COGVO

Lab / Active experiments

Build to discover constraints.

These are research instruments, not polished claims. Each project exists to make an assumption testable.
L–01MODELING

Token Credit

Create a provider-neutral unit for the economic cost of completed AI work.

Method
Benchmark tasks · repeated trials · median resource use · quality adjustment · exchange-rate smoothing
Next test
Define the first falsifiable benchmark set.
L–02PROTOTYPING

Personal Agent Harness

Give a local agent efficient perception and browser control without sending every screen to a vision model.

Method
DOM · accessibility tree · OCR escalation · visual fallback · local state
Next test
Measure accuracy and token cost across perception modes.
L–03SYSTEM DESIGN

Cogvo Router

Automatically match every task to the best open model or the lowest-cost toolchain that clears a quality threshold.

Method
Task classification · capability registry · live cost and latency · benchmark quality · automatic failover
Next test
Define routing objectives and measure switching overhead.
L–04RESEARCH

Cogvo Node

Turn this site and its notes into a portable, user-controlled research agent.

Method
Local model · structured knowledge base · replaceable inference endpoint · auditable citations
Next test
Design the content schema before choosing a model.