Temporal Queries

TensorDB’s temporal query syntax lets you query across both time dimensions.

AS OF — System Time Travel

Query the database as it existed at a specific commit timestamp:

-- See the state at commit 100
SELECT * FROM users AS OF 100;

-- Combined with WHERE
SELECT name, balance FROM accounts AS OF 500
WHERE account_id = 'acc_alice';

Behavior

• Returns only facts with commit_ts <= specified_ts
• For each key, the latest version within that window is returned
• If no version exists at that timestamp, the key is absent from results

VALID AT — Business Time Filter

Query facts that were valid at a specific business timestamp:

-- What rate was valid on June 15, 2024?
SELECT * FROM policies VALID AT 1718409600;

Behavior

• Returns facts where valid_from <= specified_ts < valid_to
• Filters on the business-time dimension, independent of when the data was recorded

Combined: AS OF + VALID AT

Use both together for full bitemporal queries:

-- What did the system know at commit 200, about rates valid in June 2024?
SELECT rate FROM policies AS OF 200 VALID AT 1718409600;

Use Cases

Pattern	SQL	Purpose
Current state	SELECT * FROM t	Latest data
Point-in-time audit	SELECT * FROM t AS OF 100	What was known at time 100
Business-time lookup	SELECT * FROM t VALID AT ts	What was valid at business time ts
Full bitemporal	SELECT * FROM t AS OF 100 VALID AT ts	What was known at time 100 about business time ts

Key-Value API

For the key-value API, temporal reads are also supported:

// Time-travel read
let value = db.get_as_of("user/alice", 500)?;

// Bitemporal read
let value = db.get_temporal("user/alice", Some(500), Some(1718409600))?;