What AI does
- Extracts text from PDFs that lack a usable text layer through an OCR fallback.
- Creates vector embeddings used to retrieve relevant document passages.
- Generates document summaries and answers to questions.
- Reformulates follow-up questions using limited conversation history.
What may be sent to Google
Depending on the feature, SynapseDoc may send PDF content, extracted text, sampled chunks, retrieved passages, your question, and recent conversation context to the Gemini API. Account passwords are not intentionally sent. File content may still contain names, contact details, confidential information, or other personal data supplied by you.
Model limitations
- Outputs may be wrong, incomplete, fabricated, biased, or inconsistent.
- A citation shows which passage was retrieved; it does not prove the generated interpretation is correct.
- Scanned, image-heavy, damaged, protected, or unusually formatted PDFs may process poorly.
- Model versions, quotas, safety filters, and availability can change.
Your responsibility
Review original pages before relying on an answer. Do not use SynapseDoc as the sole source for high-impact or professional decisions. Do not upload content that you are prohibited from sharing with an external AI provider.
No training claim
SynapseDoc itself does not train a model on your documents. Provider handling of API data depends on the applicable Google account, billing status, region, product terms, and settings. The operator must verify the production Gemini plan and its current terms before claiming that provider data is never retained or used for improvement.
Provider selection and data processing
Text-generation requests may be routed between Google Gemini and an enabled fallback provider such as Groq for reliability. Data can include a question, bounded conversation history, and retrieved passages. OCR can require transmitting PDF content to Gemini. SynapseDoc does not guarantee a specific provider. Account-specific retention, training, location, and legal terms must be verified before launch.
AI processors, gateways, and reranking
Depending on server configuration and provider health, ordinary text generation may use Google Gemini, Groq, Cerebras Cloud, SambaNova Cloud, Mistral AI, OpenRouter, Hugging Face Inference Providers, or Cloudflare Workers AI. OpenRouter and Hugging Face are gateways and may also transmit the request to an upstream inference provider. Not every provider receives every request, and SynapseDoc does not guarantee a specific provider.
Ordinary chat may send the question, bounded conversation history, and bounded retrieved passages rather than the entire document. If Cohere reranking is enabled, Cohere may receive the query and a bounded candidate set of retrieved passages. PDF OCR may send PDF content to Google Gemini, which remains the only active AI OCR provider.
Retention, training, security, region, data-residency, and compliance terms depend on each operator account and must be verified before launch. Evaluation or free-tier credentials are not assumed to be production-suitable.