Introducing the Mind Layer
Today we're publicly introducing the Mind Layer — model-agnostic infrastructure that transforms any foundation model into a believable, persistent AI employee.
Most AI employees suffer from the same fundamental problems: they forget everything between sessions, their personality is just a static system prompt, they only respond when spoken to, and they're locked to whichever model they were built on. The Mind Layer solves all four.
The engine wraps any LLM with 9+ parallel context layers — core identity, personality dynamics, evolution and growth, relationship tracking, current state and mood, hierarchical memory, proactive commitments, constellation patterns, and custom developer-defined states. These layers span different temporal intervals — real-time, near-real-time, batched, windowed, and inference-time — yet all collapse into a single unified context at the point of inference, fully automated. Your LLM handles language. We handle everything that makes a character feel alive.
The architecture separates character state from model execution entirely. Memory, personality, affect, and relational state persist independently of any specific LLM. Think of it this way: the LLM is the brain, but the Mind Layer is the mind. You can upgrade the brain without losing the mind.
This solves a real problem in a fast-moving industry. Companies that fine-tune or train character behavior into specific models get locked in. When a better model appears — and better models appear constantly — they face months of retraining, re-evaluation, and migration. With the Mind Layer, you swap in the new model and your characters instantly get smarter. No retraining. No fine-tuning. No data migration.
But model-agnosticism isn't just about flexibility — it flips the economics of AI employees entirely. The Mind Layer does the cognitive heavy lifting outside the model: conversations are processed through multiple LLM passes — extraction, consolidation, summarization, fact deduplication, personality evolution, affect construction. BM25 inverse indexes, entity indexes, temporal indexes, and type indexes enable sub-200ms retrieval across thousands of memories. Nine-plus context layers — identity, personality dynamics, evolution, relationships, current state, memory, proactive commitments, constellation patterns, custom states — are computed in parallel across different temporal intervals and collapsed into a single unified context window at inference. By the time context reaches the generation model, it's been deeply processed and refined. The result: a lightweight model receiving this orchestrated context produces responses comparable to a frontier model working from raw conversation history alone — at 1/20th the cost. ~$0.10 per conversation instead of ~$2.00. It's the difference between a solo musician improvising and a musician backed by a full production team — arrangement, mixing, sound engineering. The musician doesn't need to be a virtuoso when the production quality is excellent.
It works with OpenAI, Anthropic, Google, Mistral, and any OpenAI-compatible API including self-hosted models via vLLM, Ollama, or TGI. Swap providers without rewriting your character logic — same personality, same memories, better brain.
Key capabilities include sub-200ms memory retrieval at p95, 20x model cost reduction through externalized context, 50+ personality dimensions modeled through Big5 traits and constructed emotion grounded in core affect, and zero cold starts — characters never forget who you are.
The Mind Layer is built for two primary use cases. For gaming, it powers NPCs with persistent memory and evolving relationships — they remember player choices across sessions, develop opinions, hold grudges, and form alliances. For AI companions, it enables relationships that deepen over months with proactive outreach, personality evolution, and relationship modeling.
Deploy anywhere: mobile apps via REST API with SSE streaming, game engines like Unity and Unreal, messaging platforms including Telegram, WhatsApp, and Discord, or direct REST API access with OpenAI-compatible streaming.
The engine is already battle-tested in production powering Pocket Souls, our flagship consumer product where AI companions remember you, check in on you, and grow alongside your journey.
Explore the platform at platform.sonz.ai or reach out at hello@sonz.ai to discuss how the Mind Layer can power your AI employees.