Replay the state.
Ask the same question against what the database knew at a specific moment—even after corrections and backfills.
Explore time travel ↗ THE DECISION DATABASE
Give your AI a memory
you can stand behind.
Store the facts, relationships, and evidence behind your agents’ decisions.
Reconstruct what they knew—even after the world changes.
FIG. 01 / THE TIME LATTICEEACH CHANGE ADDS A VERSION
VALID TIME →
Bitemporal by designCypher-nativeVersioned evidenceTamper-evident history
GIVE EVERY DECISION ITS CONTEXTConnected. Relevant.
Connected. Relevant.
Remembered.
Follow relationships. Find similar evidence. Preserve the history of both.
Bring the context of a decision into one connected record.
VALID ON MAR 11, 2026KNOWN BY MAR 04, 2026
GraphFollow the relationships.
VectorFind similar evidence.
BitemporalWhen it was true. When you knew.
Illustrative concept.
01 / TRY THE INVESTIGATIONYour agent flagged a risk.
Your agent flagged a risk.
Then the facts changed.
A correction arrived after the assessment. Can you explain the original result—and show what changed? Follow the evidence below.
Interactive example · simulated data, not a live database query
Start with the evidence available at the time. The June correction is still outside this historical view.
UNDERWRITING TWINLOAN-1842
VALID TIME FIXED: MAR 11, 2026
Adverse bureau signalRecorded · March 4
ACTIVEIncome verificationRecorded · April 18
NEW EVIDENCEBureau correctionRecorded · June 1
REVISIONHISTORICAL ASSESSMENTREV 01
What the twin knewHIGH RISK
Confidence70%
The original adverse bureau signal supports the risk assessment. The later correction has not yet been recorded.
Earlier versions remain intact.Revisions add to the record. They do not rewrite it.
Now imagine answering this question for your own agent.Build your first example ↗
02 / MAKE IT YOURSStart with one agent.
Start with one agent.
Keep the evidence.
Connect your application, record its evidence, and ask a historical question. Start with a focused example before expanding to your full workflow.
01 / CREATE
Create your database
Start with your first application.
Open signup ↗02 / CONNECTFollow the quickstart
Connect your app and record its first facts.
Read the guide ↗03 / RECONSTRUCTQuery an earlier state
See what was known before a correction.
Explore the docs ↗// What was known about March 11, as of March 4?
MATCH (b:Belief {id: "belief-99"})
AT RECORDED '2026-03-04'
AT VALID '2026-03-11'
RETURN b.claim, b.confidence, b.created_atHistorical query example · not executed on this pageRead the docs ↗
03 / TWO CLOCKS. ONE RECORD.The world changes.
The world changes.
The record stays.
A corrected fact should change what you know now. It should never erase what you knew then.
InvariantDB preserves the facts, relationships, and evidence behind your decisions. Query the past without rebuilding it from scattered logs and snapshots.
01
Valid time
When the fact was true in the world.
02
Recorded time
When your system learned it.
04 / THE EVIDENCE IS THE EXPLANATIONRECONSTRUCT FOLLOW THE LINEAGE VERIFY INTEGRITY
Replay. Trace. Verify.
Trace the evidence.
Connect decisions to the beliefs, inputs, sources, and revisions that supported them. Provenance lives with the record.
See decision lineage ↗Check the history.
Tamper-evident records help establish the integrity of retained history. Start an investigation with evidence.
Explore governed workflows ↗05 / AGENTIC TWINSAgents execute.
Agents execute.
Twins persist.
Give your agent a durable model of the world—and a history of how that model changed.
People. Processes. Assets. Organizations. Preserve the relationships and beliefs an agent relied on, not just its latest answer.
Explore AI memory ↗THE AGENT ACTSTHE TWIN REMEMBERS
Evidence received
A new document enters the record.
Belief updated
Source and prior version stay connected.
Decision recorded
The evidence used is retained with the action.
Reconstruct the moment
Inspect what the twin knew when it acted.
Your next agent deserves
a better memory.
Start with one decision. Keep the evidence. Be ready when someone asks why.