Which AI models and providers does amaise use?
amaise uses the following AI services:
Azure OpenAI (Microsoft) — for LLM inference (document classification, metadata extraction, summaries, question answering). Processing is regional:
EU: Azure EU Data Zone — processing restricted to EU regions (France Central, Germany West Central, Italy North, Poland Central, Spain Central, Sweden Central, West Europe)
US: Azure US Data Zone — processing restricted to US regions (Central US, East US, East US 2, North Central US, South Central US, West US, West US 3)
CH: Switzerland North (models strictly limited to Switzerland North)
Under Microsoft's terms, prompt and completion content may be retained for up to 30 days for abuse monitoring, accessible only to authorized Microsoft reviewers, and is never used to train or improve models. amaise itself logs only token counts — no prompt or completion content.
Google Document AI — for OCR processing (optical character recognition) for EU and US customers. Processing runs in the respective region: EU documents in the EU, US documents in the US.
Microsoft Azure AI Document Intelligence — for OCR processing for Swiss customers. Hosted in Switzerland (Switzerland North) as standard. Can also be enabled for EU customers (West Europe) and US customers (East US).
Amazon Textract (AWS) — for EU and US customers, reads selected document pages (forms, tables, handwriting) in the same AWS region, to improve the extraction of summaries, diagnoses, and document dates. Not used for Swiss customers.
Optional legal research sources — when activated for a workspace, research queries are sent to external sources such as Perplexity (web research, USA) and Swiss case-law databases. They receive search queries only — no documents. The names known from the case are removed automatically, and queries that still contain identifiers such as a date of birth or social security number are blocked. LegalPilot, when activated, sends the user's question to Perplexity as entered.
The processing location per environment is specified in the Data Processing Agreement (DPA) and sub-processor list.
Document images are stored only temporarily for OCR (at the latest two days), deleted after processing, and not used for training.
The specific model versions are defined and fixed per environment in the infrastructure configuration. The platform is model-agnostic and configuration-driven — no model version is hard-coded in the application code.