CAIN Memory LIVE

Governed persistent memory for autonomous agents.

Overview

CAIN Memory gives autonomous agents persistent, adaptive, governed, tenant-isolated memory that can be written, managed, retrieved, verified, corrected, forgotten, and audited without allowing untrusted or unauthorized information to silently become trusted agent knowledge.

Trust Model

Level	Description
UNVERIFIED	Newly written, not yet assessed
OBSERVED	Seen but not validated
VALIDATED	Cross-referenced and found consistent
TRUSTED	Authoritative, can be used for consequential decisions
REVOKED	Explicitly invalidated
EXPIRED	Time-based decay

Memory Types

• episodic - Events and experiences
• semantic - Facts and knowledge
• procedural - How to do things
• preference - User preferences
• task - Task-related information
• working - Temporary working memory
• factual - Verified facts
• experience - Agent experiences
• instruction - Instructions and directives
• observation - Observations

API Endpoints

Store Memory

POST /fabric/memory/store
X-API-Key: {key}
{
  "agent_id": "agent-1",
  "memory_type": "semantic",
  "content": "The server runs on port 8399",
  "source": "tool",
  "confidence": 0.95
}

Retrieve Memory

GET /fabric/memory/get/{memory_id}
X-API-Key: {key}

List Memories

GET /fabric/memory/
X-API-Key: {key}

Update Memory

PATCH /fabric/memory/update/{memory_id}
X-API-Key: {key}
{
  "content": "Updated content"
}

Delete Memory

DELETE /fabric/memory/delete/{memory_id}
X-API-Key: {key}

Change Trust Level

POST /fabric/memory/trust/{memory_id}
X-API-Key: {key}
{
  "trust_level": "revoked",
  "reason": "Incorrect information"
}

Get Stats

GET /fabric/memory/stats
X-API-Key: {key}

Health Check

GET /fabric/memory/health

Security Features

• Poisoning Defense: Detects prompt injection, indirect injection, and malicious content
• Tenant Isolation: Memories are strictly isolated by tenant
• Admission Control: Every memory passes admission evaluation
• Provenance Tracking: Full chain of custody for each memory

Admission Control

Every memory write passes through admission control evaluating:

• Provenance: Where did this information originate?
• Source Reliability: Is the source trustworthy?
• Factual Confidence: How certain is the accuracy?
• Novelty: Does this contradict existing trusted memory?

A memory produced by an agent must NOT automatically become authoritative.

Pricing

Operation	Cost
Storage	$0.01 per KB
Retrieval	$0.001 per read
Embedding	$0.002 per operation
Consolidation	$0.005 per operation
Summarization	$0.01 per operation

Dashboard

CAIN Memory Dashboard - Live statistics, interactive demo, and full API reference.