Find files.
Run long jobs.
Four places where seekfs and vanth take the waiting out of the work, for people and for agents.
Coding agents
Give agents a file finder that answers in milliseconds
Agents look for files constantly, and a slow recursive listing costs both time and context. seekfs answers a scoped, filtered query in tens of milliseconds and returns exactly the paths, in JSON if asked.
--underkeeps results inside the current workspace- Filter syntax small enough to put in an agent's instructions
- Pairs with ripgrep: seekfs finds the file, rg finds the line
Large repositories and drives
Search millions of paths in milliseconds
Monorepos, asset libraries and data drives grow past what Explorer search or a recursive listing can handle. The seekfs service keeps whole volumes resident and current, so the hundredth query is as quick as the first.
- Whole-volume indexes built from the MFT, kept live from the USN journal
- Recursive folder sizes, so
size:>10gb type:dirfinds what fills a disk - seekfs-ui when you would rather click than type
ML and data jobs
Train overnight. Read the curve in the morning.
Training runs, ETL and batch processing outlive terminals and agent sessions. vanth keeps them running detached, charts their metrics live, and can capture logs and register outputs as versioned artifacts with lineage.
- Wake on checkpoints, not only at the end
- Compare metrics across reruns with
job_metric_compare - Configure pools to limit concurrent GPU jobs
Tests and CI
Run the suite in the background. Get woken with the verdict.
Long test suites are where agents waste the most turns, checking whether they are done yet. With vanth the agent starts the suite, goes idle, and is resumed with the exit code, a summary and the failing output.
--wake-mecovers completed, failed, timeout, cancelled and orphaned- Flaky-test analytics from repeated runs of the same job
- Kill attribution says who stopped a job and why