Context Window

The context window is Amara’s working memory — the finite budget of tokens available on each turn to hold your request, the code being examined, and the conversation so far. Understanding it helps you get better results from longer sessions.

What fills the budget

Several things compete for space. Your instructions and the current request take some. The files and search results Amara reads take more — often the largest share on a coding task. The running transcript of the conversation, including every command and its output, accumulates over time. And the standing context from your .syntic/ directory — rules and memory — is present throughout. Because the budget is fixed, Amara reads selectively rather than loading whole files, pulling in just the parts relevant to the task.

Why it matters

When the window fills, older or less relevant material has to make room for new information. If that happens carelessly, an agent can lose track of an earlier decision or forget a constraint. Syntic Code manages this deliberately so the important context — your goal, the active files, and your rules — stays present while incidental detail is trimmed.

Compaction

To keep long sessions coherent, Syntic Code performs compaction: it summarizes earlier parts of the conversation into a compact form that preserves decisions, open threads, and key facts while discarding raw noise like verbose command output. This frees space without losing the thread of the work. You can also help by starting focused sessions for distinct tasks and clearing context when you switch to something unrelated. Pair this with prompt caching to keep both quality and speed high.