Architecture Overview

Layerr is not a router. It is an orchestration runtime, a self-hosted brain between your coding agents and every LLM provider. The system is built from 1,090 source files organized into 715 functional communities (auto-detected clusters of tightly coupled code).

Layerr Architecture

#	Layer	What it does	Key files / Communities
1	Intent Classification	Determines what the incoming request is asking for	orchestration/orchestrator.ts
2	Workload Analysis	Scans the codebase to infer language, stack, complexity	repo/intelligence/scanner.ts, repo/intelligence/profiler.ts
3	Workspace Profiles	Holds per-workspace preferences, provider configs, budgets	src/features/workspaces/health/WorkspaceHealthView.tsx
4	Strategy Engine	Chooses the orchestration strategy (cost, speed, quality, balanced)	routing/enrichment.ts
5	Scoring	Scores candidate providers and models against the workload	evaluation/quality/engine.ts
6	Execution	Sends the request and handles streaming responses	server.ts (all routes flow through here)
7	Explainability	Explains why a model was chosen, with structured rationales	explainability/explainer.ts
8	Replay	Records every decision, allows trace comparison and replay	replay/comparator.ts, storage/replay.ts
9	Adaptive Learning	Learns from outcomes and adapts future routing decisions	src/features/intelligence/hooks.ts
10	Quality Evaluation	Benchmarks, calibrations, coding validators	evaluation/calibration/engine.ts, evaluation/coding/validators.ts
11	Economics	Cost dashboards, savings computation, budget protection	src/features/providers/economics/ProviderEconomicsOverlay.tsx
12	Entitlements & Governance	Access control, license management, guardrails, rate limits	security/rotation/engine.ts

Top Functional Communities

These are the largest auto-detected clusters in the codebase:

Community	Symbols	Cohesion	Description
Secrets	96	0.87	Credential management, key rotation, provider auth
Hooks	78	0.98	React hooks for data fetching and UI state
Intelligence	70	0.71	Adaptive learning, calibration, workspace health
Economics	35	0.83	Cost tracking, savings, budget pressure
Motion	32	0.90	Animation and UI motion preferences
Licensing	29	0.80	License activation, validation, snapshots
RateLimit	29	0.84	Per-provider rate limiting
Providers	24	0.81	Provider registration, resolution, health
Insights	21	0.94	Analytics and intelligence dashboards
Reliability	21	0.91	Fallback chains, retry logic, circuit breakers
Telemetry	21	0.91	Metrics, traces, observability
Scoring	19	0.77	Quality scoring, grading, confidence
Coding	19	0.98	Coding-specific evaluation and validators
Sessions	18	0.87	Session management and state
Explainability	17	0.90	Decision explanations and rationale building
Rotation	17	0.81	Secret rotation scheduling
Storage	17	0.98	Persistence layer for traces and replay
Workload	17	0.89	Workload analysis and classification

Request Flow

When a request hits /v1/chat/completions:

1. Gateway (server.ts) receives the request, validates auth
2. Intent Classification determines language, stack, security scope
3. Workload Analysis scans the project context via repo/intelligence/scanner.ts
4. Strategy Engine picks a strategy (cost-optimized, quality-first, etc.)
5. Scoring runs evaluation/quality/engine.ts to score candidates
6. Provider Resolution resolves the best provider chain (orchestration/fallback.ts)
7. Execution proxies to the chosen model
8. Explainability builds a structured rationale (explainability/explainer.ts)
9. Replay stores the trace (storage/replay.ts)
10. Adaptive Learning feeds the outcome back into future scoring (src/features/intelligence/hooks.ts)

All of this happens in a single request cycle, with full observability at every step.