Human in the loop
You inspect and approve proposed actions. Control stays with the person, not the model.
Independent development. Intentional execution.
HayanGom Labs
Building safe local AI execution systems.
GAREUM — A local-first AI orchestration platform that lets frontier AI models safely perform real tasks on a user’s computer.
Frontier AI modelsReasoning & planning
GAREUM
User-controlled execution
Your computerLocal tools & tasks
Local-firstMCP-orientedHuman-approved
01 / Why GAREUM
Powerful reasoning.
Deliberate execution.
AI models are increasingly capable of reasoning, planning, and proposing useful work. But unrestricted access to a person’s computer introduces a different kind of risk.
GAREUM adds a controlled execution layer. Models propose actions; the local runtime handles execution through explicit approvals, scoped permissions, auditability, cancellation, and rollback.
02 / How it works
A clear boundary between
thinking and doing.
The model proposes. You stay in control.
GAREUM executes within the boundaries you approve.
- 01
AI Reasoning
plan / proposeA model interprets the goal, builds a plan, and proposes actions. Reasoning is separate from device access.
- 02
User Approval & Policy
inspect / approve / cancelThe user reviews the plan. Explicit approvals and permission boundaries determine what can proceed.
- 03
Local Execution
execute / audit / rollbackGAREUM carries out approved actions locally, with a record of execution and recovery for supported actions.
03 / Principles
Control is part of the foundation.
Local-first execution
Tasks run through a local runtime on your computer, within controlled permissions.
Provider-neutral
A flexible execution layer for frontier models and MCP-compatible integrations.
Auditable & recoverable
Trace what happened, cancel ongoing tasks, and roll back supported actions.
04 / Current status
From working prototype
to a considered product.
Under active development
- Working Windows prototype under active development
- Local runtime and controlled tool execution
- MCP-oriented integrations
- Productization and startup-program preparation in progress
05 / Claude & MCP
Built for an open
AI ecosystem.
Frontier models MCP integrations
Claude can support reasoning, planning, coding assistance, and agent workflows. GAREUM provides the local safety and execution layer that connects those capabilities to real tasks under user control.
GAREUM is not Claude-exclusive. It is provider-neutral and designed around MCP-compatible integrations, so the reasoning model and the execution layer can evolve independently.
06 / Contact