Storage built for AI.

Object storage, retrieval and persistent agent memory in one engine. Applications connect through S3-compatible APIs; agents access knowledge and memory through native MCP. Your team builds the application. Origon operates the datastore service.

What did we decide about the launch? Launch plan.pdf: Begin with a small pilot group. Expand access in phases after reviewing feedback. Planning session: Keep the rollout phased. Review the pilot before opening access to the wider team. Rollout notes.md: Start with the pilot team and use their feedback to guide the wider release.

Source material and work history, together.

Store application data alongside the artifacts agents produce, execution records and the history of where information came from.

Knowledge extraction makes that material available for retrieval and use.

Find information by meaning, words and relationships.

Three illustrative matches for a question about when access opens to everyone. By meaning, a planning session passage about bringing the rest of the company in after the pilot review. By words, a release checklist line containing access and everyone. By relationships, company-wide access follows the pilot review, which the release team owns.

CapabilityWhat it retrieves
Vector searchRelevant information by meaning, using HNSW
Full-text searchMatching words and phrases, using BM25
Knowledge graphsConnections between information across your sources

Memory beyond the current task.

Agents can retrieve earlier decisions, commitments and lessons across tasks and sessions, including work completed months earlier.

Memory supports correction, configured retention and authorized deletion.

How it comes together.

A maintenance application stores approved procedures and equipment records. An agent retrieves relevant guidance, saves its work-order draft with the source history, and records the reviewer’s decision. A later task can retrieve that decision and its context through memory.

Connect applications and agents.

  • S3-compatible APIs. Give applications access to objects and artifacts.

  • Native MCP. Give agents access to memory, sessions and lineage.

  • Access and protection. Access controls, personal-information detection and redaction govern how information is used.

Storage across the cluster.

Erasure coding adds storage redundancy, replication distributes data across the cluster, and nodes communicate over QUIC. The engine has no external database dependencies.

  • Objects

    S3-compatible API for applications to store and read files and artifacts.

  • Vectors

    HNSW index for finding relevant information by meaning.

  • Full text

    BM25 index for matching exact words and phrases.

  • Graph

    Knowledge graph that connects information across your sources.

  • Sessions

    Session workspaces that keep each task’s work and its history together.

  • Memory

    Persistent agent memory that carries decisions and context across tasks and sessions.

See it work with your data.

Explore storage, retrieval and memory using representative files and tasks. We’ll work through capacity, access, retention and migration requirements for your application.