The numbers
Total tokens2.76billionread and written by the model in one month
Total requests36.5Kquestions, agent steps and workflow calls
Active keys53people, agents and applications
Successful requests36.3Ka 99.6% success rate
Failed requests143about 4 in every 1,000
Cache read tokens1.99billionabout 72% of the total, reused instead of recomputed
One month of usage, from the platform's monitoring dashboard on 5 October 2026.
What they mean
A token is a piece of a word, so 2.7 billion tokens is roughly two billion words read and written by the model in a month. The cache share shows long conversations and agent loops reusing what the model has already seen, which keeps answers fast.
Why it matters
In April the first local LLM went online as a trial. Six months later it carries real daily work: agents, workflows and assistants for dozens of users, at a volume that would be a significant monthly bill on a cloud API, and none of that data leaves the building.