Architecture & Philosophy
Folocard exists to give attention back to humans during meaningful first-touch encounters. It turns raw field encounters into structured relationships without enterprise clutter.
The First-Touch Mini CRM Frame
Most CRM software focuses on pipeline stages, quarterly reporting, and administrative record-keeping. Folocard focuses exclusively on the critical first touch: the handshake, the business card, the conversation, the quick exchange of contacts at a conference, and the critical first follow-up.
The product does not replace an enterprise sales system. Instead, it provides a local, proactive assistant that captures context when memories are fresh, structures person-level knowledge, and prepares verified next actions.
Compute Allocation Principle
Every technical decision in Folocard follows a strict rule: deterministic compute before probabilistic compute before human attention.
1. Deterministic Layer (Fast & Free)
Rules, heuristics, regular expressions, and Google ML Kit on-device optical character recognition run instantly with low battery consumption. This layer handles structured fields: phone numbers, email addresses, names, job titles, and website domains.
2. Probabilistic Layer (Adaptive & Local)
On-device vision-language models (VLMs) and large language models (LLMs) operate only where deterministic rules reach their limit: interpreting noisy card layouts, synthesizing meeting notes, researching public web context, and drafting conversational follow-ups.
3. Human Attention (Intent & Taste)
Human attention is the most expensive resource. The user provides intent, context, and approval. If the user must manually type contact details, the automated layers failed. Folocard aims to minimize keyboard typing in favor of quick tap-to-verify interactions.
Additive Architecture on Off-Grid
Folocard builds additively on the Off-Grid mobile substrate. Off-Grid provides core local primitives: model loading, tool calling, local storage, chat mechanics, and document management.
Folocard wraps these primitives in a dedicated first-touch workflow:
- CaptureSession: Coordinates camera bursts, voice transcripts, and manual notes.
- FirstTouchThread: Holds the durable relationship record for one person.
- Sourced Dossier: Compiles iterative findings, citations, and unknown elements.
- NextAction: Schedules the immediate follow-up task with a prepared draft.
System Boundaries and Disclosures
Notice on network behavior: Folocard stores dossiers and performs AI inference locally. When public-web research is enabled, search terms and requested URLs travel across public internet routes to reach target servers. Hugging Face serves model downloads, and Google ML Kit transmits SDK diagnostics. Folocard does not transmit your personal data to remote model brokers.