Konmane
Boxes Series Selling Getting started Settings 見本 Admin
This is a hands-on sample. Changes are not saved.

AI settings (writing photo descriptions)

Now: Claude (Anthropic) / Model claude-opus-5 / Key not set

⚠ The photo and that box's records are sent. Each time a description is generated, the AI provider you contracted with yourself (Anthropic or Google) receives that one photo, the box's label, the descriptions on the box's other photos (up to 5, 150 characters each), and the text you wrote under "Field". Whatever appears in the photo — delivery notes, invoices, people — is sent with it. If you do not use the AI, none of this ever leaves the device. Either way, it never reaches the supplier.

Settings

🔒 Not available in the sample (no device here). Works on a real unit.

Provider
API key (leave empty to keep the current one)
In-house AI address (only when In-house AI is selected)
Part-number search address (SearXNG; only with In-house AI; leave empty to not search)
Search hint (added after the part number when the number alone finds unrelated items; leave empty to skip)
Model

Saving takes effect from the next run. The key is stored only on this device and never sent anywhere. It is not included in backups, so after restoring a backup on another device, enter the key here again.

Getting a key (5 minutes)

For Claude:

  1. Create an account at platform.claude.com (email only; separate from the Claude app subscription)
  2. Billing in the left menu → add a payment method and put in a small prepaid balance (about $5). It is drawn down as you use it and you are never charged beyond it
  3. API Keys in the left menu → Create Key → any name will do
  4. Copy the sk-ant-… right away (it is never shown again once you close the screen)
  5. Paste it into “API key” here, press “Fetch the model list”, choose a model, then “Check connection and save”

For Gemini: aistudio.google.com (Google AI Studio) → Get API key. Always issue it from a project with billing enabled. With a free-tier key, the photos you send are used to improve Google’s products and may be read by people (their terms say so).

In-house AI: run a model that can read photos on a PC or server inside your company, in the OpenAI-compatible form (Ollama, LM Studio, llama.cpp, vLLM, etc.). Choose "In-house AI" as the provider, enter http://server-address:port/v1 as the in-house AI address, then "Fetch the model list" → choose a model → "Check connection and save". Enter a key only if the server requires one. See the manual for details.

Rough cost per photo: about 4 JPY with Claude Opus 5 (most accurate, down to small print), about 1 JPY with Claude Sonnet 5, 0.2–0.4 JPY with Gemini Flash (as of September 2026). The Activity log shows the real cost per photo.

If a model is discontinued: press “Fetch the model list” again and pick another one.