Routing

HumemDB starts with explicit routing because hidden routing rules would make the first release hard to reason about.

Routes

• sqlite: canonical write path and transactional source of truth
• duckdb: analytical read path over SQLite-backed data

Current policy

• send writes to SQLite
• allow read-only SQL on DuckDB
• allow Cypher reads on SQLite or DuckDB
• keep vector search on SQLite

Why explicit routing comes first

Automatic routing is planned later, but the runtime needs benchmark evidence and stable surface semantics before it should guess for the caller.

Example

result = db.query(
    "SELECT kind, COUNT(*) AS total FROM events GROUP BY kind",
    route="duckdb",
)

If that same call were a write, HumemDB would reject it instead of silently mutating the analytical replica.