Logbook Import

The Logbook Import feature uploads scanned aircraft logbook pages as documents and, when an AI provider is configured, can also transcribe them into structured digital entries with dates, descriptions, hours, and signoff information.

Accessing the Import Page

Click Import Logbooks in the navigation bar to access this page. This feature is available to aircraft owners.

Note

AI transcription requires at least one AI provider to be configured by the system administrator (Anthropic API key for Claude, or an Ollama instance URL for self-hosted models). If no provider is configured, the page operates in upload-only mode: images are saved as documents but no logbook entries are extracted.

Uploading Source Files

You can provide logbook pages in two ways:

Multiple image files

Select individual image files (JPEG, PNG, TIFF, WebP, BMP, or GIF). Files are processed in alphabetical order by filename, so name them so that alphabetical order matches page order (e.g., page-001.jpg, page-002.jpg).

Archive file

Upload a single zip, tar, tar.gz, tar.bz2, or tar.xz archive containing image files. Images inside are also sorted alphabetically by filename.

Setting the Destination

  • Aircraft – Select which aircraft this logbook belongs to.

  • Collection name – Groups the uploaded images into a document collection (e.g., “Airframe Log #4”). This is auto-filled from the filename but can be edited.

  • Document name – The name for the document record. Defaults to the collection name.

Import Options

Upload only (no transcription)

Check this to save the images as documents without running AI transcription. Useful if you just want to archive scanned pages without extracting entries. If no AI provider is configured, this mode is used automatically and the option is not shown.

Document type

Categorize the uploaded document: Log, Alteration, Report, Invoice, Aircraft Record, or Other.

When transcription is enabled, additional options appear:

AI model

Select which AI model to use for transcription. Available models depend on what providers your administrator has configured. Options may include Claude (cloud) or locally-hosted models via Ollama.

Log type override

By default, the AI auto-detects the log type (Airframe, Engine, Propeller, or Other) for each entry. Use this to force all entries to a specific type.

Batch size

The number of images sent per API call (1–20). Larger batches provide more context for the AI but use more tokens. The default is usually appropriate.

Running the Import

  1. Select your files and configure the destination and options.

  2. Click Import & transcribe (or Upload images if upload-only mode is selected).

The import runs in real time with a progress display:

  • A progress bar showing overall completion percentage.

  • A progress log showing detailed events as they happen:

    • Image upload confirmations (green).

    • Extracted entries (blue).

    • Warnings (yellow) – e.g., truncated output from the AI.

    • Errors (red) – e.g., failed API calls.

When the import completes, a summary shows:

  • Entries created – The number of logbook entries extracted and saved.

  • Images uploaded – The number of images saved as documents.

  • Warnings and errors counts, if any.

Click New import to start another import.

Tips for Best Results

  • Image quality matters. Clear, high-resolution scans produce better results. Avoid blurry or low-contrast images.

  • Name files in order. The AI processes pages in batches, with overlap between batches for context. Correct page ordering ensures entries that span pages are handled properly.

  • Review imported entries. AI transcription is not perfect, especially with handwritten text. Check the imported entries in the Logbook tab and correct any errors.

  • Use log type override when the entire logbook is one type (e.g., an engine logbook). This improves accuracy by removing ambiguity.

Local & Custom AI Providers (Desktop App)

The desktop build can run logbook transcription against a local Ollama model or any OpenAI-compatible endpoint (vLLM, LiteLLM proxy, OpenRouter, etc.) in addition to or instead of the Anthropic API. You can configure any combination of the three; if more than one is configured, you pick which becomes the default in the import-page model selector.

On Apple Silicon Macs, Ollama uses the GPU via Metal – but the same setup works on Windows and Linux desktop builds (Ollama uses CUDA or ROCm where available, otherwise CPU).

This is a power-user feature: you choose and size models yourself, and quality varies between providers and models.

