Stop paying models to rediscover the same state.
Kell is a local memory membrane for AI agent systems. It keeps receipts, context, verification loops, and lightweight transforms on-box so expensive model calls are reserved for decisions, not repeated archaeology.
Lower Agent Cost
Replace repeated context reloads with local K-RAM receipts and bounded tool outputs. Success metric: reduce repeated context/tool-token spend by 30%+ in long-running agent sessions.
Faster Local Proof
Run validation on local memory and disk first. Current local suite: active Kell programs execute from /dev/shm through gotokell.
Zero Egress Mode
Keep private state local. The model gets only compact decision context; secrets, logs, and bulk payloads stay on owned hardware.
Available Evaluation Kit
- Standalone native runtime:
gotokell kt-2.0. - Active Kell language suite: 23 runnable
.klprograms. - Local RAM benchmark receipt: 23/23 passed from
/dev/shm/kell_local_bench. - System health receipt: ports
9000,8000,7429OK; local door HTTP 200. - Commercial ask: one-week pilot with an AI IDE or agent infrastructure team.
Who Should Evaluate Kell
AI IDE infrastructure teams, agent runtime teams, supercomputing workflow teams, and enterprise engineering groups paying real money for long agent sessions. Kell is not a cancer product. Cancer is one future benchmark payload. The product is local-first agent memory and runtime efficiency.