# Preloop — Control any AI agent

> Human approval layer for autonomous AI agents, ensuring control and compliance. (iPhone/iPad app by Spacecode AI Inc.)

- Source: https://appshunter.io/ios/app/preloop/id6757803021 (this page in markdown: same URL + `.md`)
- Developer: [Spacecode AI Inc](https://appshunter.io/developer/1854261683)
- Category: Developer Tools, Utilities
- Price: Free
- Rating: 5.00/5 from 1 App Store ratings
- Age rating: 4+
- Requires: iOS 18.2 · 4 MB
- Languages: American English
- Released: 2026-01-16
- Data updated: 2026-08-20
- Monetization: free
- User reviews in markdown: https://appshunter.io/ios/app/preloop/id6757803021/reviews.md

## What is Preloop?

Preloop is the human approval layer for autonomous AI agents. 

When your agents attempt risky actions like deployments, refunds, or data changes, Preloop intercepts the request and routes it to you for approval right on your iPhone or Apple Watch. The right people in your organization get instant notifications with full context, approve or decline with one tap, while your organization maintains a complete audit trail for compliance.

Preloop enables event-driven automation through Agentic Flows - let AI agents handle routine work automatically while you control critical decisions. Setup takes 3 minutes as an MCP proxy, works with Claude and Cursor, and requires no infrastructure changes. 

Risk-based policies ensure only high-stakes actions need approval while low-risk operations run automatically. AI agents can't take legal or moral responsibility for their actions - Preloop ensures humans stay in control. 

Perfect for engineering teams, DevOps, and anyone deploying autonomous AI in production. Enterprise-ready, designed for SOC2 compliance, role-based access, and secure end-to-end encryption.

## Key features

- Human approval for AI actions
- Instant notifications with context
- One-tap approve/decline
- Audit trail for compliance
- Risk-based policy enforcement
- Works with Claude and Cursor


## Frequently asked questions about Preloop

### What is Preloop?

Preloop is a human approval layer designed for autonomous AI agents. It intercepts potentially risky AI actions, such as deployments or data changes, and routes them to you for review and approval directly on your iPhone or Apple Watch.

### How does Preloop ensure compliance?

Preloop maintains a complete audit trail of all approved or declined actions, which is crucial for compliance. It also allows for risk-based policies, ensuring that only high-stakes actions require human approval, while low-risk operations can run automatically.

### What AI agents does Preloop support?

Preloop currently works with popular AI agents like Claude and Cursor. Setup is designed to be quick, taking approximately 3 minutes, and requires no infrastructure changes on your part.

### Can I approve or decline actions on my Apple Watch?

Yes, Preloop is designed to provide instant notifications and allow you to approve or decline AI-driven requests with a single tap directly from your iPhone or Apple Watch.

### Does Preloop have ads?

No, Preloop does not contain advertisements. It is a free application focused on providing a secure governance layer for AI agents.

### What is the age rating for Preloop?

Preloop has an age rating of 4+, making it suitable for a wide range of users. It is available on iPhone, iPad, and Apple Watch.

### How often is Preloop updated?

The latest version of Preloop is 1.5, and it was last updated on January 24, 2026. This indicates a recent update, suggesting ongoing development and support.

### What makes Preloop enterprise-ready?

Preloop is enterprise-ready with features like SOC2 compliance, role-based access control, and secure end-to-end encryption. It ensures that humans remain in control of critical decisions made by AI agents.

## Version history (last 3 releases)

### 1.6 — 2026-07-15

### Added

- **`ask_user` questions on approval detail**: Question-style approval requests render option buttons and an optional free-text answer field. Answers submit `selected_option` / `answer_text` through the existing decision endpoints (server precedence: answer text → selected option → comment). Dismiss maps to decline.
- **Standalone Apple Watch answering**: The watch now answers `ask_user` questions on its own — one button per option (scrollable, so more than two options work), plus a free-text field that uses watchOS dictation/Scribble/keyboard when the agent allows free text. Answers go straight to the backend over the watch's direct API (phone relay is only a fallback and now carries the option/answer too). No iPhone required.
- **Listen (text-to-speech) on the watch**: A Listen button on any request reads the summary, question, and option labels aloud through the connected audio route (AirPods or watch speaker) using `AVSpeechSynthesizer`. Opt-in per tap — notifications never auto-speak — and it degrades to a visible hint instead of failing when no audio route can be activated.
- **`ask_user` questions on the main iPhone card**: The swipeable request stack now answers questions inline — one button per option, a free-text answer field when the agent allows it, and Dismiss (maps to decline) — instead of the generic Approve/Decline bar. A shared `QuestionAnswerView` component backs both the card and the detail view.
- **Dictation on iPhone question answers**: The free-text answer field has a mic button that transcribes speech on device. The speech engine was extracted into `SpeechDictationService` and is shared with agent voice sessions; permission denials degrade to a typed answer.
- **Answer questions from the iPhone notification**: New static question categories (`QUESTION_0_OPTIONS` … `QUESTION_4_OPTIONS`) expose one action per option plus a text-input "Answer" action (typed or dictated, works on the lock screen and mirrors to the watch). Option labels are read from the push payload (`question_options`, top level or nested under `data`) and submitted as `selected_option`; pushes without the question fields behave exactly as before.

### Changed

- Apple Watch request views are summary-first: the backend summary is the glance line, with the full question or tool arguments below. Question requests no longer show Approve/Decline; they route to the answering surface everywhere (main view, detail, details sheet, confirm view) and are flagged with a question icon in the list.
- Apple Watch semantic colors now follow DESIGN.md (success `#26D962`, pending `#F2A93B`, denied `#FF5D5D`, info `#30C9E8`).
- Approval cards and detail views on iPhone and Apple Watch now prefer backend-provided approval summaries as the primary request text, with the tool name shown as secondary context.
- Renamed API models from **ApprovalPolicy** to **ApprovalWorkflow** (`ApprovalWorkflowResponse`, `approval_workflow_id`) with backward-compatible decoding for legacy `approval_policy_id` payloads in approval requests, tool configurations, and WebSocket events.

### 1.5 — 2026-01-24

Websocket connectivity improvements. Smoother animation when approving / declining requests.

### 1.4 — 2026-01-16

No release notes.

## Apps similar to Preloop

| App | Rating | Price | Category |
| --- | --- | --- | --- |
| [On Device : Local AI Studio](https://appshunter.io/ios/app/on-device-local-ai-studio/id6770114399) | 5.0 (1) | Free | Developer Tools |
| [MCP Vibes](https://appshunter.io/ios/app/mcp-vibes/id6749172615) | — | Free | Developer Tools |
| [BuildFeed](https://appshunter.io/ios/app/buildfeed/id6761027122) | — | Free | Developer Tools |
| [Debug Anywhere](https://appshunter.io/ios/app/debug-anywhere/id1555924269) | 4.6 (54) | Free | Developer Tools |
| [Incident Response (Squadcast)](https://appshunter.io/ios/app/incident-response-squadcast/id1501689101) | 3.2 (6) | Free | Developer Tools |
| [Talon – Webhooks & Links](https://appshunter.io/ios/app/talon-webhooks-and-links/id1492913323) | 4.6 (29) | Free | Developer Tools |

---

*Data collected daily from the US App Store and indexed by [AppsHunter](https://appshunter.io/). User reviews are verbatim App Store reviews. Ratings, prices and chart positions refresh continuously; this snapshot is from 2026-08-20.*