A Claude Desktop extension and plugin that convenes a council with ChatGPT and Gemini for independent AI positions, votes, and recorded dissent
AI agents for specific business functions that rehearse tasks as test runs before execution
Stage
Not given
Not given
Founded
Not given
Not given
Based in
DE
DE
Pricing model
Paid
Paid
Pricing
One-time purchase for Claude Desktop and Claude Code. Usage costs paid directly to OpenAI and Google: $0.02–$0.15 per council decision depending on effort tier and input size; image generation $0.05–$0.20 per call.
Roles range from € 99 to € 499 per month excluding VAT, with included monthly tasks varying by role: Administrative Assistant and Association Secretary at € 99/mo (40-45 tasks); Secretary at € 109/mo (50 tasks); Process Documentation and Planner at € 129/mo (15-40 tasks); Customer Service at € 229/m
Key features
Three independent AI positions on questions
Majority voting with dissent recorded
Budget guard with daily spending limit
Integrates ChatGPT, Gemini, and Claude
Runs on Claude Desktop (macOS, Windows) and Claude Code
Uses own API keys with secure OS storage
Nine specialized tools for different decision types
Two-round cap and domain tie-break for consensus
Failure handling and 345 tests
Plain-language interface requiring no developer skills
Pre-built roles from a catalogue (31 roles)
Fingerprint verification on approval (SHA-256)
Full audit trail of proposals, test runs and approvals
Integration with 12+ pre-connected services
Custom REST API and MCP server connections
Servers in Germany within EU
AI Act compliance file (limited risk classification)
What makes it different
Three independent AI models vote together with dissent kept on record, avoiding single-model bias
Budget guard prevents runaway bills with daily ceiling and worst-case cost reserving
Uses your own API keys stored securely locally, no server or data transmission
One-time purchase model with no subscription or ongoing fees
Works directly in Claude without developer skills
Test runs show actual outcomes with no side effects before real execution
No automatic approvals—every proposal requires explicit human approval
Approved test run and execution count together as one task, not two
Declined proposals are free and don't count against quota
Kirat
Stilops