PF SystemsOperational authority for agentic AIStart with one workflow Information. Evidence. Action.
PF Systems helps you manage the information your systems use, keep evidence of what happened and control which actions can go ahead.
UK sovereign · model-agnostic · designed to work with existing systemsProvides relevant information and tracks where it came from.
02 · PF CoreLinks data and records so they can be checked and audited.
03 · PF KernelChecks whether an action is allowed before it happens.
One component, a pair, or all three as PF OS.
PF OS in motion
Follow an AI request from the information it uses, through the authority it needs, to the evidence it leaves.
Start with what you need.
Use PF Memory for governed information, PF Core for linked evidence, or PF Kernel for action authority. Combine them to suit your workflow. You don’t have to deploy everything at once.
0:00 · Your information
Supplier records, working procedures and operational data surround a simple task: order 12 replacement sensors. This film uses illustrative records, an example supplier and a £1,800 purchase request.
0:07 · PF Memory
The assistant receives the approved supplier reference, the current £2,000 spending limit and the stock requirement. PF Memory manages the information and its provenance; it does not decide or execute the purchase.
0:11 · AI action request
The proposed purchase names its supplier, quantity, amount and intended action. A model’s recommendation remains a request: producing it does not create permission to act.
0:17 · PF Kernel
PF Kernel checks the proposed action against the supplied authority and returns Allow, Deny, Modify, Step Up or Stop the Line. The film follows an Allow result for the illustrative £1,800 purchase. The protected execution path must apply that decision.
0:21 · Your system
The purchasing system creates the example order and supplies its own outcome record. An authorisation record is distinct from evidence of execution: an Allow decision alone does not prove that the order was placed.
0:26 · PF Core
PF Core tags and links data, source references, decisions and supplied outcome records into machine-readable evidence. This supports integrity checking, audit and accountability. Where the necessary material is retained, the captured governed evaluation can be deterministically replayed. This does not make AI generation deterministic or automatically repeat a live action.
0:37 · Your choice
PF Memory can supply governed context on its own. PF Core can link and preserve evidence for existing records. PF Kernel can evaluate action authority using supplied context. The full workflow shown in the film is one combination, not a mandatory technical pipeline.
Available combinations: PF Core alone; PF Memory alone; PF Kernel alone; PF Memory + PF Core; PF Memory + PF Kernel; PF Kernel + PF Core; All three · PF OS.
PF OS is only PF Memory, PF Core and PF Kernel. Memory knows. Core proves. Kernel decides. ClientBridge connects, Proof Harness qualifies evidence and PF Trace displays it.
This is a conceptual process film using illustrative data. These are not customer deployments. Discovery and demonstration are supportable; a controlled shadow pilot with no live action is the recommended next gate. Production use is a separate decision.
Authorisation and execution evidence remain distinct.
PF Systems can sit around existing AI models, agent frameworks, memory systems, identity services and operational software.
One governed cyber action
A synthetic incident-response agent is authorised to quarantine one endpoint. Before execution, it substitutes a critical finance server. The earlier authority no longer matches.
Quarantine DEV-481
Proposed cyber responsePF Kernel decides
No action proceeds before a decision.
Allow · Deny · Modify · Step Up · StopAn AI agent asks to quarantine endpoint DEV-481.
Illustrative synthetic scenario. It does not represent a customer deployment, certification, guaranteed security or production readiness.
The PF OS triad
PF OS contains only PF Memory, PF Core and PF Kernel. They can work independently or together without turning knowledge into authority or evidence into a decision-maker.
Manages relevant context, origin, permission and lifecycle. It can wrap an existing memory system; it does not decide or execute.
Explore PF Memory ↗02 · PF Core™ProvesLinks supplied data, records, source references and tags. Preserves machine-readable evidence and receipts to support integrity checking, audit and accountability. It does not decide.
Explore PF Core ↗03 · PF Kernel™DecidesReturns Allow, Deny, Modify, Step Up or Stop the Line before the proposed action takes effect.
Explore PF Kernel ↗PF Core preserves linked, tamper-evident evidence and the captured material needed to trace, check and deterministically replay a governed evaluation, where that material is retained. A record of authorisation is distinct from evidence of execution: the executing system must supply evidence of what actually became effective.
Replay does not make the underlying probabilistic AI deterministic or automatically repeat a live action. Discovery and demonstration are the starting point; a controlled shadow pilot with no live action is the next gate. Production use is a separate decision.
Where PF Systems begins
PF Systems works alongside IAM, model safeguards and existing security controls. Its role is to apply organisational authority when a proposed action could create a real consequence.
Choose your level of detail
Connects the existing estate.
Proof Harness™Qualifies against available evidence.
PF Trace™Displays received decision evidence.
These vendor-sponsored surveys indicate a direction of travel; they do not represent every organisation or prove PF Systems performance.
A controlled first step
A shadow pilot runs beside existing work. PF Systems evaluates what the AI would be allowed to do, but does not take the live action.
Governed workflow briefing
Tell us the action, its consequence and where authority currently sits. We will help shape a practical discovery or non-actuating shadow-pilot boundary.
