Skip to content
parelyxArchitecture

A decision layer with clear boundaries.

Parelyx separates a bounded, latency-critical live path from an asynchronous lane where evidence is collected and policies are evaluated, validated and rolled out. It sits above your inference infrastructure — it does not replace it.

01Functional map

Live path and learning lane.

Select any component to see what it is responsible for and where its boundary lies.

Functional architectureLive path · offline lane · your infrastructure
Live inference pathbounded, latency-critical
Intake
Decide
Execute
Serve

Outside Parelyx — yours or your providers'.

Return
Asynchronous & offlinenever on the request path
Live path

Decision engine

Responsibility

Chooses an execution strategy for the request — including direct execution or safe bypass.

Boundary

Operates within a bounded decision budget. Applies the current validated policy; does not learn on the live path.

02Where it sits

Above serving. Below your application.

Applications keep their model API. Infrastructure keeps its scheduling, batching and acceleration. Parelyx adds the decision in between — and the learning loop around it.

  1. AI applications, agents & RAG systems

    Your software. Integrates through a familiar model API; no application redesign.

    • Agents
    • RAG
    • Copilots
    • AI APIs
  2. Parelyx adaptive intelligence layer

    Decides how each request should be executed, within a bounded decision budget.

    • Workload intelligence
    • Latency-aware decisions
    • Model & provider selection
    • Cost & quality constraints
    • Selective optimisation
    • Continuous policy evaluation
  3. Inference infrastructure

    Owned by you or your providers. Handles serving, batching, scheduling and acceleration.

    • Compatible runtimes
    • Model providers
    • Distributed serving
    • Accelerated execution
  4. Model execution

    The selected model produces the response.

  5. Workload outcomes & feedback

    Latency, cost, quality signals and errors — recorded per decision.

  6. Parelyx workload learning

    Offline evaluation and refinement of decision policies. Outside the live path.

    • Offline
    • Validated before rollout
03Design properties

Properties the architecture is built around.

  1. 01

    Fail-open by design

    A fault inside Parelyx routes the request to its default execution path instead of failing it.

  2. 02

    Bounded decision budget

    The live decision has its own time budget. If it cannot decide in time, it steps aside.

  3. 03

    Evidence off the response path

    Decisions and outcomes are recorded without holding up the response.

  4. 04

    Explicit configuration wins

    Pinned models and disabled optimisations take precedence over any policy.

  5. 05

    Reversible policy changes

    Policy changes are versioned, validated, introduced gradually and can be rolled back.

  6. 06

    Customer-controlled data

    Outcome data is scoped and retained according to customer policy.

04Responsibility boundaries

What Parelyx does — and doesn't.

Parelyx is responsible for
  • Which execution strategy a request should take
  • Which model or provider best fits its constraints
  • Whether an optimisation is worth applying at all
  • Which outcomes were achieved, and how policies should improve
Your inference infrastructure is responsible for
  • GPU scheduling and batching
  • KV-cache and memory management
  • Distributed and disaggregated serving
  • Kernel-level acceleration
05Deployment

Runs where your models run.

Parelyx runs in your environment, alongside your applications and models — so requests, outcome data and policies stay within your boundary.

Deployment topology, data retention and access controls are agreed with each customer as part of onboarding.