Marvnor

RECORD-BASED MEMORY FOR AI APPLICATIONS

Give your AI
memory it can check.

Store your records. Get TRUE, FALSE or UNKNOWN from the evidence they contain. Flag conflicts, correct records in place and delete only what you choose.
Marvnor is a proprietary hosted API; your model writes the answer.

Python · No SDK · No dependencies

ONE ORDER. FIVE STEPS.

From conflicting records
to a clear answer.

Recorded public API run

An order has one payment status. What happens when saved records say both “pending” and “paid”?

  1. 01

    Save a fact

    Store status = pending.

    TRUEPending is supported.
  2. 02

    Add a conflicting record

    Also store status = paid.

    CONFLICTBoth return UNKNOWN. Ask for clarification.
    See the actual response
    {
      "conclusion": "UNKNOWN",
      "conflict": true,
      "reason": "conflicting_values",
      "path": ["paid", "pending"],
      "decision": "needs_review",
      "evidence_kind": "conflict"
    }
  3. 03

    Resolve the conflict

    Remove the incorrect “paid” record.

    TRUEPending is supported again. No conflict.
  4. 04

    Correct the remaining fact

    Edit the saved value to paid.

    TRUEPaid is supported. Pending becomes UNKNOWN.
  5. 05

    Delete the demo data

    Remove only this example’s records.

    UNKNOWNNo remaining support for either value.

This page displays saved results, not a live API call. The runnable example uses your API key and account quota.

LESS HISTORY. RELEVANT EVIDENCE.

Send the facts needed
for this question.

In the final question of our 160-turn synthetic-history test, relevant verified results replaced the full history. Both answers were correct.

Read the test and token accounting ↗

TRY YOUR OWN USE CASE

One file. Your API key.
A result you can inspect.

Download the example, run it and paste your key at the hidden prompt. It checks every step and removes only its synthetic data.

Get the example ↗

Python 3.9+ · standard library only

python quickstart.py

Marvnor API key (hidden):

1. Saved fact
2. Conflicting records
3. Conflict resolved
4. Corrected fact
5. Demo data deleted
Create an account ↗

Building an agent or project-memory tool?

Bring one concrete use case. Let’s test it with non-sensitive sample data.

wendelxia@gmail.com ↗