Local Ollama

  1. Install Ollama from https://ollama.com (or brew install ollama on macOS) and start it – the menu-bar app, or ollama serve in a terminal.

  2. Pull a vision-capable model. Text-only models will not work here:

    $ ollama pull llama3.2-vision
# or, for a smaller / faster option:
$ ollama pull qwen2.5vl:7b

  3. First-run setup screen (when you launch the desktop app for the first time): in the AI features section, fill in Ollama model with the tag you pulled (e.g. llama3.2-vision). Leave the Ollama base URL on its default unless you run Ollama on a non-default port. The Anthropic API key field can be blank if you only want local AI.

  4. Restart the app. The Ollama model will appear in the AI model dropdown on the import page.

Choosing a model

  • llama3.2-vision (11B parameters) is a reasonable starting point on Apple Silicon Macs or recent NVIDIA GPUs with 16 GB+ of system or VRAM.

  • qwen2.5vl:7b is faster and uses less memory but with somewhat lower extraction quality on dense logbook pages.

  • A 7B-class model needs roughly 6 GB of free RAM; 11B-class models want 12 GB+.

OpenAI-compatible endpoint (vLLM, OpenRouter, LiteLLM)

For routing through a hosted aggregator (OpenRouter), a self-hosted inference server (vLLM, LiteLLM proxy), or any other server that speaks the OpenAI Chat Completions API, fill in the OpenAI-compatible endpoint subsection of the setup form:

  • Model ID – whatever string the endpoint expects, e.g. gpt-4o-mini, anthropic/claude-sonnet-4-6 (OpenRouter-style), or your vLLM-served model name.

  • Endpoint base URL – the OpenAI-compatible base, e.g. https://openrouter.ai/api/v1, http://localhost:8000/v1 (vLLM), or your LiteLLM proxy URL.

  • API key – stored in your OS credential store (Keychain on macOS, Credential Manager on Windows, Secret Service on Linux). Leave blank for endpoints with no authentication, such as a local vLLM server on your own machine.

The model needs to be vision-capable for logbook transcription to work. Text-only models will fail at the first batch.

Choosing a default

When more than one provider is configured, the Default model radio group at the bottom of the AI section decides which provider’s model is preselected on the logbook import page. You can still pick any of the configured models per import; the radio just sets the initial value.

To flip the default later, edit the desktop config file (path varies by OS, see below) and change the default_provider line in the [ai] section to one of anthropic, ollama, or litellm. Then restart the app.

Editing config.ini after first-run setup

Already past first-run setup? You can edit the desktop config file directly. Its location depends on your OS:

  • macOS: ~/Library/Application Support/SimpleAircraftManager/config.ini

  • Windows: %LOCALAPPDATA%\SimpleAircraftManager\config.ini

  • Linux: ~/.local/share/SimpleAircraftManager/config.ini

Add an [ai] section like this:

[ai]
default_provider = ollama
ollama_model = llama3.2-vision
ollama_base_url = http://localhost:11434
litellm_model = gpt-4o-mini
litellm_base_url = https://openrouter.ai/api/v1

API keys are not stored in this file – they live in the OS credential store. To set or change them outside the first-run flow, use the Keychain Access app (macOS), Credential Manager (Windows), or secret-tool (Linux), under service SimpleAircraftManager with usernames anthropic_api_key or litellm_api_key.

Restart the app for changes to config.ini to take effect.

Tradeoffs

  • Local models are slower per page than the Anthropic API on most hardware, and JSON-schema compliance is weaker – expect occasional truncation or malformed entries that you have to clean up by hand.

  • OpenAI-compatible endpoints vary widely. Hosted aggregators (OpenRouter, etc.) typically work well. Self-hosted vLLM with a capable vision model is fast but requires GPU resources.

  • Image quality matters even more than with the Anthropic API. Crisp scans make a noticeable difference.

  • With Ollama or a self-hosted endpoint, all processing stays on your machine; nothing leaves it.