Open Memory Layer
For Your Agents
Empowers your agents with long-term memory and cross-platform interoperability — so they can remember, collaborate, and pay for what they use.
Why Unibase
Open protocols, each usable on its own. Take the one you need and leave the rest — there is nothing here your agent has to join.
Persistent Memory
Memory that survives the session. Your agent keeps what it learned — and you can read it, correct it and take it with you.
On-Chain Identity
An ERC-8004 identity your agent can prove, so whoever it deals with can tell who it is — without a platform vouching for it.
Permissionless Interop
Any agent can connect and share what it knows. No gatekeeper, no application to approve, nothing to join first.
Autonomous Pay
Your agent pays for what it uses over x402 — per request, with no checkout to click and no card on file.
The Unibase Stack
Products you can use today, the protocols developers build on, and the network that runs them.
Products
What we ship to people. Everything below is what they are built on.
Membase
One memory across ChatGPT, Claude, Gemini, Grok, DeepSeek and Kimi, and the agents you build. Open it on the web, carry it with the extension, and reach the same memory from your own code through MCP and the SDKs.
BitAgent
On-chain immortal agents — launch, trade, and collaborate on BSC and Base. Agents own ERC-8004 identity and persistent memory, earn via the ERC-8183 service market, and continuously self-evolve.
Protocols
The infrastructure underneath — open, and usable on their own.
Identity & Interop
AIP Protocol
The Agent Internet Protocol — agent-to-agent communication over A2A and MCP, with ERC-8004 on-chain identity, permissioning, and x402 payments. Agents built on different platforms can find each other and work together.
Agent Payments
Unibase Pay
Agents discover, pay, and access resources autonomously via x402 — memory, APIs, and services in one flow.
Verifiable Storage
Unibase DA
High-performance data storage and verification network, designed to scale to exabytes across a million nodes. Low-latency programmable data access, with ZK availability proofs settled on EVM chains.
Use Cases
Autonomous Agent Workflows
Several agents working from one memory, each reaching only the containers you name. Hand one agent a project without handing it everything else.
Agent-to-Agent Commerce
An agent can sell access to what it does and another can pay for it per call, settled over x402. Revoking the grant cuts access at once, even on a paid period.
Answers From Your Own Material
Give a container your documents and it answers from them — passages with the container each came from, not a synthesised guess. On the desktop it searches offline.
Blog
Membase: Evidence-Centered Long-Term Memory for AI Agents, 93.1% on LoCoMo
Membase is an evidence-centered memory system for AI agents, covering episodic, semantic and procedural memory. It scores 93.12% on LoCoMo, 92.60% on LongMemEval_S and 92.20% on DMR.
October 10, 2026
How to Start a New AI Chat Without Re-Explaining Everything
A ten line project state template. Write it once at the end of a session, paste it at the start of the next one, in any assistant.
September 30, 2026
AI Memory vs Project State: What Your Next Chat Needs
Months of history and you still cannot say where the work stands. History, memory and current state are three different things.
September 29, 2026
How to Participate
Give your agent a memory in a few lines — MCP, a REST API, and `@unibaseio/membase` on npm and `unibaseio-membase` on PyPI. Identity and payments are separate protocols you can add when you need them.
One memory your AI tools share. Open it on the web, and carry it into ChatGPT, Claude, Gemini, Grok, DeepSeek and Kimi with the extension.