reddit · SaaS · 09-16 13:50 59 分 ← 返回列表

Built an open-source PII masking proxy for LLM APIs, trying to validate if there's a B2B market for a hosted version

I'm a final-year CS student. Over the last few months I built AegisFlow - an open-source proxy that sits between your app and OpenAI/Claude, strips PII from prompts before they leave your server, and rehydrates the real values in the response. SHA-256 signed audit log included. The technical stack: Fastify gateway → Presidio + DistilBERT NER for masking → Redis for ephemeral token maps → async Kafka audit trail → PostgreSQL. Benchmarked at 4ms p95 internal overhead at 200 RPS. GitHub: ht

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：中小 AI 团队向 LLM API 发送用户数据时缺乏低成本、易用的 PII 脱敏与审计方案，企业级方案又太贵且采购周期长。

目标用户：需要处理用户 PII、但没有专职安全工程师的中小型 AI SaaS 团队和独立开发者。

证据：small AI startups either send raw user data to LLM APIs (risky) or manually strip PII with regex (fragile and inconsistent). Enterprise solutions like Private AI or Nightfall exist but start at $10K+/year and require a procurement cycle. ... Would a shareable audit log export, a PDF you can hand to a client's security team be worth paying for?

变现：按 $29-99/月提供 hosted SaaS，按请求量或席位分层收费；审计日志 PDF 导出、合规报告、企业自托管作为增值功能。

MVP：基于现有 AegisFlow 做多租户 hosted 代理：API key 接入、Presidio/DistilBERT 脱敏、Redis 临时映射、可导出的签名审计报告、Stripe 订阅计费、简单用量看板。

竞品：Private AI、Nightfall、Microsoft Presidio、Skyflow、AWS Comprehend、LLM Guard

摘要：一名大四 CS 学生开源了 AegisFlow，在应用与 OpenAI/Claude 之间代理并脱敏 PII，宣称 200 RPS 下 p95 额外开销 4ms。他发现小 AI 团队要么裸传数据，要么用脆弱正则清洗，而 Private AI/Nightfall 等企业方案起价 $10K+/年。他想验证 $29-99/月的自助托管版是否有 B2B 市场，但目前没有评论或付费用户验证，需求仍偏假设。