What we do
We build AI products and the developer tooling that makes them cheap to run. Small, local-first, and shaped to one job rather than every job.
Large models and heavy frameworks are easy to reach for and expensive to keep. We work the other way: pick the smallest thing that holds the job, measure what it costs, and cut the rest.
Where a script, a test or an exit code can decide something, we let it. The model is asked only for the part that genuinely needs judgement. Deterministic work is cheaper, and it fails the same way twice.
This work rests on thirteen-plus years of engineering across payments, logistics, health and SaaS. The same discipline shapes how we build with models.
We are open to engagements with enterprises and startups on modern tooling, agent workflows, and token and cost efficiency.
Work
-
Describe a goal. It plans, edits, tests and commits, spending the big model only where it counts.
-
Sits inside a coding agent and keeps it from reading or writing more than the job needs.
-
Reads a job posting, works out whether it is worth your time, and writes the application.
-
Photos and notes go in. It works out what belongs together and publishes the result.
-
We put Gemini inside a terminal coding agent back when the vendors had not shipped one.
-
One editor that scores, captions, grades and cleans a video, instead of four separate tools.
Get in touch
If any of this lines up with what you are building, write to us. We answer plainly.