OhnePapier v0.1 — Debian 13 GPU Stack

What This Is

OhnePapier ("Without Paper") is a fully local, privacy-first document management stack running on Docker Compose. It ingests scanned documents via Paperless-ngx (OCR + storage), routes them through an AI layer (paperless-ai → local or cloud LLM) for automatic tagging and classification, and stores everything in PostgreSQL — zero cloud dependencies unless explicitly configured. Part of the Crumbforest ecosystem.

Core Value

Every document gets intelligently tagged and searchable, fully offline, on your own hardware — with cloud LLM as an opt-in escape hatch tracked by token cost.

Requirements

Validated

  • ✓ Paperless-ngx OCR + document storage running — existing
  • ✓ PostgreSQL 18 + Redis 8 backend running — existing
  • ✓ Ollama local LLM connected (llama3.2 + llama2) — existing
  • ✓ paperless-ai auto-tagging pipeline active (~467 documents processed) — existing

Active

  • [ ] Fresh Debian 13 install guide / setup script with GPU support
  • [ ] Ollama containerized inside Docker (not host-side) with GPU passthrough (nvidia-container-toolkit)
  • [ ] Gitea service added to the stack (self-hosted Git, co-located)
  • [ ] Anthropic API as configurable AI provider alternative to Ollama
  • [ ] OpenRouter as configurable AI provider alternative to Ollama
  • [ ] Token cost tracking for cloud LLM usage (#token)

Out of Scope

  • OZMBridge / FPGA integration — future milestone, tagged #ozm
  • Nullfeld-Integration (FPGA × SoC star maps) — #ozm dependent
  • Vector-Bridge to Eule/Qdrant cross-system RAG — #ozm dependent
  • Custom Fields — deferred until v0.1 stable

Context

  • Current host: macOS, Ollama running on host via ollama serve, accessed by containers via host.docker.internal:11434
  • Target: Debian 13 server with NVIDIA GPU; Ollama moves into a container with nvidia-container-toolkit GPU passthrough
  • paperless-ai currently supports Ollama and OpenAI-compatible APIs — Anthropic and OpenRouter require either native support or a proxy adapter
  • Token tracking (#token): cloud LLM calls must log model, tokens in/out, and estimated cost per document processed
  • OZMBridge lives at /Users/bmt/Documents/OZMBridge — FPGA/OZM layer, integration is a separate future milestone

Constraints

  • Philosophy: Local-first, no surveillance, no cloud by default — CKL.md / HHL.md ethical licenses apply
  • GPU: NVIDIA assumed (nvidia-container-toolkit path); AMD/ROCm is not in scope for v0.1
  • Gitea: Must be accessible within the Docker network and optionally exposed externally
  • AI providers: Switching between Ollama / Anthropic / OpenRouter must be a config-level change, not a code change

Key Decisions

Decision Rationale Outcome
Containerize Ollama GPU passthrough in Docker is now stable on Linux; removes host dependency — Pending
Add Gitea to stack Self-hosted Git co-located with document store; Crumbforest-native — Pending
Cloud LLM as opt-in Privacy-first default; cloud only when explicitly configured + token-budgeted — Pending

Evolution

This document evolves at phase transitions and milestone boundaries.

After each phase transition (via /gsd-transition):
1. Requirements invalidated? → Move to Out of Scope with reason
2. Requirements validated? → Move to Validated with phase reference
3. New requirements emerged? → Add to Active
4. Decisions to log? → Add to Key Decisions
5. "What This Is" still accurate? → Update if drifted

After each milestone (via /gsd-complete-milestone):
1. Full review of all sections
2. Core Value check — still the right priority?
3. Audit Out of Scope — reasons still valid?
4. Update Context with current state

Last updated: 2026-04-17 after initialization