Architecture intelligence for complex systems

Architecture Decision IntelligencePrivate design partner program

Before changing one service, know what else you're changing.

Wattra investigates code, contracts, tests, dependencies, configuration, and ADRs across repositories. It shows affected components, risks, and unknowns — then preserves the reviewed result for the next engineer or AI agent.

  • Evidence-linked
  • Explicit unknowns
  • Reviewable memory
  • Agent-neutral
How it works

One change request. A governed investigation. A reviewable result.

Illustrative investigation

Input & evidence

Change request

Add partial refunds across the payment system.

Evidence sources

  • PaymentController.cs

    Code

    Verified
  • refund-requested.schema.json

    Contract

    Verified
  • appsettings.Production.json

    Configuration

    Inferred
  • SettlementWorkerTests.cs

    Tests

    Unknown
  • ADR: Payment event compatibility

    ADR

    Stale

Architecture investigation

  1. 01Payments APIVerified
  2. 02Refund ContractVerified
  3. 03Event BusInferred
  4. 04Settlement WorkerInferred
  5. 05Ledger StoreStale
VerifiedInferredUnknownStale

Result

Open unknown

Legacy reconciliation flow has no confirmed owner.

Compatibility risk

Refund idempotency differs between contract versions.

Impact brief

5

affected components

3

contracts

8

relevant tests

2

open questions

1

compatibility risk

Reviewed result saved to Architecture Memory.

The shift

From repeated discovery to reusable architecture knowledge

Without Wattra

  1. 01Search repositories
  2. 02Ask senior engineers
  3. 03Read stale documentation
  4. 04Build a one-off explanation
  5. 05Repeat the investigation later

Context remains fragmented across tools, chats, documents, and people.

With Wattra

  1. 01Investigate evidence
  2. 02Understand architecture
  3. 03Expose risks and unknowns
  4. 04Review the result
  5. 05Preserve reusable context

Each accepted investigation improves the shared architecture understanding.

Platform

One architecture intelligence core. Every engineering decision.

Wattra does not build separate AI features that rediscover the system independently. Investigations, specifications, diagrams, meetings, and external agents work from the same reviewed architecture understanding.

Living Architecture Model

The shared, evidence-backed understanding every application draws from.

ComponentsBoundariesFlowsContractsIntentEvidenceConfidenceUnknownsFreshnessReview state
SelectedArchitecture InvestigationsCurrent focus

Input

A system question or proposed change

Output

Architecture & Impact Brief

5

components

3

contracts

8

tests

2

questions

The starting point. Every other application builds on a reviewed investigation.

Real-world workflows

One architecture intelligence core. Four critical workflows.

General-purpose AI can help engineers read code faster. Wattra turns that investigation into persistent, evidence-backed, freshness-aware architecture memory — then uses it to support decisions, specifications, diagrams, meetings, and implementation agents.

AI assistants accelerate a session. Wattra preserves and evolves the architectural understanding behind it.

Trace a core change across hundreds of repositories

A change to a shared core component can affect services, contracts, tests, deployment paths, and past architectural decisions across a large repository estate. Wattra investigates the relevant paths, preserves the evidence, and records what is fresh, stale, or still unknown.

Change requirement

What changes if the shared-core event contract evolves?

Investigation model:LocalCloudPersistent session — no rebuild per questionPlanInspectRecord evidenceEvaluate gaps

Repository estate

Scanning…
platform-coreorders-apibillingrisknotificationsanalyticscontract-tests
  • Confirmed by evidence
  • Inferred
  • Unknown / needs investigation
  • Stale context

Evidence ledger

  • shared-core/events/OrderSettled.cs
  • order-settled.schema.json
  • SettlementConsumerTests.cs
  • ADR-021 event versioning

Impact brief · Investigation Session saved

7

affected repos

3

event contracts

12

relevant tests

2

open unknowns

  • Fresh
  • Changed since investigation
  • Needs revalidation

Ad-hoc coding assistant

Re-scan repositories → rebuild context → spend tokens again → receive a point-in-time answer.

Wattra

Investigate → preserve evidence → track freshness → reuse and update architecture memory.

