Skip to content

LLM Assessment

Coming soon

LLM Assessment is on the roadmap (milestone #0045). The description below reflects the planned capability based on the current design specification.

LLM Assessment evaluates large language models as the subject of a sovereignty assessment, rather than using them as analysis tools. It connects to multiple LLM providers simultaneously and benchmarks each against the SWAO_LLM_ASSESS community framework, producing a side-by-side comparison to support model selection decisions for sovereign workloads.

Planned CLI: swao assess --type llm

TUI: Main Menu > Run Assessment > [4] LLM Assessment (coming soon)


Why it matters

Before deploying any AI application to a sovereign environment, organisations must evaluate whether their chosen LLM meets the jurisdiction's requirements for data residency, transparency, safety, and regulatory compliance. Today this relies on vendor claims and informal testing. LLM Assessment makes it structured, repeatable, and auditor-defensible.


What it assesses

The SWAO_LLM_ASSESS framework covers seven domains:

DomainWhat it evaluates
Data sovereignty and residencyContractual data residency guarantees, zero-retention endpoints, EU hosting options, GDPR DPA compliance
Model transparencyPublished model card, EU AI Act conformity assessment, per-call audit metadata, licensing terms
Performance and reliabilityLatency, hallucination rate, context window, token throughput, cost
Cultural and language fitTarget-language fluency, domain vocabulary accuracy, regulatory body awareness
Safety and biasBOLD/WinoBias benchmarks, red-team disclosure, EU AI Act prohibited-use test, prompt injection resistance, jailbreak resilience
Agentic capabilitiesTool calling reliability, multi-step orchestration, RAG grounding fidelity, memory coherence, structured output compliance
People and team readinessLLM engineering skills, data management, AI governance roles, continuous learning commitment

What it produces

OutputWhat it contains
Sovereignty Readiness Score (SRS)A 0-100 score per model, weighted by domain
Side-by-side comparison tableEach framework control with RAG status (pass/partial/fail) for every assessed model
Regulatory fit summaryWhich GDPR-linked controls pass for each model
Benchmark reportLatency, throughput, cost estimates, and grounding test results

Configuration

LLM Assessment is configured in .swao.yml under the llm_assessment: block, where you specify which models to assess, thresholds, target locale, and budget ceiling. API keys are read from environment variables -- never stored in the configuration file.

Released under the Apache 2.0 Licence.