. On a scheduled run it only drafts and never schedules a send.
• Needs your approval: every send. Scheduling one waits for a person, and the consent gate drops anyone not opted in to the channel. Its weekly Monday routine is created paused and runs only after you resume it.
• Getting started: “Find customers who ordered in the last year but not in the last 90 days, and draft an email win-back campaign for the ones who opted in to email.”

Customer Intelligence

Its job: scores customers by recency, frequency and order value and reports who is buying, slowing down or gone quiet.

• Works from: order events synced from your store (order.placed). Every report opens with how far back that order history goes.
• Never does: changes a contact, a segment or a connected app: every tool it holds reads. Nor does it predict churn with a model: overdue is a fixed rule, more than twice the customer’s own average gap between orders.
• Needs your approval: nothing. It only reads. Its weekly Monday routine is created paused and runs only after you resume it.
• Getting started: “Give me the customer report from the last 365 days of orders.”

Your own agents

You can hire another agent in the app under Team. You give it a name, a role, instructions, and pick its tools and skills from a fixed list. The tools on that list are: request human approval, compute a segment, draft a template, and schedule a send. Scheduling a send keeps its approval gate whichever agent holds it.

There is no file format for agents and no way to hand an agent arbitrary API tools. Routines run on a schedule. Routines started by an event, such as a new order, are not built yet.

What the old agents page listed

An earlier version of this page listed agents that Tajo does not run. Where they stand:

• Campaign Orchestrator is the Orchestrator and the Engagement Designer above. They draft campaigns; they do not launch them.
• Data Sync Agent is the Brevo Integration Architect and the Sync Operator. Each rule moves data one way, from a source into Brevo or HubSpot.
• Custom Agents are the agents you hire under Team, from the fixed list of tools above.
• Cart Recovery is not built. Tajo does not receive abandoned checkouts from Shopify today, so there is no data for it to work from. The recovery rate the old page quoted had no measurement behind it and is gone.
• Win-Back and Customer Intelligence are now in Tajo, as the Win-back and Customer Intelligence cards above.

Get access

Tajo workspaces are not open for self-serve sign-up yet. Sign up for early access with your first name and work email: you confirm by email, then we send access details when a workspace can be opened for you. If you run a Shopify store, start with Tajo for Shopify stores.