HUPI is a self-hosted memory gateway for AI apps — durable, fact-checked, and provider-independent. Point any OpenAI-compatible client at it and every conversation gets captured, verified against its source before being trusted, and recalled with a full audit trail of what was used and why. Self-hosted on your own Postgres, works with any LLM vendor, and supports real multi-tenant team workspaces with per-team encryption. MIT licensed — memory you own, not a black box.

What real task does your product handle with GPT-6 Astra?

GPT-6 Astra just like any other LLM needs memory retention beyond some the full context window. HUPI will help with that.

Hey PH 👋 — maker here. I built HUPI after noticing how much time gets lost re-explaining context to AI tools, at both the personal and team level. Every existing memory feature I looked at was locked to one vendor, opaque about what it actually stored, and built for a single user. The part I care most about isn't "it remembers things" — plenty of tools do that. It's that every memory goes through an independent fact-check before it's trusted, retrieval logs exactly what it used and why for every single response, and corrections never silently rewrite history. If you've ever had an AI confidently state something wrong that it supposedly "remembered," this is the problem HUPI is built to solve. It's MIT licensed and self-hosted by design — I wanted this to be infrastructure people actually own, not another subscription. Would love feedback, especially on the setup experience and what's missing for your use case.

What real task does your product handle with GPT-6 Astra?

GPT-6 Astra just like any other LLM needs memory retention beyond some the full context window. HUPI will help with that.

HUPI

Self-hosted memory for AI that verifies before it trusts

Self-hosted memory for AI that verifies before it trusts