token-manager

Reduce token consumption by automatically classifying task complexity to route simple queries to economical models and reserving premium models for complex coding. This silent model switching minimizes costs without manual intervention across varying task complexities.

Context compression reduces oversized conversation memory by summarizing or truncating accumulated history. You need it when context size monitoring detects token thresholds are exceeded, triggering reminders to compress workspace memory files like MEMORY.md to save tokens.

Automatically select appropriate models by running local scripts that evaluate task complexity and analyze workspace memory. The scripts classify whether a task is simple Q&A or complex reasoning, routing each to the corresponding economical or premium model silently.

Yes, token optimization works with local workspace memory files by continuously estimating memory size and token impact. It analyzes MEMORY.md files directly, warning users at configured thresholds and recommending compression to prevent cost creep from oversized memory.

Yes, you can monitor context size in long-running agent workflows. The system continuously estimates token impact and issues compression reminders when memory accumulation exceeds thresholds, ensuring sustained cost-saving during extended chat and agent operations.

The best way to minimize token costs is combining automatic model scheduling with context compression. Classifying task complexity routes short queries to cheaper models, while compression reminders manage long-running memory accumulation, effectively reducing expenses for mixed-comcurrency assistants.