↓Skip to main content
  1. Tools/
  2. Publications/

docbank

·2 mins·
ai-partial Publications external Documents Pdf Pdf Documents Sqlite Content-Addressed Search Versioning Local-First
Language: Go
License: Apache-2.0
Maturity: beta
Verified: 2026-08
Table of Contents

docbank is a “self-sovereign document system” for managing records that users and AI agents need to organize, locate, modify, and verify. It handles PDFs, images, and text files, keeping them in a local catalog without requiring cloud storage.

Storage and Indexing #

  • Catalog: SQLite database (CGO or pure-Go implementation)
  • Content store: Local filesystem or S3-compatible object storage, physically separate from the catalog database
  • Stable node IDs: Documents retain consistent identifiers even after being moved or renamed
  • Immutable versioning: Each content state receives a permanent SHA-256 digest
  • Virtual filesystem: Familiar folder structure over the content store
  • Tags: Supplement search with categorical labels

Ranked full-text search over document names and extracted text. The page does not describe the text extraction pipeline or which document formats support full-text extraction beyond PDFs.

Preservation Features #

  • Recoverable deletion with explicit garbage collection
  • Incremental backups with verification
  • Complete content authentication (SHA-256 digests)
  • Optional permanent audited history (all versions retained)

Relation to git-annex / DataLad #

docbank’s content-addressed storage and immutable versioning overlap conceptually with git-annex’s design, but the two systems are independent. docbank manages its own catalog and content store rather than delegating to git-annex or DataLad.

For the con/serve mission, docbank is relevant as a local document access layer – particularly for scenarios where researchers want to manage a personal PDF library with AI-friendly search without the overhead of a full DataLad dataset. The ~/.docbank/ store could in principle be snapshotted into a DataLad dataset for off-site backup, but no established workflow for this exists.

Compared to Zotero (which manages scholarly references with metadata, citation keys, and publisher integration), docbank is format-agnostic and does not attempt citation-level metadata – it is closer to a content-addressed file store with search.

See Also #

  • Zotero – reference manager with citation metadata and PDF management
  • citations-collector – discovery and archival of scholarly citations
  • agentsview – sister kenn-io tool for AI session artifacts