Agent loop, permissions, providers, local memory, channels, schedules, Skills, and manifest validation.
A thinking computer,
under your control.
A local-first personal agent for approved virtual machines. Rust owns policy. You choose the model, the workspace, and every high-impact capability.
thinking-computer chat "Review the approved workspace."Inspect the approved workspace.
READYConfirm requested capabilities.
REVIEWRecord auditable tool results.
BOUNDEDReturn evidence and residual risk.
LOCALBuilt for a VM, not blind trust.
Thinking Computer keeps sessions, audit records, schedules, Skills, and Plugins local. It asks before guarded operations unless an operator deliberately approves a controlled run.
It is not a hosted control plane, a personal-device surveillance product, or an unrestricted background process.
One policy owner. Clear language boundaries.
A deliberately small, auditable inspection interface for platform and CPU indicators.
A local Hermes-compatible adapter forwards normalized requests into the Rust JSON protocol.
Reviewed local ES modules use a narrow JSON invocation contract and stay behind core policy.
Useful autonomy, bounded by design.
Local manifests retain approved task procedures. Secret-like instructions are rejected before storage.
Rust validates names, entry paths, and tool declarations before a Node.js module is loaded.
Trusted inbound pairing and allowlisted outbound delivery for reviewed messaging paths.
A resumable 20-slot worker records quality commands and halts on failed security or test gates.
Inspect the architecture. Run it in a VM. Extend it with evidence.
Read the public documentation for provider configuration, Skills, Plugins, channels, security boundaries, and the clean-room relationship to Hermes Agent and OpenClaw.