Operator Console
AI-powered business operating system
269,000+ lines of code · 494 endpoints · 41 agents · 10 domains · 2,000+ vault notes
A complete business automation platform built from the ground up as a solo project by a marketing and e-commerce professional with no formal CS background.
A single-page application with 12 navigation tabs providing real-time oversight across every business function.
Each domain is fully operational with its own routes, database models, agent coverage, and UI integration inside the operator console.
Elektra is not a chatbot. It is a structured AI agent with awareness, memory, behavioral flags, and proactive capabilities. It delegates tasks to 41 specialized sub-agents across 3 execution backends and 9 LLM providers.
Every autonomous action is bounded by governance rules, cost tracking, and quality gates. The system is designed to be safe by default.
Multi-model deliberation for high-stakes decisions, combining outputs from multiple LLM providers.
Continuous risk monitoring with emergency halt capability when drawdown thresholds are breached.
All published content requires an 8+ quality score. No automated publishing without passing the gate.
Full execution history with per-call LLM cost tracking. Every agent run is logged and traceable.
End-to-end content lifecycle management across 11 platforms and 3 active brands. From topic scouting to batch publishing, every stage is automated.
Purpose-selected tools, each earning its place through proven utility in a production system running real business operations.
Identified gaps and the open-source solutions queued to close them - all free, all self-hostable, all compatible with the existing stack.
Self-hosted tracing across all 41 agents. Proxy-based cost tracking with zero code changes. Diagnoses why KPI tracking reads zero and which agents silently fail.
Cross-session persistent memory with ChromaDB vector search. Vault RAG grounds retrieval in 2,000+ notes. Semantic search replaces keyword matching.
All 41 agents exposed as MCP tools. System callable from Claude Code, Claude Desktop, Cursor, and any MCP-compatible client.
Radial dendrogram mind map and 3D force-directed graph for 2,000+ vault notes. Virtual scrolling renders only visible items for 10x faster navigation.
Replace the interval-based scheduler with retry-safe, observable Prefect flows. LangGraph for complex multi-agent coordination - each of the 41 agents becomes a stateful graph node.
Zero-cost local inference via 8 installed models. Ollama promoted to 3rd in the LLM fallback chain with native streaming. Qwen2.5 7B for reasoning, gemma3 4B fast fallback, nomic-embed-text for vault RAG embeddings.
Stanford DSPy systematically optimizes prompts across the SEO and blog pipelines - replacing trial-and-error prompt engineering with measurable, reproducible quality improvements.