Turn-granular memory for AI agents.
ICE, the Indexless Context Engine, is turn-granular memory for AI agents that makes the context you already have useful.
Embedded.
ICE runs entirely within your application process. No external servers, database daemons, or network round-trips required.
Optimized.
Core memory operations are handled by an optimized, compiled binary kernel, bypassing Python runtime overhead.
Type-Safe.
Surgical control over cognitive defaults, budgets, and query limits via a clean SDK validated with strict Pydantic models.
How ICE applies systems thinking to agent memory.
Turn-granular storage
Each conversational turn is stored as its own frame, like a frame in a film, so the order of events survives. Fixed-window chunking loses it.
Multi-lane retrieval
Retrieval is organised like the sections of a library: episodic for what happened, semantic for what it means, procedural for how to do it, and working for what the agent is focused on now.
Cross-session persistence
Memory carries across sessions, so each interaction can build on the last, like a feedback loop.
Useful, not just long.
While others chase longer contexts, we ask what makes the information in the window useful. ICE works on the signal-to-noise ratio of the context an agent already has.
Session 1 → Session 47. Same memory.
Zero-Conf.
Embedded Memory.
Boots instantly with an Indexless memory file (`~/.ice/ice_memory.db`). No Docker, no PostgreSQL, no daemons required. Just wrap your OpenAI client and get started.
- 100% Embedded (runs completely locally)
- Optimized Binary Core (bypasses Python overhead)
- Type-Safe Enterprise API (Pydantic validated)
Three lines. That's the whole integration.
Memory capabilities.
Pure, embedded memory engine for Large Language Models. Mount virtual memory directly into your application.
Embedded & Private
ICE runs entirely within your application process. No external servers, database daemons, or network round-trips required.
High Performance Core
Core memory operations are handled by an optimized, compiled binary kernel, bypassing Python runtime overhead.
Type-Safe SDK
Surgical control over defaults, budgets, and query limits via a clean SDK validated with strict Pydantic models.
Enterprise Escape Hatches
Inject your own logic before or after a query using middleware hooks for PII scrubbing, toxicity filtering, or custom audit logs.
Memory Guardian
The scheduler automatically monitors VRAM pressure and down-quantizes precision to prevent out-of-memory system crashes.
Seamless Integration.
Deploy ICE exactly how your stack demands. Compiled native core.
ICE for Python
Compiled SDK for AI engineering and LangChain.
Simple Licensing
Start free. Scale when you need to.
🧊 Community
Free for individuals and research. Local-first, zero-config.
- Individual & research use
- Local-first, no cloud required
- Full core features
Enterprise
For teams needing compliance, support, and scale.
- SSO, RBAC & audit logs
- Priority support & SLAs
- Air-gapped deployment option
Built for Production.
If you are shipping persistent enterprise copilots or multi-tenant AI products, stop rebuilding brittle Postgres/Redis memory workarounds. Drop ICE into your stack today.
GoingMerry is the next product we are building after ICE.