Private Beta

Observability Controller

Evaluate workflow readiness before expensive reasoning, retrieval, tool use or agent execution begins.

A lightweight infrastructure layer that sits ahead of AI models, RAG systems and agent workflows.

Without controller Standard LLM workflow
C
Customer · 10:30

My deployment failed.

AI
AI Assistant · 10:30

There are several possible causes: network issue, configuration, permissions, recent deployment or resource exhaustion. Checking documentation...

≈ 820 tokens
C
Customer · 10:31

It’s the payments service. Production. Started after upgrading to v1.32. CrashLoopBackOff.

AI
AI Assistant · 10:31

Thanks. That changes the diagnosis. Let me check the logs and common causes for this error in EKS...

≈ 760 tokens
Total ≈ 1,580 tokens Clarification occurred after reasoning began.
With Observability Controller Clarify first. Reason once.
C
Customer · 10:30

My deployment failed.

OC
Observability Controller · 10:30

Before we start, I need a few more details. Which service is affected? Which environment? What changed recently? Any error messages?

≈ 40 tokens
C
Customer · 10:31

Service: payments API. Environment: production EKS. Changed: updated to v1.32. Error: CrashLoopBackOff.

OC
Observability Controller · 10:31

Request sufficient. Proceeding to assistant.

AI
AI Assistant · 10:31

Thanks. Based on the deployment details, here’s what’s happening and how to resolve it...

≈ 860 tokens
Total ≈ 900 tokens ≈43% fewer workflow tokens in controlled evaluation.

Evaluate first. Execute when ready.

Many AI workflows begin reasoning before they know whether a request contains enough information to produce a useful result.

The Observability Controller acts as a lightweight decision layer before downstream execution. It evaluates request sufficiency and routes requests to clarify or proceed before models, retrieval systems, tools or agents consume additional compute.

The controller is model agnostic and works alongside existing AI infrastructure. It integrates ahead of AI models, RAG systems, agent workflows and internal assistants through a simple API, allowing controlled evaluation before wider production deployment.

Controlled evaluation results

Beta benchmark

Request readiness evaluation

Controlled evaluation across labelled support, troubleshooting, clarification and workflow-efficiency benchmarks.

Request full benchmark results
Controlled evaluation 97.3% Real-world benchmark accuracy 43% Workflow token reduction 107 Labelled evaluations

Results shown are from controlled evaluation environments. Independent production validation is ongoing.

Where it applies

Workflow
Typical issue
Outcome
Customer support
Incomplete tickets or missing troubleshooting context.
Clarify requests before investigation begins.
RAG systems
Retrieval begins before the request contains enough information.
Evaluate request sufficiency before retrieval executes.
Agent workflows
Agents plan, call tools or execute actions from ambiguous instructions.
Pause execution until sufficient context exists.
Operational triage
Incident reports lack scope, severity or reproduction detail.
Improve request quality before analysis begins.

Data handling

The controller evaluates each request and returns a routing decision such as proceed or clarify.

The current beta API is stateless by design. It does not require conversation memory or stored prompt history and does not intentionally retain prompts between requests.

Request evaluation access

Discuss workflow fit, benchmark results, private beta access or a controlled evaluation with the FoundScript team.

Request beta access Request benchmark results