Dumping the repo vs Hexum
Last updated 12 September 2026
| Repo dump | Hexum | |
|---|---|---|
| What the model sees | The repository, or a large slice of it | This task, trimmed |
| 24-task context tokens | 220,393 | 27,352 |
| You keep | The editor | The same editor |
Numbers and method: how MBPP and HumanEval were run. Tiny prompts may save little.
FAQ
How does Hexum cut what I pay the model?
Your agent stops dumping the whole repo into the prompt. Hexum sends this task, not every file. On the published card, MBPP context dropped 85.2% and the combined 24-task cut is 87.6%, with 24/24 tests still passing. Tokens are ceil(utf8_bytes/4), not a provider invoice. See /benchmarks.
What is Hexum?
Hexum is a developer ecosystem for coding agents in Cursor, Claude Code, and the rest. One API key. Five products and twelve tools today: spend less on each prompt, block bad imports, review pull requests, stop the agent looping, and only keep building when checks pass. We do not keep your code.
What does the dashboard mean by tokens saved?
How much smaller the reply was than what the agent sent, when that is a cut. Your invoice is still based on what the agent sent Hexum, not on the savings.