Sovereign Command Center

AI coding assistants lock teams into cloud SaaS with per-token costs, vendor lock-in, and data-privacy risk, a non-starter for proprietary codebases that can't be shipped to a third party.

• An orchestrator that routes between local Ollama models and optional cloud escalation, with per-org spend caps and cost estimation
• A six-stage code-safety pipeline: Crucible (Docker sandbox test) → Burn Book anti-pattern detection → review agent → approval token → commit
• A vector-backed Vault that indexes the whole workspace via AST analysis (ts-morph) for repo-wide context
• Ed25519 org-bound licensing and a privacy-first referral engine (HMAC contact matching), covered by 88 tests
• An MCP tool server that exposes the platform's capabilities to agents

An npm-workspaces monorepo. A React 18 + Vite operator UI talks to an Express + Mongoose API. Local services are orchestrated with Docker: MongoDB, Ollama for LLM inference, ComfyUI for image generation, and an MCP server. The UI uses a Monaco editor and Graphology + Sigma for graph visualization; Zod validates throughout.

A local-first AI coding platform whose data never leaves the developer's machine: self-hosted, multi-model LLM orchestration for proprietary codebases that can't be shipped to a cloud vendor. It's my production multiplier: how I ship whole systems solo, faster and safer than teams renting per-seat cloud AI. The intended model is a one-time, org-bound license rather than a subscription. Nothing has been sold: this is a single-operator platform I run myself, and the licensing is built, not proven.