Meminia

Private betaPatent pending

Memory that knows what changed—and why.

Meminia is a governed memory and knowledge substrate that preserves provenance, contradiction, correction history, and decision evidence across models and deployment surfaces.

Request a private review Explore the system

Built for on-device and self-hosted deployment paths.

Meminia iOS knowledge view showing a contradiction-aware graph and active memory nodes
Native Apple surface Recorded internal demo
Memory state Conflict preserved

Current codebase

Token-budgeted context selection

Spend the context window on evidence.

A configurable context runtime compiles retrieved candidates into a bounded evidence pack. The current implementation can score gain per token, reward topical diversity, preserve contradiction-bearing evidence, and emit a selection trace.

  • 01
    Bound the pack

    Set explicit token and item budgets instead of relying on uncontrolled truncation.

  • 02
    Protect disagreement

    Keep relevant conflicting evidence available for the model to explain.

  • 03
    Inspect the choice

    Return the backend, selected records, token total, objective, and fallback state.

meminia / context

$ meminia context optimize

backendgreedy_diverse budget52 tokens selected43 tokens fallbackfalse
retrieve→score→select→trace

$ python -m rdcl.quickstart --quiet

✓ quickstart: OK

retrieve · supersede · audit verify · tamper detect

Observed from the current local codebase Model-free quickstart

The governed record

Memory is more useful when its history survives.

01

Provenance

Keep source and supporting context connected to important memory.

source → record
02

Contradiction

Retain meaningful disagreement instead of silently overwriting it.

old ↔ new
03

Correction

Preserve a reviewable path from an earlier record to the current one.

v1 → v2
04

Audit

Record decision evidence in a model-free, tamper-evident journal.

verify ✓
Meminia iOS audit view showing recorded memory events
Recorded internal demo surfaceAudit feed · iOS

One layer, several interfaces

Meet the system where the work happens.

The current repository exposes the same governed-memory concepts through native Apple, Python, command-line, and HTTP surfaces.

Apple

Native app

iOS and macOS targets with local memory stores and on-device model paths.

Python

Library facade

Remember, supersede, retrieve, verify, and export through a compact API.

CLI

Operational interface

Ask, ingest, export, inspect graphs, and verify ledger state in-process.

REST

Integration surface

A FastAPI service for self-hosted and customer-managed deployments.

Qualified diligence

Examine the product, evidence, and limits.

Qualified reviewers can examine the current product, filed intellectual-property portfolio, technical documentation, validation records, and known limitations under appropriate confidentiality.

Discuss technical diligence