Same agent work. A fraction of the tokens.

The coding harness that finishes the job and stays in scope.

jecoha hands routine decisions to Jev, a fast small model, so the big model reads less, runs fewer turns and never works outside your intent.

Illustrative rundecision log

intent Add CSV export to the invoices page. Existing tests must pass.

jev? which_file → src/invoices/export.ts0.97 · acted

jev? in_scope "rename InvoiceRow" → noblocked

jev? which_tool → edit0.95 · acted

check invoices.spec.ts ✗ date formatfailed

jev? retry_or_stop → retry: fix date format0.89 · acted

jev? in_scope "upgrade date-fns" → noblocked

done 3/3 obligations met · each linked to evidence · 42/42 tests

01 · small model

Small choices go to a small model.

At each fork — which file, which tool, retry or stop — Jev answers a typed question. Code acts on confident answers.

02 · fixed intent

Your intent is the fence.

Actions outside the stated intent don't run. No drive-by refactors, no surprise upgrades.

03 · evidence

Done comes with a receipt.

A short report links each requirement to its evidence. A failed check isn't the end — the harness steers to a new move.

Planned at £20/month.

For individuals. Team pricing on request. Give us your spend and we'll measure what jecoha would save you.

Questions

jecoha is in development. Features on this page are planned, not live.