Building an Information Firewall for AI Agents

Published: true

Description: TrustBoost is an open source middleware that redacts PII and verifies payments on Solana for autonomous AI agents.
tags: aiagents, opensource, privacy, solana.

⚠️ UPDATE April 27, 2026 — TrustBoost v2.0 is live.
Migrated from Make.com to FastAPI + Supabase + Render (AWS).
New endpoint: https://api.trustboost.dev/sanitize
Open source server: github.com/teodorofodocrispin-cmyk/trustboost-api
All examples below have been updated to reflect v2.0.

🛡️ TrustBoost: Information Firewall for AI Agents
I've been building AI agents (local and cloud-based) and noticed a recurring problem: agents leak sensitive data.
Emails, passwords, private keys, internal hostnames — all sent directly to LLMs without any control.
So I built TrustBoost.
What It Does
TrustBoost sits between your agent and the LLM. It acts as an information firewall with three layers:
LayerFunctionPayment verificationChecks Solana Network transactions (149 USDC for 10,000 sanitizations) — no human approvalPII redactionUses GPT-4o-mini to detect and redact emails, passwords, private keys, national IDs, crypto walletsAudit loggingSaves every operation to Supabase PostgreSQL for traceability
Why Solana?
Agents transact with each other without humans. Solana is fast, cheap, and Helius API makes verification simple.
How to Test It (50 Free Requests Per Wallet)
bashcurl -X POST "https://api.trustboost.dev/sanitize" \
-H "Content-Type: application/json" \
-d '{
"tx_hash": "TRIAL",
"wallet_address": "your_wallet",
"text": "My email is [e-mail] and my AWS key is AKIAIOSFODNN7EXAMPLE"
}'
Response:
json{
"status": "success",
"data": {
"sanitized_content": "My email is [REDACTED] and my AWS key is [REDACTED]",
"safety_score": 0.85,
"risk_category": "CRITICAL",
"usage_metrics": {
"quota_remaining": 49,
"quota_limit": 50
}
}
}
Multilingual Support
TrustBoost v2.0 detects PII in 5 languages automatically:

🇺🇸 English — SSN, API keys, credit cards, passwords
🇲🇽🇨🇴 Spanish LATAM — RFC, CUIT, DNI, CURP, Cédula
🇧🇷🇵🇹 Portuguese — CPF, CNPJ, RG, NIF
🇩🇪 German — Personalausweis, Steuernummer, IBAN DE
🇯🇵 Japanese — マイナンバー, 運転免許証, 住所

For AI Agents (MoltBook Compatible)
TrustBoost exposes a molt.json schema so agents can discover and use the service autonomously.

Endpoint: https://api.trustboost.dev/sanitize
Health: https://api.trustboost.dev/health
Trial: "tx_hash": "TRIAL" — 50 free requests per wallet_address
Paid: real Solana tx_hash (149 USDC for 10,000 sanitizations)
Anti-replay: automatic via Supabase PRIMARY KEY

v2.0 Infrastructure — Open Source
TrustBoost v2.0 runs on:
ComponentTechnologyAPI FrameworkFastAPI (Python)DatabaseSupabase PostgreSQLHostingRender (AWS)Payment OracleHelius on SolanaPrivacy EngineOpenAI GPT-4o-mini
Full server code auditable at:
github.com/teodorofodocrispin-cmyk/trustboost-api
Important Limitations

Not suitable for HIPAA zero-transmission environments — text is sent to a remote API
Payments are irreversible — test with TRIAL before paying
TRIAL is trust-based — not cryptographically verified (paid mode uses on-chain verification)
This is a prototype — not a certified enterprise product

Open Source
MIT licensed. Full source code available.
🔗 GitHub: github.com/teodorofodocrispin-cmyk/TrustBoost-PII-Sanitizer
🔗 Server: github.com/teodorofodocrispin-cmyk/trustboost-api
Feedback Welcome
This is a prototype built in public.
What would make you trust this for production use?

Integration with LangChain? Firecrawl? OpenClaw?
Solana payment model — yay or nay?

Leave a comment or open an issue on GitHub.
Thanks for reading.