Memobase – Universal memory that works across all your AI tools

Name: Memobase – Universal memory that works across all your AI tools
Availability: InStock
Author: chsitter

by chsitter·Mar 3, 2026·2 points·15 comments

Visit Project View on HN

AI Analysis

●●SolidSolve My ProblemBig Brain

MCP-native memory beats vendor lock-in, but memory import already exists in Claude.

Strengths

•MCP protocol choice is genuinely smart—sidesteps the silo problem by using open transport rather than fighting each platform's APIs.
•Sub-100ms vector retrieval with Postgres RLS isolation is production-ready, not hand-wavy.
•Free tier with 50 ops/month + trivial integration lets you try it immediately—no friction.

Weaknesses

•Claude's native memory import (dropped yesterday) solves cross-session persistence for the primary use case; unclear what Memobase adds beyond portability.
•Depends entirely on MCP adoption—if Claude/Cursor don't prioritize MCP tools, this becomes a niche integration that most users never discover.

Post Description

Hey HN — I'm the builder behind Memobase.

Timing: Anthropic just launched memory import for Claude yesterday. You can export your ChatGPT memories and bring them over. It's a step in the right direction, but it's still moving your data from one silo to another. You don't really own that memory.

The problem as I see it: there's no standard protocol for AI memory. You can't say "here's my MCP server, use it for memory in every session." Each platform builds its own walled garden. Number portability took regulation. Email interoperability took SMTP. AI memory needs something similar.

What Memobase is: a universal, AI-agnostic memory layer. It builds a structured profile — your preferences, context, project history — that any connected AI tool reads from. Not locked inside ChatGPT, Claude, or any single platform.

Technical approach: - Profile-based memory, not raw conversation logs. Compact and fast (sub-100ms lookups). - You own your data. Full visibility, editing, deletion, export. Self-hosted option coming. - Working toward an open protocol so any tool can plug in — not just our integrations.

What's live: open beta with the core memory and integrations for the major tools. What's still patchy: Agents don't automatically use it all the time without being prodded, the protocol spec is still being formalized, and we need more tools to adopt it for this to really work.

I'd love to hear: - Would you want your AI memory to live outside any single platform, or do you prefer each tool handling it? - What would the protocol need to look like for you to build against it? - Technical feedback on the approach — we chose profile-based RAG vs knowledge graphs etc, happy to go deep on that.