Every database stores what.
NEDB also stores when, when it was true, and why.

Four axes. Everyone else ships one.

A relational database answers what is the value now? NEDB answers four questions natively — not with triggers and side tables, but with queryable, indexed, hash-chained primitives.

from nedb import NEDB

db = NEDB("./mydata")

# Time travel — the value at any past sequence, forever
snap = db.seq
db.put("users", "alice", {"age": 32})
db.get("users", "alice", as_of=snap)      # → the prior version

# Causal provenance — an edge in a hash-chained DAG, not a comment field
db.put("inputs", "msg_1", {"text": "user prefers dark mode"})
seq_msg = db.seq
db.put("beliefs", "dark_mode", {"value": True},
       caused_by=[seq_msg], evidence="user_message", confidence=0.95)

db.query('FROM beliefs WHERE _id = "dark_mode" TRACE caused_by')     # → msg_1
db.query('FROM inputs  WHERE _id = "msg_1"     TRACE caused_by REVERSE')  # → dark_mode

# Tamper evidence — verifiable by anyone, offline
assert db.verify()

History you can audit, not just history you have.

Most systems log events. NEDB chains them. Each write's hash commits to the write before it, the running Merkle head commits to everything, and the state root commits to the present — three separate commitments with different jobs.

How NEDB stacks up.

Judged on the four axes plus the operational basics. NEDB is a single-node embedded/daemon database — it does not pretend to be a distributed cluster; the comparison is about what a database can be asked and prove.

SQLite row is the stock engine; extensions (sqlean, SEE) can add pieces. Redis ≥6 has TLS in transit but no native at-rest encryption. MongoDB encrypted storage engine requires Enterprise Advanced. PostgreSQL row reflects core + common extensions (temporal_tables). Every cell is stated to the best of our knowledge as of Sept 2026 — corrections welcome, we apply them publicly.

Provenance that doesn't cost you the database.

Every number below is dated, names its machine, and ships with the command that reproduces it. Absolute figures aren't cross-machine comparable — read the ratios. Time travel costs ~30%, not 10×.

It speaks what you already speak.

NEDB is one Rust core with five ways in. No bespoke protocol lock-in — the PostgreSQL grammar is vendored wholesale (19,513 lines, 492 keywords, from PostgreSQL 17.4) and extended with the clauses only a permanent store can answer.

shop=> UPDATE orders SET total = 999 WHERE _id = 'o1';
shop=> SELECT total FROM orders WHERE _id = 'o1';                        -- 999
shop=> SELECT total FROM orders AS OF SYSTEM TIME 0 WHERE _id = 'o1';   -- 120 — the old value, forever

Built for systems that must show their work.

NEDB was designed in production for AI agent memory — where "the model said so" is not an audit trail. It ships with the machinery to keep one honest.

The limits, stated rather than buried.

An evaluator's first job is finding out what a product won't claim. Here is ours, in writing.

Three storage generations in 95 days, zero API breaks.

Every entry below is a real tag in this repository's history — git tag is the audit. Same hash-chain API across all of it: stores migrate forward automatically, client code never rewrote.