OutcomeImpact briefEvidence ledgerPersistent contextFreshness tracking

One continuous architecture loop

  1. Investigate
  2. Discuss
  3. Decide
  4. Specify
  5. Implement
  6. Re-index
  7. Learn
How Wattra runs

Model-flexible by design. Agentic under the hood. Open at the edges.

Wattra does more than send repository context to a single prompt. It plans an investigation, explores the relevant evidence, evaluates gaps, records findings, and updates persistent architecture memory. Teams can use the result through the Wattra interface, MCP server, or REST API.

The model is replaceable. Repository intelligence is the product.

Choose a local or cloud LLM according to your infrastructure and data boundaries. Wattra orchestrates an iterative, evidence-governed investigation flow and exposes architecture context through MCP and REST.

Model runtime

Consumers

  • Engineers
  • Architecture teams
  • Meeting copilot
  • Coding agents
  • Internal engineering tools

Interfaces

  • Wattra UI
  • MCP Server
  • REST API

Wattra agentic runtime

Orchestration
Persistent investigation session

Governed loop — repeats until evidence is sufficient

  1. 1Plan
  2. 2Inspect
  3. 3Record evidence
  4. 4Evaluate gaps

Then conclude

  1. Conclude
  2. Generate artifact
  3. Update memory
Replaceable model runtime
Local LLMselected
Cloud LLM

Evidence sources

  • Source code
  • Multiple repositories
  • APIs & contracts
  • Tests
  • ADRs
  • Documentation
  • Configuration

Living Architecture Model

Persistent core
  • Evidence ledger
  • Decision memory
  • Freshness state
  • Investigation artifacts

Outputs

  • Impact brief
  • Change specification
  • User stories & tasks
  • Architecture diagrams
  • Meeting context
  • Context for external agents

Run investigations through a model endpoint in your environment.

Change the model without rebuilding the architecture intelligence layer.

Open at the edges: MCP & REST

MCP Server

Give coding agents evidence-backed architecture context and Wattra tools without manually copying investigation history between sessions.

Coding agent requests

Architecture context for a backlog task

Wattra returns

  • Relevant specification
  • Affected components
  • Architectural constraints
  • Source evidence
  • Current unknowns

Agent handoff

  1. 1An external coding agent receives a backlog task
  2. 2The agent requests architecture context through Wattra MCP
  3. 3Wattra returns the specification, affected components, constraints, evidence, and unknowns
  4. 4The agent starts implementation with that context
  5. 5After the change merges and repositories are re-indexed, Wattra updates the architecture model

External agents consume Wattra's architecture context instead of rediscovering it for every task.

How the layers relate

  1. Local or cloud model
  2. Wattra agentic investigation
  3. Evidence-backed architecture memory
  4. UI · MCP · REST · Artifacts

The model is an outer, swappable dependency. Wattra's agentic investigation and its evidence-backed architecture memory are the persistent product.

Evidence & governance

Architecture memory you can inspect, review, and trust

Wattra does not merely answer questions about code. It creates and continuously updates evidence-backed architecture memory that teams and agents can use to investigate, decide, specify, implement, and learn.

Architecture claim

Settlement Worker consumes RefundRequested events.

  • VerifiedHandler and passing test confirm consumption.
  • InferredRetry path inferred from configuration.
  • UnknownDead-letter behaviour is unconfirmed.
  • StaleADR predates the last contract change.
  • ContradictedTwo contract versions disagree.
SettlementWorker.csrefund-requested.schema.jsonSettlementWorkerTests.csADR-014rev a1b2c3d

Wattra keeps facts, assumptions, uncertainty, and stale knowledge visibly different.

How the knowledge stays trustworthy

  • Evidence over plausibility
  • Explicit uncertainty
  • Reviewed and versioned knowledge
  • Neutral context for existing agents
  • Human review before implementation

Honest product status

  • Change-impact investigationCurrent focus
  • Spec & agent handoffNext
  • Living architecture diagramsNext
  • Architecture meeting copilotFuture vision

Roadmap statuses are indicative. Investigation is where Wattra is focused today; the other workflows build on the same architecture memory over time.

Have a change that crosses system boundaries?

We are looking for platform teams and modernization specialists willing to validate Wattra on real architecture investigations and change-impact decisions.