nyxcore-systems — open source
Open-source tools for the layer beneath your editor — live code graphs, architectural analysis, LLM-native format handling, AI-powered shell, multi-agent orchestration, and a keyboard-first file manager.
nyxCore-Systems on GitHubProduction-grade dashboard for the nyxCore platform.
Full-stack internal dashboard built on Next.js 14 App Router with tRPC, PostgreSQL,
and Prisma. Auth via NextAuth v5 with JWT and Resend email. Real-time updates over
SSE — no WebSocket overhead. Rate limiting via Redis token bucket. REST + SSE
external API under /api/v1.
Vitest unit tests, Playwright e2e, row-level security enforced at the database layer.
Multi-tenant RAG platform. API-first, provider-agnostic, embeddable.
Modular retrieval-augmented generation platform with strict multi-tenant data isolation. Swap LLM providers via LiteLLM — OpenAI, Anthropic, Google, local models. Streaming chat over SSE, HMAC-signed webhooks, scheduled source re-ingestion, conversation export (JSON/CSV), feedback analytics, and an embeddable Shadow DOM widget that drops in with a single script tag. FastAPI backend, Qdrant vectors, Redis + ARQ worker queue, no-build Alpine.js dashboard.
The live code intelligence layer your tools are missing.
A persistent, incremental code intelligence daemon that keeps a live, queryable graph of your entire repository and updates only the blast radius of each change — the files and symbols actually affected. An open protocol for the layer every AI coding tool builds privately: cross-session persistence, sub-millisecond freshness, federated dependency indexing, and a first-class MCP server for AI agent integration.
The GPS with traffic data for your codebase.
A structural intelligence engine that maps your codebase's architecture, monitors its health, and predicts the ripple effects of changes before you make them. Detects architectural bottlenecks via bridgeness centrality, enforces layer boundaries with configurable rules, scores health 0–100, and tracks structural evolution over time.
Split structured content at the right seam for LLM compression.
Zero-dependency TypeScript library that detects format across 26 types — JSON, XML, HTML, YAML, TOML, MDX, Markdown, CSV, Dockerfile, and more — then splits content into what must stay verbatim versus what can be compressed. Designed for LLM pipelines that receive structured documents: keeps the schema, shrinks the prose. 226 tests. Works everywhere JavaScript runs.
Talk to your terminal.
Natural language to shell commands with no subscriptions, cloud dependency, or login required. Six modes: Ask, Explain, Fix, Debug, Build, and Chat. Safety-first: risk analysis and command preview before any execution. Fully offline via Ollama, or connect your own API keys. Ships as a single 7 MB binary. Includes an MCP server for Claude Desktop and Cursor, plus a workflow marketplace of pre-built DevOps playbooks.
Orchestrate LLM agents without losing control of cost.
Budget-first multi-agent orchestration library for TypeScript. Coordinate agents through typed, composable patterns — pipeline, parallel, debate, and more — with hard cost limits, conflict resolution, and adaptive routing. Library, not framework: zero lock-in, works standalone or with ANCS as its memory backend. Ships as an npm package with full TypeScript 5.5+ types.
Two panels, full keyboard, no Electron.
Keyboard-first dual-pane file manager for macOS and Linux in the tradition of Total Commander
and FAR Manager — native Rust backend via Tauri 2, Svelte 5 frontend. Two panels with
independent tab stacks, archive browsing, hex and diff viewers, a live process explorer, and
a plugin system that hot-reloads from ~/.nyx/plugins/.
Every AI coding assistant, every static analysis tool, every architectural review builds its own private model of your codebase from scratch. There's no shared substrate — no live, queryable graph that persists across sessions and updates only what changed.
That's the layer we work on. Persistent code graphs. Structural health analysis. LLM-native format handling. Multi-agent orchestration with hard cost limits. AI-powered shell that works fully offline.
Everything we ship is open source, documented, and designed to compose. We don't build platforms or lock-in — we build primitives that other tools can integrate with. If we build an MCP server, it works with any client. If we define a graph protocol, anyone can implement it.
We're a small team and we build carefully. For teams that want the intelligence layer fully managed, there's nyxcore.cloud — our hosted platform.
Martyna Kwiecień
Data Sovereignty Counsel