Kraterion

Docs

Run agents on storage you own

Kraterion is an S3-compatible storage platform where every file is owned on-chain by you and encrypted by default — and where AI agents read, write, and reason over that storage with access you can revoke at any time.

What is Kraterion

Put files in a bucket the way you already do with S3. Underneath, each file is stored on Walrus as a blob owned by your Sui account, encrypted before it ever leaves the gateway. Kraterion can read your data only because your on-chain bucket grants it that access — and a single transaction takes it away. No file is custodial; nothing is readable once you revoke.

Agents, first

The reason most teams come to Kraterion is the agent layer. Define an agent once — a system prompt, a model, the buckets it can see, the tools it can call — then talk to it over an OpenAI-compatible endpoint. It searches your documents, cites every claim back to the exact bytes on Walrus, remembers what matters, and can be embedded on your own site with scoped, rate-limited access.

Storage, knowledge, and MCP are the foundation that makes those agents trustworthy: the data they reason over is owned and verifiable, and the access they hold is on-chain and revocable.

How it fits together

  • Storage — an S3-compatible API for buckets and objects, encrypted by default. Use any S3 client.
  • Knowledge — turn a bucket into a searchable index with hybrid keyword + vector retrieval and verifiable citations.
  • Agents — configurable assistants with tools, memory, and knowledge, reachable over an OpenAI-compatible chat API.
  • MCP — the same buckets, search, and agents exposed to Claude Desktop, Cursor, and any MCP client.

Where to go next

New here? Start with the concepts, then the quickstart.