Projects

Built in the open

Alongside client work, we build our own products in public. NithronAI is open source and NithronOS is source-available, and the code and licence of each are on GitHub.

Nithron on GitHub

NithronOS (nOS)

Source-availablev0.9.5 pre-alpha, public preview

A Linux-based operating system for NAS and homelabs. Local-first storage management, snapshots, shares and backups, with a modern web dashboard and an optional app catalog, without cloud lock-in.

  • The admin dashboard is served on your local network by default, and remote access is opt-in.
  • Btrfs first, with snapshots and send and receive, and optional ZFS where the platform allows it.
  • Snapshots before every update, dry-run plans and an easy rollback.
  • One-click apps through Docker and Compose, and NithronSync for file sync across devices.
Inside NithronOS. The React dashboard in the browser, and the desktop and mobile sync apps, reach the system through Caddy, which serves the interface, proxies the API and offers WebDAV for sync. nosd, the Go API, handles disks, pools, snapshots, shares, sync, jobs and live health. Anything that needs root goes through nos-agent, a helper that accepts only allow-listed actions over a Unix socket. Underneath are Btrfs storage pools, Docker apps and an nftables firewall that only lets the local network in by default. systemd timers run snapshots and pruning, and health metrics refresh every second.BrowserReact dashboardSync appsDesktop and mobileCaddyServes the UIProxies the APIWebDAV for syncnosdREST API, OpenAPIPools and snapshotsShares, sync, jobsLive healthnos-agentRuns as rootUnix socketBtrfsPools, snapshotsDocker appsFirewallJobs and monitoringSnapshot timers, metrics every secondInside NithronOS. The React dashboard in the browser, and the desktop and mobile sync apps, reach the system through Caddy, which serves the interface, proxies the API and offers WebDAV for sync. nosd, the Go API, handles disks, pools, snapshots, shares, sync, jobs and live health. Anything that needs root goes through nos-agent, a helper that accepts only allow-listed actions over a Unix socket. Underneath are Btrfs storage pools, Docker apps and an nftables firewall that only lets the local network in by default. systemd timers run snapshots and pruning, and health metrics refresh every second.BrowserReact dashboardSync appsDesktop and mobileCaddyServes the UI, the API and WebDAV for syncnosdGo API for pools, snapshots, shares and syncnos-agentRoot helper, allow-listed actions onlyBtrfsDocker appsFirewallJobs and monitoringSnapshot timers, metrics every second
Inside NithronOS. The highlighted line is one snapshot, from the dashboard down to the disks.
Built with
  • Go
  • React
  • TypeScript
  • Caddy
  • Debian
Licence
NithronOS Community License. You can read, modify and contribute to the code; selling, hosting or redistributing binaries needs a commercial agreement.
Activity

NithronAI (nAI)

Open source

A local-first document Q&A system. It ingests PDFs, Markdown, text, HTML and code, searches them, and answers questions with citations. Private by default, and it runs well on a homelab.

  • Full-text search with BM25, plus optional semantic search through embeddings in Qdrant.
  • Optional answers from a language model through Ollama, OpenAI or Anthropic.
  • Multi-turn chat with conversation history, document management, JWT authentication and rate limiting.
  • Runs as a Docker Compose stack with Qdrant and Ollama.
How NithronAI answers a question. Documents (PDF, Markdown, text, HTML and code) are ingested and indexed for full-text search with BM25, with optional semantic search through embeddings in Qdrant. A question, asked in a chat that keeps its history, is searched against the index, and the answer comes back with citations, optionally written by a language model through Ollama, OpenAI or Anthropic. It all runs as one Docker Compose stack, with JWT authentication and rate limiting.DocumentsPDF, Markdown, HTML, codeIngestEvery formatIndexBM25 full textQdrantEmbeddings, optionalQuestionChat with historySearchFull text and semanticAnswerWith citationsLanguage modelOptional, through Ollama, OpenAI or AnthropicOne Docker Compose stackJWT auth and rate limitingHow NithronAI answers a question. Documents (PDF, Markdown, text, HTML and code) are ingested and indexed for full-text search with BM25, with optional semantic search through embeddings in Qdrant. A question, asked in a chat that keeps its history, is searched against the index, and the answer comes back with citations, optionally written by a language model through Ollama, OpenAI or Anthropic. It all runs as one Docker Compose stack, with JWT authentication and rate limiting.DocumentsPDF, Markdown, HTML, codeIngestEvery formatIndexBM25 full textQdrantOptionalSearchFull text and semanticQuestionChatAnswerWith citationsLanguage modelOptionalOne Docker Compose stackJWT auth and rate limiting
How NithronAI works. The highlighted line follows a document from upload to an answer that cites it.
Built with
  • Python
  • FastAPI
  • Qdrant
  • Ollama
  • Docker
Licence
GNU AGPL-3.0. Commercial add-ons and support licences will be offered separately.
Activity