FIT

Arkheia · Self-Qualification

Is This
For You?

Self-qualify in two minutes. Save everyone's time.

§ 01 · What You Need

Three things — that's it

Arkheia operates at the API boundary. Deploy as a cloud proxy — change one URL — or as an on-prem engine within your own infrastructure.

01

AI APIs in Production

You call LLM APIs — Claude, GPT-4, Gemini, or self-hosted models — in a production workflow. Arkheia sits between your application and the model.

02

An Observable API Layer

You can observe the API boundary — either by routing calls through the Arkheia cloud proxy, or by deploying the Arkheia engine within your own infrastructure.

03

A Use Case Where Reliability Matters

Finance, legal, healthcare, customer-facing automation, agent pipelines — anywhere fabricated outputs create real risk.

§ 02 · You Don't Need

Common blockers that aren't blockers

These are concerns teams raise. None of them prevent deployment.

Model Weight Access

Detection operates on observable API behaviour — token timing, response statistics, log-probabilities where available. We never need access to model internals or weights.

Special Hardware

Detection runs on the observable surface — timing, token statistics, response metadata. No GPU or specialised compute required.

To Change Your Models

Works with whatever you're already using — cloud APIs or self-hosted inference stacks.

To Share Your Data

We never see your prompts or responses. Detection is behavioural — not content-based.

§ 03 · Hard Stops

When it won't work

These are genuine blockers. If they apply, say so — we'd rather tell you now than waste a pilot.

No API boundary to observe

Arkheia works at the API layer — between your application and the model. If your AI runs as a fully embedded binary with no network interface (no HTTP calls, no API responses), there is nothing for the proxy to sit in front of. This is rare in production systems.

Strict air-gap with no on-prem deployment path

If your environment prohibits all network traffic routing — including within your own infrastructure — the cloud proxy cannot be used. An on-prem deployment resolves this in most cases; contact us if you have a true air-gap constraint.

Requirement to skip baseline characterisation

Detection profiles are built from your models, your use cases, your workload. A model we haven't characterised on your data cannot produce reliable risk signals. We don't deploy into production without a validation phase — and we won't pretend otherwise.

§ 04 · Strong Signal

You're a strong fit if…

Regulated Industry

Finance, healthcare, legal, government — anywhere AI fabrications create compliance exposure.

High-Stakes Decisions

AI output that influences medical, legal, financial, or safety-critical decisions.

Agent Pipelines

Multi-step agentic workflows where a single fabrication can cascade. Per-invocation detection catches drift early.

Privacy Requirements

GDPR, data sovereignty mandates, or internal policies that prohibit content leaving your infrastructure.

Claude Code, Cursor, or MCP Frameworks

Teams using AI coding tools or MCP-compatible agent frameworks — the MCP Trust Server provides tool-native detection with no proxy setup or endpoint changes required.

§ 05 · Pilot Process

What to expect

01

Week 1

Strategic Briefing

Technical deep dive with your team. Map use cases, API stack, and success criteria.

02

Weeks 2–3

Integration

Integrate Arkheia — either by routing API calls through the cloud proxy or deploying the on-prem engine. Configure model profiles and establish behavioural baselines.

03

Weeks 4–6

Validation

Run on real workloads. Measure detection quality, false positive rates, and latency overhead. Tune for your domain.

04

Weeks 7–8

Go / No-Go

Review results. Did we meet success criteria? Decide on production deployment — no commitment required before this.

§ You Qualify

Ready to start?

Schedule a strategic briefing. We'll map your requirements and confirm feasibility — no commitment needed.