Coming soon
Sentinelas icon

Sentinelas

Autonomous agents on duty

Explore the full Library and run up to two sentinels for free. Each one has its own manual, schedule, queue, and settings.

Access is not available yet · We will let you know by email

Pro
Search
All15
Attention
Recent3
Running
Catalog
Groups
Test2
Flutter4
Design System1
React5
Simulators
Component Library2
No group1
New group
Test2
Discord Test: pipeline smoke-test sentinel
on 7/23 at 11pm · Kimi
Prove the pipeline end to end (cron → run-sentinela → claude → discord-send): sends a short status message to Discord on every run.
448e7e11
Approval Test: human-gate loop with Discord send
Mondays at 9am · Kimi
Demonstrate the human approval loop, executable: on each run the sentinel drafts a short phrase and asks whether it may be sent to Discord.
5b189ce5
Pro
Search
All15
Attention
Recent3
Running
Catalog
Groups
Test2
Flutter4
Design System1
React5
Simulators
Component Library2
No group1
New group
Test2
Discord Test: pipeline smoke-test sentinel
on 7/23 at 11pm · Kimi
Prove the pipeline end to end (cron → run-sentinela → claude → discord-send): sends a short status message to Discord on every run.
448e7e11
Approval Test: human-gate loop with Discord send
Mondays at 9am · Kimi
Demonstrate the human approval loop, executable: on each run the sentinel drafts a short phrase and asks whether it may be sent to Discord.
5b189ce5

Multi-CLI Support

Use Claude Code, Kimi, Codex, or another supported agent CLI for each sentinel.

Recurring Schedules

Simple cron to run audits and checks as often as you want.

Automatic Audits

Sentinels scan code, dependencies and configs and deliver a concrete report.

100% Local

Sentinelas uses cron on your Mac. Your code and credentials stay on the machine.

You're in Control

Pause, resume, run now or edit each sentinel's manual and schedule anytime.

Lives in the Notch

Closed, it shows fleet status, alerts and sneak peeks; hover or click and the full panel drops down.

Human Approval Gate

Sentinels pause before sensitive actions. Approve or reject with a comment, and the same run resumes where it stopped.

Suggests Sentinels Automatically

It reads session metadata and proposes sentinels for repeated work. Every proposal starts paused and waits for your approval. Pro.

Catalog + Skills Hub

Explore and preview the entire Library for free, install up to two sentinels, and discover thousands of Agent Skills on skills.sh. The complete Skills Hub is also included in Free.

Resume Any Session

Every run records the CLI resume command. One click opens Terminal in the right folder, continuing the conversation.

Parallel Multi-CLI Execution

Split a queue of independent tasks across multiple CLIs and models. SQLite coordinates the rounds, and a configurable interval staggers each worker's start.

Version History

Every edit to the manual or memory becomes a snapshot. Restore versions from the panel; the CLI remains open to everyone and preserves the current state before rolling back.

Real-world use cases

Twenty real examples from dozens of ready-made sentinels, spanning revenue protection and security through quality and daily operations. The complete catalog has manuals in Portuguese, English, and Spanish.

Pro
Search
All1
Attention
Recent3
Running
Catalog
Groups
Test2
Flutter4
Design System1
React5
Simulators
Component Library2
All1
critical-journey-watchdog
every day at 6am · claude
Continuously test the product's essential journeys with synthetic data and reproducible evidence.
sentinel
Search the manual
CLIClaude
Timeevery day at 6am
Last✓ today
Executions42
Failures2
Duration8min 14s (average)
Critical Journey Watchdog: Catch checkout failures before customers do
Status: ~/sentinelas/critical-journey-watchdog/ | Cron: 0 6 * * *
It continuously runs the journeys that keep the product alive, using synthetic data and leaving reproducible evidence.
Objective
Discover the journeys that sustain the product and run them continuously with evidence, catching real regressions without relying on a hand-maintained route list.
Scope (what it does)
  • Derives journeys from the router, navigation, E2E tests, documentation and the domain model.
  • Starts a safe environment, creates synthetic data and checks the observable outcome of each journey.
  • Captures sanitized traces and screenshots, then repeats failures once in a clean context.
What it does NOT do
  • Never uses production, human accounts, real payments, messages, uploads or destructive actions.
  • Never weakens assertions, updates baselines or fixes the product to force a passing result.
Procedure (one run)
  1. Discover the stack, boot command, fixtures, routes and existing tests.
  2. Build a sourced inventory and prioritize authentication and the product's central object.
  3. Run up to five journeys, capture evidence and repeat each failure once.
  4. Classify the result, merge it into state and notify only new fingerprints or status changes.
State (state.json)
{"lastRunAt":null,"inProgress":null,"queue":[],"completed":[],"lastResult":{"summary":"","url":""}}
Stop condition
Stops after five journeys, 30 minutes, an empty queue or any sign of production or a real-world side effect.
01 / 20

Catch checkout failures before customers do

Tests signup, login, checkout, and other critical flows in a safe environment. If something fails, it delivers reproduction steps, a trace, and a screenshot before customers encounter the problem.

Frequently asked questions

Is Sentinelas only for developers?

Sentinelas is for people who work with software and already use AI agents, including developers, designers, technical product managers, founders, and operations teams. You do not need to build an agent from scratch, but you should review what each sentinel may do.

How technical do I need to be?

You do not need to know cron or write shell scripts. Start from the built-in library or create a sentinel with your agent CLI, review its Markdown manual, choose a schedule and run a test. You need macOS 14 or later and at least one supported agent CLI installed and signed in.

Which AI agents does it work with?

Sentinelas supports Claude Code, Kimi, Codex, GitHub Copilot, Qwen Code, Gemini CLI and Factory Droid. You choose the CLI and model for each sentinel and continue using the provider's own authentication and account.

What can I automate with a sentinel?

Any recurring, well-defined task that produces a concrete result: code audits, PR reviews, dependency checks, design-system validation, reports, notifications and project maintenance. Every sentinel has an explicit scope and stop condition in its manual.

Will sentinels run when the app is closed?

Yes. They run through macOS cron, so the panel does not need to remain open. Your Mac must be awake when the run is scheduled; sleeping Macs do not execute cron jobs. Sentinelas includes an optional Keep Awake control.

Can a sentinel take sensitive actions without me?

A sentinel can perform only what its manual and CLI permissions allow. For sensitive steps, its manual can require the native human-approval gate. You review the request, approve or reject it with a comment, and the same agent session resumes where it stopped.

Where do my code, prompts and logs go?

Manuals, state, queues and logs stay on your Mac. Sentinelas does not send their operational content to Viralizai. The agent CLI you choose may send context to its model provider according to that provider's terms.

Where can I read the Privacy Policy and Terms of Use?

Both documents are available before purchase. The Privacy Policy explains local data, analytics, licensing and third parties; the Terms of Use cover the license, eligibility, responsibilities, updates and refunds.

Privacy PolicyTerms of Use