Welcome credit: $2 (personal) or $10 (Instinct). Prepaid packs: $14, $40, or $60 USD. Pay-per-call pricing varies by service: from <0.1¢ for LLM answers, ~1¢ for web search and scraping, 10-30¢ for deep research, from 5¢ for lead enrichment, 25¢ for company signals, from 33¢ for image and media gene
Key features
Continuous model training from user interactions
Automatic quality improvement
Reduced latency and cost over time
User-owned learning flywheel
Buy and hold tokenized stocks at 100% loan-to-value
Access hundreds of services via single balance
Per-call price disclosure before execution
MCP, TypeScript SDK, and Python SDK integration
Per-agent spending limits and controls
GitHub-based credit line with scoring
Access to LLMs, search, scraping, compute, storage, and media services
Holdings remain invested while credit is used
Weekly or on-demand billing via card on file
What makes it different
Models learn inside the harness rather than remaining frozen
Eliminates the limitation of frozen models with notepads
First credit line backed by tokenized stock holdings at 1:1 LTV
One unified balance for all agent services and payments
No subscription or monthly minimum; pure pay-per-call model
Price transparency with exact cost shown before each call
Holdings never automatically sold; credit is separate repayment obligation