spnc.ai builds a continual learning layer for AI assistants that persists knowledge across sessions with Claude, ChatGPT, Cursor, Codex, and Claude Codeaine. The system creates typed, signed knowledge objects with provenance and supersession edges, enabling AI to compound understanding rather than resetting between conversations. Its architecture resolves information staleness at write time, with benchmark results showing 100% accuracy on current-state probes versus 60% for naive retrieval.
Funding
Funding not disclosed
Founders
Product
Problem
AI assistants operate with no persistent memory across sessions, forcing users to re-explain context and losing accumulated knowledge between conversations. Current approaches to agent memory fall into two categories—in-context retrieval or in-weights training—both of which forget structure, provenance, and which facts superseded others, preventing knowledge from compounding over time.
Solution
Sapience provides continual learning at the system level, creating a persistent knowledge layer that lives across all major AI assistants including Claude, ChatGPT, Cursor, Codex, and Claude Code. The system captures typed, signed knowledge objects with supersession edges that record what replaced what at write time, ensuring the AI always accesses current information rather than stale records. Every fact carries provenance and is owned by its creator, with knowledge consolidating and synthesizing while users are away. When users choose, knowledge can connect across minds, enabling collective wisdom while maintaining attribution and credit. The system arrives in minutes via invite code and grounds its answers in the user's personal notes, lighting the specific sources it used in each response.
Target Audience
Primary users are professionals and teams who work extensively across multiple AI assistants and need persistent, owned knowledge that compounds across sessions—including researchers, engineers, analysts, and knowledge workers managing complex, evolving information domains.
Features
- Continual learning layer that persists across Claude, ChatGPT, Cursor, Codex, and Claude Code sessions
- Typed, signed knowledge objects with provenance tracking on every fact
- Supersession edges recorded at write time, preventing stale-value retrieval
- Automatic consolidation and synthesis of knowledge while the user is away
- Optional cross-mind connections with attribution, enabling shared knowledge with credit to creators
- Answer grounding that references and displays the specific notes and sources used
- Benchmark-validated performance: 100% accuracy on current-state probes versus 60% for naive similarity retrieval (p<0.0001)