AI Agent Workflows

AI Agents

Test a controlled AI Agent path from discovery to results without giving it unrestricted control. Global rollout controls may pause protected operations while Mailrith validates the complete workflow.

  • Connect OpenAI, Claude, Codex, n8n, Pipedream, or another standards-based client without changing Mailrith's resource model.
  • Start read-only and add only the Permissions a task needs before it can run higher-impact operations.
  • Use request IDs, idempotency safeguards, and current resource state to verify results safely.
Mailrith Broadcast preflight showing the checks required before a send.

Practical Use

What This Feature Helps You Do

See the everyday task this feature supports and the small decisions it helps you avoid handling by hand.

Reporting And Research

Let a read-only agent inspect workspace capabilities, Subscribers, campaigns, and progress without granting write or send access.

Drafting And Synchronization

Create Broadcast drafts or synchronize Subscriber data with typed inputs, bounded pagination, request IDs, and idempotency safeguards.

Scoped Operations

Grant only the resource Permissions and Work Profile that the workflow needs.

When To Use This Feature

  • Use AI Agents when another application or agent runtime needs to inspect or operate Mailrith on a user's behalf.
  • Use AI Agents when you need repeatable email marketing work across reporting, Subscriber sync, drafting, execution, and progress checks.
  • Use a dedicated workspace with no real Subscribers when first testing a workflow with writes or external effects.

When This May Not Be The First Thing To Set Up

  • Use AI Writing instead when you only want help writing copy inside a Mailrith editor.
  • Do not connect an agent before you can describe its task and choose the smallest matching Work Profile.
  • Do not use a live workspace for early workflow experiments.

Setup Steps

Simple Setup Steps

Follow these steps before you add more complex setup or automation.

  1. 1

    Choose the task: read-only reporting, Subscriber sync, Broadcasts, Sequences, or another focused resource workflow.

  2. 2

    Connect through OAuth when a user is authorizing an agent client, or use a task-specific API key only for a trusted server-side runtime.

  3. 3

    Run capability discovery and expose only the tools required for that task.

  4. 4

    Test the workflow in a dedicated workspace with no real Subscribers, then inspect every request ID and result.

  5. 5

    For mutating work, grant only the required Permission, run any documented preflight check, and read the changed resource or progress endpoint to confirm the result.

Avoidable Problems

Common Mistakes To Avoid

Watch for these mistakes because they can make basic email setup and sending harder to troubleshoot.

Granting Broad Access First

Begin with read-only tools and expand the connection only when a tested workflow has a clear need for another permission.

Granting One Credential Too Much

Use separate task-specific credentials so a reporting workflow does not also receive resource write access or Full Email Marketing Access.

Retrying An Uncertain Result

Use the request ID to inspect the current resource or progress state. Do not repeat a mutation until you know whether the first attempt changed anything.

Works With Other Features

How This Connects With The Rest Of Mailrith

Email work often uses several Mailrith screens. Use this section to see where AI Agents connects to related Mailrith features.

  • Subscribers, tags, and custom fields support reporting and synchronization tasks.
  • Broadcasts, sequences, and automations support draft-first campaign workflows.
  • Request IDs and progress endpoints help reconcile uncertain results.
  • Email delivery connections and preflight checks remain the only path to provider delivery.