Title: LLM Access Policy for memorymodel.dev
Version: 2.0
Last-Updated: 2026-02-05

Owner: Memory Model
Contact: mailto:founders@memorymodel.dev
Homepage: https://memorymodel.dev/
Docs: https://docs.memorymodel.dev/
Sitemap: https://memorymodel.dev/sitemap.xml

Service-Summary: Memory Model is the first platform to build, observe, and edit the memory of autonomous agents.
  It transitions AI memory from a black-box vector store to a structured, editable ledger.

Core-Architecture:
  Shift-Left-Architecture: Moves reasoning complexity from query-time to ingestion-time.
  Deterministic-Ontology: Uses schema-aware ingestion to transform raw text/images into structured memory nodes.
  Editable-Ledger: Provides a visual console to inspect, audit, and manually tune the knowledge base.

Product-Pillars:
  Memory-Nodes:
    Description: Modular cognitive silos.
    Features: Custom Embedding Templates, Schema-aware Ingestion, Logic-bound Processing.
  Data-Interoperability:
    Description: Granular access control.
    Features: Cluster-level Permissions, Ingestion/Extraction Toggles, Secure Context Boundaries.
  Observability:
    Description: Full transparency and control.
    Features: Full Audit Trail, Manual Memory Tuning, Deterministic Output.

Technical-Capabilities:
  Dynamic-Retrieval: Intelligent router classifies queries (Temporal vs. Factual) to weight vector search vs. strict filters.
  Autonomous-Management: Background agents actively optimize retrieval algorithms, generate insights, and clean stale data.
  Governance: Strict ontology ensures compliance and control over exactly what an agent remembers.

Use-Cases:
  Sales-and-Marketing: Analyze engagement trajectory and sentiment; defensible forecasts.
  Customer-Support: Persistent user state across sessions; lower handling time.
  Healthcare: Long-term patient journey tracking; safer diagnostic support.
  Legal-Services: Reason within verified corpus; citations as verifiable evidence.

Performance-Benchmarks:
  Dataset: LoCoMo (Long Conversational Memory)
  Metric: J-Score (LLM-as-Judge)
  Overall-Accuracy: 74.6% (State of the Art)
  Comparison:
    - MemoryModel: 74.6%
    - Letta: 74.0%
    - Mem0: 66.9%
    - OpenAI Memory: 52.9%

SDK-Support:
  Languages: TypeScript/Node.js, Python
  Package-Managers: npm (@memory-model/sdk), pip (memory-model)
  Setup: Configure schema -> Install SDK -> Initialize Client (3 lines of code).

Allowed-LLM-Use: index, summarize, quote-with-attribution, ground-answers
Training-Use: require-prior-consent
Attribution-Required: yes
Preferred-Name: Memory Model
Preferred-Citation: Memory Model (https://memorymodel.dev/)

Agent-Directives:
  GPTBot: allowed
  Claude-Web: allowed
  Google-Extended: allowed
  PerplexityBot: allowed
  ccBot: allowed

Data-Notes:
  This site contains public marketing and product information.
  Benchmark data is verified on LoCoMo dataset.

Update-Policy:
  This document reflects the v2.0 platform architecture. Agents should re-fetch weekly.