SQLite-backed tape store plugin for bub.

• Bub plugin entry point: tapestore-sqlite
• A provide_tape_store hook implementation backed by SQLite
• One database file for all tapes instead of one JSONL file per tape
• Native SQLite indexes for tape listing, anchor lookup, and kind filtering

uv pip install "git+https://github.com/bubbuild/bub-contrib.git#subdirectory=packages/bub-tapestore-sqlite"

You can also install it with Bub:

bub install bub-tapestore-sqlite@main

Note: bub-tapestore-sqlite currently supports bub 0.3.1 and later.

The plugin reads environment variables with prefix BUB_SQLITE_:

• BUB_SQLITE_PATH (optional): database path
  • Default: <BUB_HOME>/tapes.sqlite3
• BUB_SQLITE_BUSY_TIMEOUT_MS (optional, default: 5000)
• BUB_SQLITE_JOURNAL_MODE (optional, default: WAL)
• BUB_SQLITE_SYNCHRONOUS (optional, default: NORMAL)
• BUB_SQLITE_EMBEDDING_MODEL (optional): any-llm model identifier, for example openai:text-embedding-3-small

• The plugin overrides Bub's builtin file-based tape store through provide_tape_store.
• Default database location: <BUB_HOME>/tapes.sqlite3.
• Entry IDs come from the autoincrement tape_entries.id column.
• Query behavior matches Bub tape queries:
  • all()
  • after_anchor(...)
  • last_anchor()
  • between_anchors(...)
  • kinds(...)
  • limit(...)
• Optional vector search:
  • Message entries are indexed lazily when you call await TapeQuery(tape, store).query("...").all().
  • Configure BUB_SQLITE_EMBEDDING_MODEL.
  • The first successful embedding call persists vector dimensions automatically from the returned embedding size.
  • Semantic queries reuse the standard fetch_all() path and still honor kinds(...), limit(...), anchor bounds, and date bounds.

• Enables PRAGMA foreign_keys = ON
• Uses PRAGMA journal_mode = WAL by default
• Uses PRAGMA synchronous = NORMAL by default
• Applies PRAGMA busy_timeout for concurrent readers/writers in one process
• Loads the bundled sqlite-vec extension when vector search is enabled