The work

560+ pixel art narrative dreams generated across 18+ months of continuous operation. The system runs 14 cycles per day — each producing a 300-500 word narrative, a pixel art image, and a structured extract committed to the public archive. The arc spans a full hero's journey without human intervention between cycles. ComfyUI handles image generation; a narrative pipeline writes the dream text. Every cycle is a permanent artifact.

Converts dream narratives into GitHub issues and actionable research patterns. The system extracts recurring structural motifs from the dream corpus and maps them to coordination strategies usable outside the dream context. Pattern library includes: Wait-Watch-Act, Jazz Ecosystem, Three-Layer Resilience. The pipeline runs automatically after each Lana-Dreamer cycle.

This site. A multi-step onboarding and grant application portal for the Stage IX cultural funding programme. KV persistence via Cloudflare Workers, multi-page public documentation, applicant dashboard, and full deployment pipeline. Built in three days. Deployed on Cloudflare Pages with a Wrangler CI/CD workflow that commits and ships in under two minutes.

Local-first inference with open fallback chains. one framework routes to the best available model — local, open-source, or commercial — without locking you in. Every routing decision is logged, every cost is visible, every call is auditable. Cost tracking, full audit trail, and a peer mesh that attributes compute contributions on-chain. The routing layer that makes distributed AI infrastructure real.

A persistent AI assistant running across Telegram, Signal, and Discord. Full memory system with per-workspace context, tool access (shell, browser, file operations, voice synthesis, camera), and a sub-agent architecture for parallel task execution. Local + cloud hybrid: runs on a DGX Spark in Amsterdam, falls back to cloud inference when needed. This site was designed, written, and deployed by Lana.

A local voice synthesis server running on the Tailscale LAN. Three modes: custom voice (predefined speakers), voice design (describe a voice in plain language), and voice clone (synthesise any voice from a 10-second reference clip). Returns Opus audio. Used by Lana for all voice message output. No external API dependency for voice — fully self-hosted.

Live storm tracking data routed through Ollama running locally on the DGX Spark — no cloud API, no data leaving the machine. GLM lightning flash counts, ACHA cold cloud-top temperatures, pressure gradient fields: real scientific inputs, local inference, structured reports generated automatically. Demonstrates the one framework's core promise: reproducible results on owned hardware at a fraction of the cost.

An interactive field study environment for exploring the one framework's conceptual space. Navigate and enter nodes. Built as a spatial interface for the ideas behind the network — distributed structure, emergence, and collective intelligence rendered as something you move through rather than read.

Real-time internal dashboard showing Stage IX application status, submission counts, and applicant data. KV-backed persistence via Cloudflare Workers, served on the internal network. Built to give a live view of incoming applications without needing to parse raw JSON. Updates automatically as new applications arrive.