Coretext
AI Tooling Platform
Fractional CTO for an AI tooling startup, joined immediately post-funding. Built team and product from zero in 6 months with 3 engineers. Two products: Sidekick (desktop MCP client built personally with Tauri 2 + Rust) and Console (web app for configuring and building custom MCP servers). Implemented AI-first team practices that produced 2-3x velocity gains engineers carried to future roles. Company closed abruptly (CEO personal reasons) — retained all IP and releasing Sidekick as open source.
What Was Built
Multi-workspace interface with isolated state per workspace
Multi-MCP server connections with tool namespacing
Real-time streaming chat with full conversation context
File management (drag and drop, Google Drive via MCP)
Dynamic theming and gesture navigation
Console: web app to configure and build custom MCP servers, context engineer them for specific needs
Template Orchestrator: AWS Lambda with complexity-adaptive routing (SQS chains vs. Step Functions)
How It's Built
Tauri 2 framework with Rust backend for desktop app
React 18 + TypeScript + Vite 6 frontend
SQLite local persistence per workspace
rmcp SDK v0.2.1 for MCP protocol handling
OAuth 2.0 with PKCE, JWT auth
NestJS v11 + GraphQL + Apollo Server v4 backend
PostgreSQL (Prisma) + DynamoDB + Elasticsearch data layer
Multi-tenant SaaS with RBAC and circuit breakers
AWS Lambda + CDK v2 infrastructure as code
Terraform IaC, ECS Fargate, RDS Multi-AZ, CloudFront CDN
Technical Stack
Results & Impact
Near-complete ecosystem built with 3 engineers in 6 months
Engineers self-reported 2-3x capacity vs. previous workflows
AI-first practices carried to future roles by team members
Automated bug-fix pipelines: agents pull, branch, fix, push, create PR for human review
Retained all IP — releasing Sidekick as open source