Meet Glove Core — the deterministic state engine underneath every Glove surface. Patent pending.
Glove is the execution layer where AI, humans, and software coordinate from a single shared state. You design the process — Glove Core assembles the agent at runtime, from your business context.
Every actor — human, AI, software — reads and writes the same business context. Nothing asks twice, nothing contradicts.
Assembled → executed → discarded. Glove Core builds each agent from the process definition and the current state, runs the turn, and discards it. The state persists; the agent does not.
The LLM speaks; Glove Core routes. The model writes the language of each turn, but never decides which state comes next.
Memory lives in state, not in conversation history. Turn 10,000 costs what turn 1 costs, and accuracy does not drift as history grows.
Every action takes the same path. The model proposes it; Glove Core checks it against your rules and the current state; only then does it execute — through an API, an MCP tool or the legacy operator. Nothing reaches the outside world without passing the check, and every step is logged.
Every state transition and every human intervention is written to the log as the process runs — nothing is reconstructed afterwards. Auditable end to end and ready for the EU AI Act.
Input, decision, action and outcome are sealed into one record: who acted, what changed, when and why.
Transitions are written by Glove Core as they happen, timestamped and replayable.
Approvals, edits and takeovers are logged like any other transition, with the person who made them.
The model writes the words; the path is the one you defined, executed the same way on every run.
Glove Core runs under three surfaces — Tandem, Dualdrive, AI Voice. See what it does for you → Product
Memory lives in the state, not in a transcript that is replayed on every turn. Each turn reads the state relevant to that step, so turn 10,000 costs what turn 1 costs — and accuracy does not drift as history accumulates.
Every state transition and every human intervention is written to the log as it happens, not reconstructed afterwards. Any run can be replayed step by step, which is what an EU AI Act audit asks for.
Yes. The channel is a transport, not a memory. A process can start on WhatsApp, continue by email and finish on a call, reading and writing the same state throughout.
The process definition and the accumulated state belong to you and stay put. Models are called as bounded actions, so swapping one does not rewrite how your business operates.