| What it does | Policy-driven spending controls for AI agents
| An AI panel debates your goal; a referee rules on re-checkable evidence
|
|---|
| Stage | Not given
| Not given
|
|---|
| Founded | Not given
| Not given
|
|---|
| Based in | Not given
| Not given
|
|---|
| Pricing model | Freemium
| Freemium
|
|---|
| Pricing | Free: $0/month (open source, self-hosted). Pro: $24/month with hosted dashboard, 3 agents, 1,000 spend checks/month. Team: $99/month with 10 agents, 10,000 checks/month, team approvals. Business: $299/month with unlimited agents and checks.
| Free: $1.00 starting credit once. Credit plans from $10 top-up with no expiration. Your own key: 5% infrastructure fee after the first $5.00 of usage.
|
|---|
| Key features | - Policy Engine with YAML configuration for amount limits, merchant allowlists, category blocks, business hours, and duplicate windows
- Duplicate Detection using SQLite to prevent the same purchase twice within a configurable time window
- Human-in-the-Loop approval with Telegram alerts and dashboard queue for purchases exceeding threshold
- MCP-Native integration for OpenClaw, Hermes, and any MCP-compatible agent
- Zero Credential Exposure with no access to cards, tokens, or Stripe credentials
- Local-first architecture with all policy files, logs, and databases stored on your machine
- Dry-run and simulation mode for testing policies before deployment
| - Panel of up to six AI models from different vendors
- Referee rules on re-checkable evidence only
- Searches academic databases: arXiv, OpenAlex, Crossref, Europe PMC
- Executes Python code to verify claims
- Customizable acceptance standards
- Matches pause and resume without losing work
- Shared ledger prevents re-derivation
- Spending limit per match
|
|---|
| What makes it different | - Policy-driven rules that the agent cannot override, unlike prompt-based guardrails
- Zero credential exposure means the gateway never sees payment cards or API keys
- Duplicate detection with SQLite prevents duplicate purchases within a configurable time window
- Native MCP integration works out of the box with OpenClaw, Hermes, Claude Code, and Cursor
| - Referee decides on evidence, not argument quality
- Panel from different vendors fails in different directions
- Stuck is a targeted question, not model thrashing
- Cost grows with ledger, not round count
- Published matches become research record
|
|---|
| Community votes | 0 | 0 |
|---|
| DR (Domain Rating by Ahrefs) | 23 ▲ +1 (up) since last check | 2 ● no change since last check |
|---|
| Trust Flow (Majestic) | 6 | 0 |
|---|
| Citation Flow (Majestic) | 19 | 13 |
|---|
| Referring domains (Majestic) | 12 | 31 |
|---|
ActOnce Spend Gateway
Referee.chat