RAWTHINK

Persistent memory for AI development.

Every AI session starts from zero. Your context window is your entire relationship. Close the tab, lose the thread.

RAWThink fixes this.

Three layers. One memory.

01

Raw Input

Sessions, documents, sources. Immutable data that flows in and never changes.

02

Knowledge Graph

Entities, relations, observations. JSONL-backed, connected across every session.

03

Memory Layer

MEMORY.md + MCP schema. Configuration, lifecycle, personality. The interface.

Built for real workflows.

Hybrid Search

Vector + BM25 via Qdrant. Dense and sparse retrieval with RRF fusion. Find anything across your entire history.

Knowledge Graph

Typed entities, semantic relations, temporal observations. Turkish character normalization built-in. Your knowledge, structured.

Session Lifecycle

Automatic session management, handoffs, context restoration. Every session builds on every previous one.

From the blog.

Your AI Doesn't Remember You. That's About to Matter More Than You Think.

How two engineers independently arrived at the same architecture — and what it reveals about the next layer of personal AI.

Read on dev.to →