vis-Neuclea: Scoped Agent Infrastructure for APIs

THE WEB WASN'T BUILT FOR AGENTS Modern web architecture was engineered for humans. Forcing AI agents into this model creates three massive bottlenecks: 1. Financial Waste: Traditional agent frameworks cycle through bloated loops, burning thousands of unnecessary tokens per request. 2. Security Risks: Existing solutions require storing sensitive credentials on third-party backend servers. 3. Integration Friction: Every API requires a custom, hardcoded wrapper, slowing development. THE SOLUTION: SCOPED AGENT INFRASTRUCTURE vis-Neuclea introduces a sandboxed agent paradigm where the LLM is strictly limited to: * Tool Selection: Intelligently determining exactly which API endpoint to call. * Response Formatting: Polishing retrieved data to present cleanly to the user. All non-essential computation is rejected by design, keeping execution fast and predictable. CORE ARCHITECTURE 1. vis (MCP Server Generator) An automated pipeline that ingests any OpenAPI 3 spec (JSON/YAML/URL) and outputs production-ready Model Context Protocol (MCP) servers. For enterprise readiness, it features a containerized Observability Stack packaging metrics/logs via Prometheus and Loki, distributed tracing via Tempo and Promtail, and live analytics through Grafana. 2. Neuclea (Scoped Agent Orchestrator) A high-performance Go backend managing the real-time loop between the frontend, LLM, and MCP servers. It uses WebSockets to stream reasoning steps to the user console and switches between local execution (Ollama) or scalable cloud setups (Fireworks AI). THE RESULTS * 2x Token Efficiency: Cut total LLM token usage by ~50% compared to standard, unconstrained loops. * Instant API Onboarding: Standardized discovery into a lightweight agents.json root file. * True Observability: Out-of-the-box telemetry providing clear enterprise governance. TECH STACK Go, React, TypeScript, GLM 5.2 (Fireworks AI), Ollama, MCP, WebSockets, OpenAPI 3, Prometheus, Grafana, Loki, Tempo, Promtail.