5.7.3. AI configurations per job (OpenAI, Gemini, Ollama)
An AI configuration tells ChronoScan which model to call, with which prompt, for which documents and where to put the answer. A job can have several configurations (one per provider, per document type or per task) and chooses which one each document type uses.
Providers
Provider
Models
Notes
OpenAI
GPT-4.1 and GPT-5 families, including mini and nano
Reference provider. Supports deferred batch processing at a discount.
Azure OpenAI
The same models deployed in your Azure tenant
Data stays in your Azure subscription and region.
Google Gemini
Gemini Flash, Flash-Lite and Pro
Good price and quality. Also supports deferred batches.
DeepSeek
DeepSeek API
Low cost alternative.
Ollama (local)
Llama and other open models on your own server
Fully on-premise, no usage cost, needs a suitable GPU.
Local OpenAI-compatible server
LM Studio or any server exposing the OpenAI API on the local network
Option "Local server" of the OpenAI module. Local addresses only, no key, no credits. Enabled on request for ADVANCED and Enterprise licenses.
Where they are configured
Open the job configuration (Administration tab, Configure job) and the AI section: one page per provider with the list of configurations. Each configuration has an alias (it names the response files and the AI Assistants refer to it), the model (and an optional secondary model from a number of tokens upward), the system prompt, the JSON template whose nodes are named after the fields of the job (an array node is a capture grid), the field mapping, which pages are sent and whether the page images are sent too, the split options (the model marks where each document starts), the conditions to run (masterkey, field condition) and the call parameters (temperature, timeout, retries).
Import AI configuration copies a configuration from another job or converts it between providers: if OpenAI and Gemini share the same alias and template, switching engine only means changing the stage in the processing chain.
How it runs
The configurations run in the processing chain of the job as the OPENAIMODEL or GEMINIAIMODEL stages (Process and Tasks settings), typically after OCR and before or after the type detection. The answer is saved in the document (oai.data.<alias>.json or gemini.data.<alias>.json) and written to the fields and grids according to the mapping; a reprocess reuses the saved answer unless the response is deleted (ChronoDocument.AI_DeleteResponse or the reset option of the job).
Fields filled by AI get the OPENAI or GEMINIAI status. The AI does not overwrite values read from zones or typed by the user; it can be disabled per document type and per masterkey. The per-field AI Assistant can match the values returned against your master lists.
Keys, credits and models
With your own API key (entered in the program options) you pay the provider directly. Without a key the requests go through the ChronoScan HUB and are paid with ChronoScan credits; see AI cost and credits. When a provider retires a model ChronoScan warns about the deprecated model and proposes the replacement without rewriting the prompts.
