Long browser tasks are mostly repetition
A 100-step enterprise workflow often looks like this: log into a portal, open the next claim, copy three fields into a tracker, set a status, and repeat forty times. The difficulty comes from length, not depth. Step forty usually looks like the steps before it.
Context rot, not model size, breaks the run
Each step adds to the context. Over time, stale pages and dead history pile up, and the model starts to focus on information that no longer matters. We call this context rot. It affects frontier models too.
Ramain keeps late steps as reliable as early steps
Ramain keeps compact summaries of the work so far and compresses history that the task no longer needs. The agent sees what matters for the next step, and nothing more.
Every point on the frontier is a Ramain configuration
We ran the same harness on three models. Each one sets a point on the cost and success frontier, so teams can choose the trade-off they need.
Each point is the same Ramain harness on a different model. Ramain used 37,000 context tokens per task, against 165,000 for Codex.