Usage Insights without cloud analytics

Developers need usage visibility. They do not need every local agent move turned into a management dashboard. The useful question is not “how do we track everything?” It is “what signal helps a developer understand their own workflow without making the tool creepy?”

Usage Insights starts from local evidence: sessions, token activity, response load, tools, agents, models, changed files, and daily rhythm. It is there to help the person doing the work understand cost pressure, context shape, and agent behavior.

Local evidence beats invented precision

AI tooling is full of fake certainty. If metadata is missing, MendCode should not pretend it knows. If a provider exposes useful token data, show it. If it does not, mark the gap. Good telemetry is honest about its source.

That honesty matters because Usage Insights is not a vanity chart. It is an instrument panel. It should help answer practical questions: which models dominate a day, which tools are overused, where sessions get expensive, and whether the workflow is improving or just getting louder.

Not surveillance dressed as productivity

There is a big difference between local operational insight and a cloud feed of developer behavior. MendCode keeps the default posture local-first because the developer should be able to inspect the evidence and understand the calculation.

The goal is not to score humans. The goal is to make the agent workflow legible enough to tune.