Vector-graph database
with memory semantics.
AI agents need memory that works like memory — where old information fades, important things stay vivid, and context flows through connections.
Not another vector store. Actual memory infrastructure.
Currently in development
The Problem
Your agent remembers everything.
That's the problem.
Think about how you remember things. A conversation from yesterday is vivid. A conversation from last year? Fuzzy, unless it was important. A conversation from five years ago? Gone — unless you've thought about it since.
This isn't a bug in human memory. It's a feature. Forgetting the irrelevant is what makes remembering the important possible.
Now look at how AI agents handle memory today. Every conversation, every fact, every interaction — stored forever with equal weight.
Outdated information competes with current knowledge
Storage grows without bound
Retrieval gets slower and noisier over time
The agent can't distinguish what matters from what doesn't
Vector databases and graph databases weren't built for this. They store data. They don't understand memory.
The Solution
Memory that works
like memory.
Aizu is a database built around how memory actually works. Not storage with search — actual memory semantics baked into the core.
When your agent recalls a memory in Aizu, it's not just finding similar vectors. It's weighing recency, importance, connection strength, and access patterns — just like your brain does.
How It Works
Four memory primitives.
Infinite possibilities.
Memories fade
Every memory has a natural lifespan. Store a conversation with a 30-day TTL, and it gracefully expires unless something keeps it alive. Old, unused memories don't clutter your agent's mind — they simply fade away.
Importance decays
A memory from yesterday should rank higher than the same memory from last year. Aizu applies time-based decay to relevance scores automatically. Recent memories surface first — unless older ones have been reinforced.
Access reinforces
When a memory is recalled, it gets stronger. Just like how thinking about something cements it in your mind. Important memories that keep getting accessed stay vivid. Unimportant ones fade. No manual curation needed.
Connections expand context
Memories don't exist in isolation. A fact came from a conversation. A decision was based on several observations. Aizu traverses these connections automatically, giving your agent the full context — not just a list of similar documents.
Under the Hood
Built for agents.
Designed to scale.
Hybrid queries
Vector similarity and graph traversal in a single query. Find memories by meaning, then expand context through connections. Sub-millisecond performance.
Time-aware ranking
Relevance isn't just similarity. It's similarity weighted by recency, access frequency, and explicit importance. The ranking your agent needs.
Automatic cleanup
Expired memories are garbage collected in the background. No manual pruning. No unbounded growth. Your agent's memory stays lean.
Concurrent access
Multiple agents can read and write simultaneously. Snapshot isolation ensures consistent views. Scales with your workload.
Use Cases
What you can build.
Personal assistants
Assistants that learn preferences over time — but don't get confused by outdated information.
Document retrieval
Company wikis and docs where outdated versions cause real problems. Old documents fade. Updated ones reinforce.
Customer support
Support agents that remember customer history without drowning in it. Recent tickets surface first.
Knowledge bases
Enterprise knowledge that stays fresh. Frequently-accessed articles rank higher. Stale content deprioritizes.
Meeting memory
Every meeting transcribed, every decision tracked. Recent discussions surface first. Old meetings fade unless referenced.
Research & analysis
Track sources and extracted facts with natural prioritization. Frequently-referenced papers stay top of mind.
Give your agents
real memory.
Aizu is currently in development. Join the waitlist to get early access and help shape the future of agent memory.