| What it does | An approval layer for unattended AI agents that halts risky actions for human approval before execution
| Open-source runtime for computer-use agents
|
|---|
| Stage | Not given
| Startup
|
|---|
| Founded | Not given
| 2026
|
|---|
| Based in | Not given
| Not given
|
|---|
| Pricing model | Freemium
| Open source
|
|---|
| Pricing | Self-hosted: Free (MIT licensed). Hosted Indie: $49/mo (2 seats, monthly governed-action ceiling, 30-day retention, email support). Hosted Team: $199/mo (10 seats, higher governed-action ceiling, 90-day retention, org-scoped rate limits, priority support).
| Not given
|
|---|
| Key features | - Intercepts risky agent actions before execution
- Works with Claude Code, Codex, Hermes, OpenClaw, and MCP
- Policy-based risk scoring and decision making
- Remote approvals from phone or browser
- Signed audit ledger with replayable records
- Calibrated interruptions that reduce false positives
- Enforcement liveness monitoring via synthetic probes
- Self-hosted (MIT licensed) or managed hosting options
- Approvals inbox with decision ledger and policies
- Agent defense documentation with assumptions and shields
| - Works with any OpenAI-compatible model
- Ready-wait mechanism instead of fixed sleeps
- Batches actions and bounds model calls
- Desktop pause, resume and fork with memory
- MCP server for Claude Code and Cursor
- Graded rollouts for computer-use RL
- Works with full desktops
- Supports multiple sandbox providers
|
|---|
| What makes it different | - Only approval layer that protects unattended runs that walk away
- Signed ledger with replayable audit records and explicit signature verification
- Self-hosted with no infrastructure lock-in or vendor dependency
- Learns from your approvals to reduce interruptions automatically
- Honest about enforcement boundaries across different integrations
| - 1.48x faster than Anthropic's computer-use reference loop on the same model
- 19 points more success than OSWorld's Qwen agent at same speed
- Reduces rollout time by 40-53% by eliminating unnecessary waiting
- Published benchmarks run same model on same desktops with reproducible results
- Works with any model and sandbox provider
|
| Community votes | 0 | 0 |
| DR (Domain Rating by Ahrefs) | 7 ▼ −1 (down) since last check | Not checked yet ● no change since last check |
| Trust Flow (Majestic) | 3 | 0 |
| Citation Flow (Majestic) | 30 | 0 |
| Referring domains (Majestic) | 88 | 0 |