A small model is usually short of context, not intelligence

Memory is not only a saving. It decides which models can do the job at all.

The Problem

A smaller model is quick and cheap, but on your project it behaves like a stranger. It does not know your rules, your history or what was already settled.

So you reach for the biggest model every time, and pay for capability when what was missing was context.

How Stompy Helps

Keep the record, and let the model read what applies to the task in hand.

It stops being a stranger. In our own six-agent benchmark, Haiku 4.5 completed none of a set of 40 tasks without the record and 39 of them with it. In the same results a mid-size model with the record beat a larger one working without it. The tables, and the single run they come from, are on the deep dive page.

Over months, model choice becomes a question of fit and cost rather than of who remembers the most. You can use the cheaper, faster model for most of the work and keep the expensive one for the problems that need it.

Limits, plainly. These are our own benchmarks on our own kind of work, not a promise for yours. Memory can lose on a small, fresh project or on single-session work, where the overhead is not worth it. And how well a model reasons with what it retrieves is still the model's job. The limit, plainly: only what you lock is kept.

Integration Walkthrough

Let the model read what applies

Start from the brief, then read the entries that bear on the task by name or by search.

project_brief(project="my-project")
recall_batch(project="my-project", topics=["working_rules", "api_conventions"])
context_search(project="my-project", query="how we name migrations")