ViduraNīti
A spending governance platform for autonomous AI agents: budgets, ML risk checks, human approval and verified x402 + Algorand settlement.
An organization defines its agents, approved services, budgets and policies. Agents run company tasks, but never touch money directly: every payment passes deterministic policy, ML anomaly detection, human approval where needed, and the mandatory x402 and Algorand settlement workflow, with a ledger recording each decision.
Proves
- FastAPI backends
- React + TypeScript
- State machines
- Payments safety
- ML in the loop
Decisions that carry it
- Agents never hold the wallet. Every payment goes through the spending controls and the mandatory x402 path. Policies are deterministic; the optional Groq model only interprets tasks and never decides money.
- Layered decisions. Budget and policy first, ML anomaly scoring second, a human third when needed. Approval never bypasses revalidation of policy and budget.
- Settlement is a state machine. Task, approval and payment lifecycles are separate. An accepted x402 response and a confirmed Algorand transaction are different milestones; nothing is SETTLED until confirmation is verified.
- Idempotent and auditable. Idempotency keys, immutable decision records and atomic budget reservation make retries, duplicates and reconciliation safe.
- Fast without trusting the cache. TanStack Query, IndexedDB and Redis keep the interface quick, while PostgreSQL stays the source of truth for approvals, payments and the ledger.
Built with
- FastAPI
- React + TypeScript + Vite
- Supabase PostgreSQL
- Redis
- x402
- Algorand
- Cloudinary
- TanStack Query
- Groq (optional)