Open Format · v0.2.0 Draft

Portable Knowledge Containers

An open format for storing, searching, and transporting knowledge. SQLite for search. Filesystem for blobs. One bundle per scope. Move it anywhere.

Read the Docs Quickstart

What's inside a .omnidata bundle?

A directory that behaves like a file. Each component uses the right tool for its job.

my-knowledge.omnidata/ ├── manifest.json identity, schema version, owner ├── index.db SQLite: resources, chunks, embeddings, FTS5 ├── memory.db SQLite: collections, edges, tags, graph ├── blobs/ content-addressed files (SHA-256 fanout) │ └── ab/ab3f7c... raw PDFs, images, audio, video ├── adapters.json what feeds this instance └── ingress.log append-only audit trail

One container, one scope

Each role, project, or client gets its own .omnidata bundle. No data leakage. No shared state. Isolation is structural.

Filesystem as runtime

Blobs are real files. Get btrfs compression, ZFS snapshots, APFS clones, and S3 backends for free. The OS does the work.

Search built in

RRF fuses vector similarity and FTS5 full-text search. Every resource is findable by meaning, not just keywords.

Adapter-fed

Browser captures, voice recordings, messages, emails, files. Adapters discover content and feed it into the bundle.

ManyHats aware

Each hat gets its own .omnidata. Switch roles, switch knowledge. The right context for the right work.

Machine-first

Designed for AI agents. Verbose field names, documented schemas, llms.txt. Any LLM can read and search it.