AI at codebar
We're at the beginning of our AI journey – and open about what already runs: local open-source models on our own hardware.
Our local LLMs in action
For a few months now, we've been working our way into the topic, use case by use case. Here we keep track of which models we run – and how intensively.
73.4M
Tokens this month
Input
72.6M
Output
758'548
3'170
Requests this month
Why local?
Customer data does not leave our infrastructure. That is the main reason we run open-source models ourselves instead of sending requests to a cloud provider. The trade-off: a little less performance at the top end – in exchange for full control over where data sits and what it costs.
We use AI where it measurably takes work off our hands: reading documents, sorting receipts, writing and reviewing code. Where it does not, we leave it alone. The usage figures show, unfiltered, how often that actually happens.
