Studio
EA Intelligence Under the Hood Run History

EA Studio · Governance-Grade AI for Enterprise Architecture

Enterprise architecture,
generated with a paper trail.

EA Studio drafts and debates architecture artifacts on demand — then validates them, hashes them into a tamper-evident ledger, and prices every run. Two lanes: a fast generative lane for template-shaped work, and a deliberative six-advocate lane for high-stakes trade-offs.

Open EA Intelligence Suite

Separate page for daily trend chips, trusted Q&A, and RSS subscriptions.

Model Router

llm-tiebreaker

API keys & env vars ›
NVIDIA NIM ⭐NVIDIA_NIM_API_KEY — build.nvidia.com/free-credits (150+ models with one key) OpenAIOPENAI_API_KEY AnthropicANTHROPIC_API_KEY DeepSeekDEEPSEEK_API_KEY — platform.deepseek.com Kimi / MoonshotMOONSHOT_API_KEY — platform.moonshot.cn Qwen / AlibabaQWEN_API_KEY — dashscope.aliyuncs.com (also set DASHSCOPE_API_KEY) Google GeminiGEMINI_API_KEY — aistudio.google.com
Who

Enterprise architects and CTOs at institutions where every AI output needs an audit trail.

Regulated finance, multilateral development, healthcare, public sector.

Problem

AI outputs today are unaccountable. Copilots hallucinate; consultants leave.

Decisions get made without recorded rationale, sources, or defensibility.

Why now

Board-level scrutiny on AI meets executive appetite for speed.

You need artifacts that survive audit — and cost pennies, not billable weeks.

How it works
01 · INTAKE
You describe the ask

Natural language, or click a pathway card below.

02 · ROUTE
A classifier picks the lane

Keyword signal + optional LLM tiebreaker. You can override.

03 · GENERATE
Lane A drafts · Lane B debates

Template + RAG for A. Six advocates + guardrails + judge for B.

04 · PROVE
Validate · hash · price

MADR + WAF checks, chain hash, tokens & cost per run.

Choose a pathway

What do you need to decide, defend, or document?

Fast Draft — structured artifacts (fast, low cost). | Committee Debate — multi-agent trade-off analysis (slower, defensible).

Decision basis: if your ask is to draft or assess one artifact, use Fast Draft. If your ask is to compare options and justify a choice, use Committee Debate.

Architecture Decision Record Fast Draft (A)

You've made a decision and need to write it down with context, options, and consequences — MADR 3.0 format.

≈ $0.001  ·  ~5 s

ArchiMate + TOGAF ADR Fast Draft (A)

You need an ADR with a model-first ArchiMate workflow and structured chain prompting for TOGAF views.

≈ $0.001  ·  ~5 s

Well-Architected Review Fast Draft (A)

You need a workload scored across the five WAF pillars (Reliability, Security, Cost, Operational Excellence, Performance).

≈ $0.001  ·  ~5 s

Framework Selection Committee Debate (B)

You need to pick between agent orchestration frameworks (or any comparable platform set) with a defensible scorecard.

≈ $0.08  ·  ~60–90 s

Migration Study Committee Debate (B)

You're moving a workload to Azure and need a debated recommendation across Functions / ACA / App Service / AKS / VMs.

≈ $0.08  ·  ~60–90 s

Integration Pattern Committee Debate (B)

You're choosing between REST, event-driven, queue, iPaaS, streaming, or GraphQL federation for a system-to-system link.

≈ $0.08  ·  ~60–90 s

Landing Zone Committee Debate (B)

You need to pick a subscription / network topology (ALZ, Virtual WAN, hub-spoke, sovereign) for a new estate.

≈ $0.08  ·  ~60–90 s

Selected
RAG Pattern Committee Debate (B)

You need to choose a retrieval architecture — naive vector, hybrid, reranker, HyDE, GraphRAG, agentic, or long-context.

≈ $0.08  ·  ~60–90 s

Pathway workspace

RAG Pattern

You need to choose a retrieval architecture — naive vector, hybrid, reranker, HyDE, GraphRAG, agentic, or long-context.

mode Committee Debate (B)  ·  est. ≈ $0.08 · ~60–90 s

Crew context for this lane

What these labels mean: A## are option advocates, GR is the governance/guardrails reviewer, and JD is the decision judge.

A01 Naive vector RAG Advocate
A02 Hybrid retrieval (BM25 + vector) Advocate
A03 Reranker-augmented RAG Advocate
A04 HyDE / query-rewriting RAG Advocate
A05 GraphRAG (entity+relation index) Advocate
A06 Agentic RAG (tool-using agents) Advocate
A07 Long-context no-RAG Advocate
GR Responsible-AI / IAM / Observability Reviewer
JD Enterprise Architecture Judge
Role responsibilities (plain English)
  • A##: each advocate argues one candidate pattern with benefits, risks, and fit assumptions.
  • GR: checks responsible AI, IAM, observability, policy, and compliance implications.
  • JD: applies weighted criteria, resolves trade-offs, and recommends a decision with rationale.
What problem this multi-agent setup solves
  • Problem: one-agent responses can anchor on one option. Fix: advocates force explicit option-by-option debate.
  • Problem: governance concerns are missed late. Fix: GR inserts policy/IAM/RAI checks before recommendation.
  • Problem: decisions are hard to compare over time. Fix: JD scorecard makes trade-offs measurable and repeatable.
  • Problem: outputs are hard to defend in committee. Fix: validators + provenance + HITL create an auditable approval path.
Scoring dimensions
20% · answer quality 10% · freshness 10% · latency 15% · cost per query 10% · build complexity 10% · evaluation ease 10% · guardrail fit 15% · scalability
How to read the score weights
  • 20% answer quality — How well responses satisfy stakeholder intent and architecture quality standards.
  • 10% freshness — How current the retrieved or generated knowledge is relative to change velocity.
  • 10% latency — End-user and system response-time impact under realistic load.
  • 15% cost per query — FinOps impact per request at scale, not just pilot traffic.
  • 10% build complexity — Implementation and operational complexity for delivery teams.
  • 10% evaluation ease — How straightforward it is to test, compare, and audit outcomes.
  • 10% guardrail fit — Alignment with security, policy, responsible AI, and compliance controls.
  • 15% scalability — Ability to scale across business units, domains, and workload growth.
Interaction tracer preview (before you run)
  1. Step 1: Advocates A01..A0N generate competing recommendations with assumptions and trade-offs.
  2. Step 2: GR reviews each recommendation for policy, security, and responsible-AI compliance.
  3. Step 3: JD applies weighted scoring dimensions and ranks options.
  4. Step 4: Validator spine checks TOGAF principles, schema, and provenance readiness.
  5. Step 5: Final artifact is published with hash chain and executive bundle outputs.
Enterprise architecture value signal
  • Converts strategic asks into governed, traceable decision artifacts.
  • Balances innovation with risk, policy, security, and operational constraints.
  • Produces evidence-backed trade-off analysis suitable for architecture review boards.
LLM context composer

Your prompt stays pinned to the selected pathway context.

Lane selector: Auto picks the best lane for your prompt. Lane A is faster and lower-cost. Lane B runs a deeper multi-agent debate with guardrails and judge scoring.