CASCADE Memory API

Semantic compression with geometric verification (τ-gating)

Bytes per signature

≥0.96

τ verification threshold

96%+

Semantic fidelity

⚠️ Responsibility Boundaries

CASCADE Memory API — Verification & Storage

This API provides verification and memory compression. For model training services, see our upcoming CASCADE Training offering (coming soon).

If your model produces repetitive outputs ("the the the") — that's a training issue, not CASCADE
If your embeddings are garbage — compressed garbage is still garbage
τ verifies geometric consistency, not factual accuracy of your training data

CASCADE Provides (ERN)	You Must Provide (User)
✓ Semantic compression (embedding → 78 bytes)	✓ Quality embeddings from a well-trained model
✓ τ-gated verification (geometric confidence)	✓ Proper model fine-tuning / RLHF
✓ Memory retrieval with similarity scoring	✓ Meaningful data to store
✓ Hallucination detection (τ < 0.96 = warning)	✓ Acting on warnings appropriately
✓ Post-quantum secure storage (Dilithium3)	✓ Secure API key management

Authentication

All requests require an API key in the Authorization header:

Authorization: Bearer YOUR_API_KEY

Endpoints

POST /v1/compress

Compress an embedding to a 78-byte signature with fidelity estimate.

// Request
{
  "embedding": [0.1, -0.2, 0.3, ...],  // Your model's embedding
  "format": "base64"
}

// Response
{
  "signature": "base64_78_bytes",
  "tau": 0.97,  // ≥0.96 = verified
  "fidelity": 0.96
}

POST /v1/memories

Store a memory in your semantic memory bank.

// Request
{
  "key": "user_preference_001",
  "embedding": [0.1, -0.2, ...],
  "importance": 5.0,
  "metadata": { "type": "preference" }
}

// Response
{
  "stored": true,
  "signature": "base64_78_bytes",
  "tau": 0.96
}

POST /v1/retrieve

Find similar memories by embedding.

// Request
{
  "embedding": [0.1, -0.2, ...],
  "k": 5,
  "min_tau": 0.96  // Only verified memories
}

// Response
{
  "memories": [
    { "key": "user_preference_001", "similarity": 0.92, "tau": 0.97 }
  ],
  "latency_ms": 0.8
}

What τ Means

τ (tau) is a geometric confidence score measuring consistency in semantic space:

τ ≥ 0.96  →  Geometrically verified, high confidence
τ < 0.96  →  Warning: potential hallucination or inconsistency
τ < 0.80  →  Reject: do not trust this output

Important: τ measures geometric consistency of YOUR model's outputs. If your model is poorly trained, τ will correctly identify the inconsistencies.

🔧 Need Help Fixing Your Model?

We offer τ-RLHF Training Services for enterprise clients. Our team can diagnose and correct geometric inconsistencies in your model's latent space.

Requires: Enterprise agreement + liability waiver Contact Sales →

Questions? Matthew@ern-inc.com