Pillar Two — Intelligence

The Intelligence Model

Proprietary, self-hosted, and completely isolated from the outside world. The model runs in your cloud environment. Not ours. Not anyone else's.

Architecture

Core Specifications

The PanOps intelligence model is an open-weights AI model deployed and run entirely within your cloud environment. It is provisioned and managed by PanOps on your behalf — but your data never leaves your infrastructure to reach it.

Hosting

Self-Hosted Architecture

The model runs entirely on compute inside your cloud environment. Inference is local. No query, prompt, or response is transmitted to an external system — including PanOps systems. Your data stays where it is.

Data Isolation

Zero External Transmission

PanOps personnel have no access to your data by design — customer encryption keys never leave your cloud environment. The architecture is designed to make external access impossible — not just contractually prohibited. Isolated compute with no outbound data paths for communications or query content.

Performance

Benchmarked Performance

The self-hosted model is evaluated against leading commercial models on organizational intelligence tasks specific to PanOps use cases — not general benchmarks. It must meet or exceed defined thresholds before deployment to any customer environment.

Roadmap — Phase 3

Enterprise Dedicated Tier

A dedicated model instance per customer — separate compute, separate weights storage — available in Phase 3 for organizations in regulated industries or with strict compute isolation requirements beyond the standard architecture.

Architecture Detail

How the Shared Model Architecture Works

The standard PanOps deployment uses a shared model weights approach with fully isolated customer data. This is the same pattern used by nearly every self-hosted enterprise AI deployment.

Architecture — Standard Tier

Shared

Model weights — the underlying AI model parameters are the same across customer deployments. The model itself is not personalized per customer.

→

Isolated

All customer data — your communications archive, query history, embeddings, and persistent memory are stored and processed only in your environment. No cross-customer data access is possible.

Shared

PanOps-managed model updates — PanOps deploys model updates into your environment as improvements are made. You can defer updates if needed.

→

Isolated

Your encryption keys — all stored data is encrypted under keys you control. PanOps has no key material and cannot decrypt your data.

Memory

Persistent Memory Architecture

PanOps builds an organizational memory over time — drawing on prior queries, identified patterns, and accumulated context. Memory is stored within your isolated environment and never leaves it.

Session Memory

Context Across Queries

Signal tracks what has been asked before. Prior queries inform new answers — so you don't repeat context and responses grow more precise over time as the system learns your organization's patterns and terminology.

Organizational Memory

Builds Continuously

The PanOps system identifies recurring topics, relationships, and threads across your organization's communications. This organizational context accumulates over time and surfaces patterns a single query wouldn't reveal.

Technical Notes

Implementation Details

For teams doing technical diligence on the model architecture.

Embedding Pipeline

The embedding pipeline runs entirely within your cloud environment. Communications are embedded locally — no third-party embedding API. Vector storage is isolated per customer.

Transcription

Audio transcription for voice and video connectors uses a self-hosted open-source transcription and translation pipeline deployed inside your cloud environment. No audio is sent to external transcription services.

Compute Cost

Estimated infrastructure cost at typical organizational scale (20–150 employees, moderate communication volume) is approximately $50–150 per month in your cloud environment. This is your cost, not billed by PanOps.

Inference Model

PanOps uses a PanOps-hosted AI model — an open-weights model tuned and validated for organizational intelligence tasks. Specific model identifiers are not disclosed in public documentation.

No External APIs

No third-party model APIs are called at any stage — not for inference, embedding, transcription, or summarization. The entire pipeline runs on self-hosted components inside your environment.

Understand how the security architecture works.Data isolation, encryption, and the zero-access model — in full detail.

View Security Architecture