Costs

Every prompt Syntic Code sends and every reply Amara streams back consumes API credits against your Syntic AI account. Understanding how that consumption works is the difference between a predictable line item and an alarming surprise at the end of the month. This page explains what drives credit usage and how to keep it under control without getting in your developers’ way.

How credit consumption works

The Syntic model bills by the amount of text it processes, measured in tokens, split between the input it reads and the output it generates. A request’s input includes not just your latest message but the accumulated context the agent carries: the conversation so far, the files it has read, and any tool results it has gathered. Long-running sessions on large codebases therefore cost more per turn than short, focused ones, because the context sent with each request grows. Output tokens, the text Amara produces, are counted separately and typically weigh more per token than input.

What drives cost up

Three factors dominate. The model tier matters most, since a more capable tier costs more per token than a lighter one, so reserve the strongest tier for genuinely hard work. Context size matters next, because feeding the agent an entire repository when a few files would do inflates every request. And task shape matters: automation that loops or retries can multiply usage quickly if it is not bounded. Autonomous auto-mode jobs deserve special attention, since there is no human to notice a runaway loop.

Keeping spend predictable

Pin a sensible default model tier through settings so expensive tiers are a deliberate choice, not the norm. Encourage focused sessions and scoped context rather than open-ended exploration of huge trees. Route traffic through a gateway so you can attach cost-center tags, enforce per-team quotas, and see spend broken down by group. Review usage against your baseline regularly, and set an alert on unusual jumps so a misbehaving pipeline is caught in hours rather than weeks.