How It Works

Describe. Preview. Approve. Run.

You type what you want in plain English. GloriaMundo builds a structured workflow, shows you every step and the cost before anything executes, and keeps the workflow running reliably once you approve it. This page walks through each part of that loop.

Step 1

From a sentence to a workflow

You open the builder and describe the task. "Every morning, check my Gmail for invoices, log the amounts to Google Sheets, and send me a Slack summary." The architect translates that into a structured workflow: typed steps, the data flowing between them, and any conditional logic, choosing the right actions from 1,000+ integrations via Composio.

The result is not a chat transcript. It is an inspectable workflow on a visual canvas that you can edit, save, schedule with cron or webhooks, and rerun. You can adjust it by chatting ("make the Slack message more concise") or by editing steps directly.

Step 2

The Virtual Run: preview before anything executes

Before a workflow runs for real, you preview it. The Virtual Run treats operations differently by type:

  • READ Operations that fetch data execute live, so the preview works with your genuine emails, rows and issues rather than placeholders.
  • WRITE Operations that would change something (send an email, post a message, create a record) are intercepted and shown as structured previews. Nothing is sent until you explicitly approve a live run.

Alongside the step previews you get a cost breakdown for the run, shown before you approve, and a connection check that tells you which services still need to be linked. The Virtual Run itself is usually a few cents; heavier AI workflows cost a bit more.

Every run, virtual or live, is logged step by step: inputs, outputs, timings and costs, so you can see what happened during and after the run, not just at the end.

Iron Layer

Reliability Injected Automatically.

Retries, circuit breakers, and exponential backoff are built-in. You focus on the logic; we handle the infrastructure.

Automatic Retry

Exponential backoff with jitter. Failed requests retry intelligently.

Circuit Breaker

Failing services get isolated. Your workflows keep running.

Rate Limit Detection

429s get queued and retried. No manual intervention needed.

Idempotency Keys

Safe retries guaranteed. No duplicate actions, ever.

Error Recovery

Failed steps resume from where they left off. No data loss.

Coming from n8n? No more Docker containers, no more Redis setup, no more debugging retry logic at 2am.

Not coming from n8n? Even better โ€” you don't need to know what any of that is. The reliability is just there.

iron_layer.log
[10:30:01] POST /api/slack/send
[10:30:02]
429 Too Many Requests rate_limit_exceeded
[10:30:02]
IRON LAYER Rate limit detected, queueing request
[10:30:02]
retry_after: 2.5s attempt: 1/5 idempotency_key: ik_a8f3k2
[10:30:04]
RETRY Retrying with exponential backoff...
[10:30:05]
200 OK message_ts: 1704456605.123456
[10:30:05]
SUCCESS Request completed with automatic retry handling
Error absorbed by Iron Layer
Zero manual intervention required
Governance

Five layers between a proposal and an action

Every action passes through five checks by design before it executes, and every action is logged.

01

Kill Switch

An emergency stop that halts every running workflow at once.

"Emergency halt: all workflows paused"
02

Blocked Actions

Your own list of actions that must never execute, regardless of context.

"Never delete production data"
03

Constitutional Rules

Guardrails that can't be bypassed by users, by agents, or by us.

"Credentials never exposed in outputs"
04

Policy Evaluation

Configurable handling per action type. Financial, social and email actions require approval by default.

"Financial transactions require approval"
05

Approval Requirements

Actions above cost thresholds, or touching sensitive operations, queue for your explicit sign-off.

"Approval required for sends over a set amount"

See it on your own task

$10 in free credits to start, no card required.

Get Started