Choose where intelligence runs

Structor · AI Model

Understand the two layers

Structor AI

Choose Structor AI (Subscription) when you want the managed provider attached to your Structor plan.

1. Sign in under Settings and confirm an active plan.
2. Select Structor AI (Subscription).
3. Press Test.
4. Confirm the sample files receive structured category results.

No endpoint, model name, or API key is required in this mode.

Bring Your Own AI

BYO has two modes:

Settings save as you edit. Changing provider, endpoint, API key, or model cancels an in-progress provider test and triggers a fresh capability check.

Choose a provider quickly

Ollama

Ollama is the shortest local setup on macOS, Windows, or Linux.

1. 01

   Install and start Ollama

   Install Ollama using its official quickstart. Keep the desktop service running.
2. 02

   Download a model

   Choose a model with tool/function calling support and pull it locally:

   ollama pull <model>
   

3. 03

   Connect Structor

   In AI Model, choose Bring Your Own AI → Local → Ollama. Keep the default endpoint:

   http://localhost:11434/v1
   

4. 04

   Select and test

   Refresh available models, select the downloaded model, and press Run test. The API key can remain empty; Ollama's local OpenAI-compatible endpoint ignores a dummy token.

See Ollama's OpenAI compatibility reference for supported endpoints and model behavior.

LM Studio

LM Studio provides a desktop model browser and OpenAI-compatible local server.

1. 01

   Download a capable model

   Install LM Studio, open Discover, and download an instruct/chat model whose card lists tool or function calling. CLI users can run:

   lms get <model>
   

2. 02

   Load the model

   Load it from Chat or Developer, or run lms load. Keep enough memory available for the selected quantization.
3. 03

   Start the API server

   Open Developer and switch on the server, or run:

   lms server start --port 1234
   

4. 04

   Connect Structor

   Choose Local → LM Studio and use:

   http://localhost:1234/v1
   

   If LM Studio API authentication is enabled, enter its token in Structor's API key field.

5. 05

   Refresh and test

   Refresh models, choose the loaded model, and run the provider test. If the list is empty, confirm http://localhost:1234/v1/models is reachable.

Reference: LM Studio local server, model downloads, and tool use.

oMLX

oMLX is an OpenAI-compatible inference server for Apple Silicon Macs. Current upstream requirements include macOS 15 or newer.

1. 01

   Install oMLX

   Install the macOS app from oMLX releases, or use Homebrew:

   brew tap jundot/omlx https://github.com/jundot/omlx
   brew install omlx
   

2. 02

   Choose storage and download a model

   The app welcome flow asks for a model directory and offers a model downloader. Choose an MLX-format instruct model with tool support. The admin dashboard can search Hugging Face and download models directly.
3. 03

   Start the server

   Start it from the oMLX menu-bar app, use its Homebrew service, or run:

   omlx serve --model-dir ~/models
   

4. 04

   Configure Structor

   Choose Local → Custom local endpoint and enter:

   http://localhost:8000/v1
   

   If oMLX started with --api-key, put that same value in Structor's API key field. Otherwise leave it empty for localhost.

5. 05

   Select the API model

   Refresh models and choose the name or alias returned by oMLX's /v1/models endpoint. Run the provider test and confirm structured output support.

Reference: oMLX installation, models, API, and authentication.

Other custom providers

Choose Bring Your Own AI → Custom, then select the contract your endpoint exposes:

Enter the complete base URL, model identifier, and bearer token when required. Localhost, 127.0.0.1, and [::1] OpenAI-compatible endpoints can operate without an API key.

Provider test

The test sends sample filenames through the selected model and expects category values. Status can progress through connection, capability probing, categorization, and completion.

• Ready: structured categories were returned.
• Warning: provider responded but Structor used a fallback or found limited capabilities.
• Error: endpoint, authentication, model name, or structured output failed.

When debugging, verify in this order: server running, endpoint correct, model visible, token accepted, model loaded, provider test successful.

Local search models

The bottom of AI Model manages:

Press Download once. Status moves through pending, downloading, ready, or failed. Progress includes downloaded and total bytes when available.