Know Which Facts to Trust

Turn your database into a geometric truth engine

Upload for a free structural audit. Pro users download the full .ramish engine - a compressed geometric model that scores every edge, runs locally forever, and plugs straight into your pipeline. Free open-source reader included.

πŸ“Š

Drop your database here

or click to browse

Supports: SQLite - CSV - JSON - Excel - Parquet - 50MB / 50K entities max, pro users 250mb / 500k max

How It Works

1

Upload

Drop your SQLite database or CSV file. We detect the schema automatically.

2

Analyze

Ramish builds a 4D geometric model and scores every edge by structural consistency.

3

Review

Get a report showing which facts have low support β€” the ones to double-check.

What You Get

🎯 Risk Snapshot

Instant assessment: Low, Moderate, or High risk based on structural consistency patterns.

πŸ“‰ Trust Distribution

Histogram of edge weights. Understand the shape of your data quality.

⚠️ Suspicious Edges

Top 10 lowest-support facts. These are the ones to review first.

πŸ“Š Relation Analysis

Breakdown by relation type. Find which relationships are most problematic.

Simple Pricing

Start free. Upgrade when you need full power.

Free

$0
3 audits/month • 50MB / 50K entities per audit
  • βœ“ Risk snapshot (Low/Med/High)
  • βœ“ Trust distribution histogram
  • βœ“ Top 10 suspicious edges
  • βœ“ Markdown report export
  • βœ— Full .ramish file
  • βœ— 500,000 entities
  • βœ— Local execution
  • βœ— CLI explorer tool
Try Free

Free Open-Source Explorer

Query, validate, and audit your .ramish file locally. No account required. Runs offline forever.

pip install ramish-explorer PyPI ? GitHub ?

The Ramish Advantage

What the full .ramish file unlocks for Pro users

πŸ”’

Fully Local Execution

Your .ramish file runs entirely on your machine. Query your knowledge graph without sending data anywhere. True data sovereignty.

⚑

Frozen-Key Inference

Each .ramish engine embeds relational keys as frozen geometric rotations. Score new edges against existing structure without retraining - your baseline stays locked while your data evolves.

πŸ“Š

Production Scale

Up to 500,000 entities and 250MB per dataset. Covers most production knowledge bases, RAG systems, and research datasets. 10x the free tier.

πŸ”Œ

RAG Pipeline Ready

Every edge in your .ramish file carries a trust score. Query scores with the explorer, filter low-confidence facts, and keep hallucination-prone data out of your retrieval pipeline.

πŸ“‰

Built-in Compression

The geometric structure of a .ramish file naturally compresses your data. With int8 quantization, up to 30% smaller still. Ship your knowledge graph on a thumb drive.

πŸ•°οΈ

Drift Detection

Audit your data as it evolves. Each .ramish engine captures a geometric snapshot of your knowledge graph. Compare snapshots across versions to measure how your data's structural integrity changes over time. Native implimentation coming soon.

Frequently Asked Questions

What is the ramish-explorer?

A free, open-source command-line tool for reading and querying .ramish engine files. Install it with pip install ramish-explorer. It runs entirely on your machine - no account, no server, no internet required.

What file formats are supported?

Currently: SQLite databases (.db, .sqlite, .sqlite3) and CSV files. We auto-detect your schema, including foreign key relationships, junction tables, and entity types. Parquet and JSON support coming soon.

Is my data secure?

Yes. Uploaded files are processed in isolated containers, never written to permanent storage, and automatically deleted within 24 hours. We don't train on your data or share it. Pro users get a .ramish file that runs entirely locally.

What does "low support" actually mean?

Every fact is verified by how many independent paths confirm it geometrically. "USA is a country" has thousands of supporting paths. A suspicious edge like "Einstein invented Bitcoin" has noneβ€”the graph doesn't agree. Low support means structurally isolated, worth reviewing.

How is this different from Neo4j or vector databases?

Neo4j stores and queries graphs but doesn't verify them. Vector databases embed content but lose relational structure. Ramish combines both: we preserve topology and embed it geometrically, then score every edge by structural consistency. Truth verification that neither offers alone.

How accurate is the fake detection?

Ramish flags structurally inconsistent edges - facts that don't fit the geometric pattern of your data. In testing with injected false edges, they consistently rank in the lowest support tiers. Real-world performance depends on graph density. The audit report shows your specific risk distribution.

What's the math behind this?

Ramish uses Quaternion Knowledge Graph Embeddings (QuatE), where relations are rotations in 4D space. When you traverse a triangle (A→B→C→A), the rotations should compose to identity. The "deficit" measures deviation. High deficit = structural inconsistency.

Can I use this with my RAG pipeline?

That's a primary use case. Run your data through Ramish, use the free explorer to query trust scores, and filter low-confidence facts before retrieval. Low-support facts cause hallucinations - Ramish helps you catch them first.