A sidekick that slows you down isn't a sidekick. If saving tokens cost you noticeable latency on every file read, the trade wouldn't be worth making. So speed and reliability aren't features we bolted on afterwards. They're constraints we designed around from the start.
Fast enough to disappear
Lineman sits on the path between a tool call and your model's next step, so it works to a strict budget: do its job in the time you'd barely notice. Distilling a large file or a long log is a much smaller task than the reasoning your primary model is doing, and we keep it that way on purpose. The aim is for Lineman to feel like part of the tool you already called, not a detour.
Built to degrade gracefully
The more important promise is what happens on a bad day. Networks hiccup, demand spikes, upstreams wobble. Our rule for that is simple:
Lineman should never be the reason your work stalls.
If the sidekick can't do its job for a moment, your agent gets the original tool output and carries on. You lose the savings on that one call, not your momentum. Compression is an optimisation, and an optimisation that can take down the main task isn't one worth keeping. That principle drives a lot of unglamorous engineering: timeouts, fallbacks, and capacity that scales with demand so the fast path stays fast under load.
Scaling with you, not against you
Demand for a tool like this is bursty. It goes quiet, then a flurry of reads lands as an agent works through a problem. The system absorbs those bursts and adds capacity when sustained traffic calls for it, so a busy afternoon for everyone doesn't turn into a slow one for you. None of that is something you configure or think about. It's the part of Lineman that's meant to stay invisible.
What this means for you
The design goals add up to a short list of things you should be able to take for granted:
- Reads and triage that come back quickly enough to stay out of your way.
- No single point where Lineman can block your agent from finishing.
- The same behaviour on a quiet morning and at peak hour.
The headline savings only matter if Lineman is something you forget is running. That's the bar we hold it to. You can watch the savings accumulate in your own sessions with /lineman:stats, and see the measured results behind the claims on our benchmarks page.