Four purpose-built layers. No vendor abstractions. Every component exists because we built it for a specific reason.
The intelligence layer. Six specialized agents coordinated by a meta-planner that decomposes developer intent into structured execution plans.
Receives natural language requests, classifies complexity (simple/moderate/complex), routes to appropriate model tier, decomposes into sub-tasks.
ContractWriter, SecurityAuditor, DeploymentPlanner, TestGenerator, DocWriter, ChainAnalyst — each with specialized system prompts and tool access.
20+ tools: file operations, compiler invocation, chain queries, test runners, deployment scripts. Each tool has typed input/output schemas.
Suggest → Approve → Execute. AI generates proposals, humans review diffs, only approved actions execute. Full audit trail.
Routes requests based on complexity: simple → fast models (Haiku), moderate → balanced (Sonnet), complex → frontier (Opus). Cost optimization built-in.
Two React applications: a browser-based IDE for development and a dashboard for organization management. Both share a design system but are independently deployable.
Monaco-based editor with LSP integration, integrated terminal (xterm.js), file explorer, and persistent AI assistant panel.
7 modules: Projects, Workspaces, AI Command Center, Deployments, Analytics, Settings, Export/Portability.
6 modules: Org Management, AI Governance, Infrastructure Overview, Security Dashboard, Feature Flags, Forensic Tools.
Zustand stores per domain. No global god-store. Permission-aware rendering via usePermission hook and PermissionGuard component.
The coordination layer. Fastify API gateway handles auth, rate limiting, and routing. Go workspace manager handles container lifecycle and resource allocation.
JWT authentication, RBAC middleware, rate limiting (per-org, per-user), request validation, audit logging. All API routes go through here.
Container provisioning, resource allocation, health monitoring, automatic cleanup. Communicates with Kubernetes API for pod lifecycle.
Typed event system with 35+ event types across 11 domains. Pub/sub for decoupled service communication. WebSocket bridge for real-time frontend updates.
PostgreSQL with row-level security per tenant. Drizzle ORM for type-safe queries. Migration management with versioned schemas.
Managed blockchain infrastructure. RPC nodes, chain validators, deployment pipelines, and monitoring. Multi-cloud with Kubernetes orchestration.
Dedicated and shared RPC endpoints for 8+ chains. Health monitoring, automatic failover, load balancing. Managed Solana validators for devnet.
Chain-specific deployment strategies: Anchor for Solana, Hardhat/Foundry for EVM. Gas estimation, simulation, verification, and post-deploy monitoring.
Each workspace is an isolated Kubernetes pod with network policies, resource limits (CPU/memory/storage), and automatic garbage collection.
Prometheus metrics, Grafana dashboards, structured logging, distributed tracing. Health checks at every layer with circuit breakers.
Every request is authenticated, authorized, rate-limited, and audit-logged. No implicit trust between services. Network policies isolate all workspace pods.
Row-level security in PostgreSQL. Separate Kubernetes namespaces per org. Network policies prevent cross-tenant communication. AI context is workspace-scoped.
Export your entire workspace — code, configs, CI/CD, deployment scripts — in standard formats. Use our infrastructure, but never depend on it.
