opencode-plugboard

Appearance

paperless-gpt ​

paperless-gpt is a companion to paperless-ngx. An LLM suggests titles, tags, correspondents and document types, and a vision model can run OCR.

Upstream: source.

Plugboard options ​

Give paperless-gpt its own key. vision handles OCR and must accept images. Docker needs a reachable address (Serving other machines).

jsonc
{
  "options": {
    "listen": { "host": "0.0.0.0", "port": 4010 },
    "keys": { "paperless-gpt": "{env:PLUGBOARD_KEY_PAPERLESS_GPT}" },
    "models": ["opencode-go/glm-5.3-flash", "anthropic/claude-sonnet-4-6"],
    "aliases": {
      "text": "opencode-go/glm-5.3-flash",
      "vision": "anthropic/claude-sonnet-4-6"
    },
    "apps": { "paperless-gpt": { "allowedModels": ["text", "vision"] } }
  }
}

Connect ​

ini
PAPERLESS_BASE_URL=http://paperless-ngx:8000
PAPERLESS_API_TOKEN=<your paperless-ngx API token>

LLM_PROVIDER=openai
LLM_MODEL=text
LLM_LANGUAGE=English

# The plugboard app key, not an OpenAI key.
OPENAI_API_KEY=<the paperless-gpt app key>
OPENAI_BASE_URL=http://host.docker.internal:4010/v1

# LLM OCR, if you use it. The model must accept images.
VISION_LLM_PROVIDER=openai
VISION_LLM_MODEL=vision

LOG_LEVEL=info

OPENAI_API_KEY takes the plugboard key. OPENAI_BASE_URL ends in /v1.

The .env and a compose file are in examples/self-hosted/paperless-gpt/.

What works ​

  • Suggestions for title, tags, correspondent and document type, through LLM_MODEL.
  • LLM OCR, through VISION_LLM_PROVIDER and VISION_LLM_MODEL. Set the OCR provider to llm to use it.

Limits ​

  • paperless-gpt has no login. Anyone who can reach its port can use the UI and /api/*. Publish the port on loopback or a trusted network.
  • Keep the paperless-ngx API token in an env file, not in the compose file.