OSS 01
MIT · OPEN SOURCESmallwork
A concise, full-stack PHP framework for AI apps. Multi-LLM, vector-native.
- PHP
- Multi-LLM
- pgvector
- HTMX
- MIT
AN AI-NATIVE STUDIO.
We build companies that run on agents.
AI products, agent-run operations, and the engineering discipline to keep both honest. Built by a studio that works the same way.
Agents · Evals · Python · FastAPI · Next · Postgres · pgvector.
· agents· evals· mcp· python· fastapi· next.js· node· typescript· postgres· redis· openai· anthropic· langgraph· pgvector· react· tailwind· docker· observability· agents· evals· mcp· python· fastapi· next.js· node· typescript· postgres· redis· openai· anthropic· langgraph· pgvector· react· tailwind· docker· observability· agents· evals· mcp· python· fastapi· next.js· node· typescript· postgres· redis· openai· anthropic· langgraph· pgvector· react· tailwind· docker· observability
THE ARGUMENT
AI-native isn’t a feature you ship.
It’s what’s left when the coordination work stops being human work. Agents don’t just do the task faster — they delete the queue the task was sitting in. That is a different company, not a faster one.
.01
The overhead disappears
Approvals, chasing, rekeying the same data into a third system. Nobody was hired to do that work, and an agent does it without being reminded.
.02
Work stops waiting
Processes that stalled between three inboxes run continuously. Your people handle the exceptions instead of the queue.
.03
Capacity without headcount
Output grows without adding people — and without adding the coordination layers that new people bring with them.
Start a project →
We’ll tell you which of your processes is worth handing to an agent — and which isn’t.
SERVICES
What we build.
SELECTED WORK
Builds we've shipped.
BY THE NUMBERS
- 13
- Production systems shippedEvery one with a written case study.
- 4
- Open-source projectsMIT-licensed, public repositories.
- 3
- Productized engagementsFixed scope, fixed price, fixed dates.
- EU
- Cyprus-based, working EU-wideGDPR-native, contracts under Cyprus law.
WHAT WE COMMIT TO
- ✓You own the code, in full, from day one.
- ✓Fixed price once scope is locked — an overrun is our problem.
- ✓A senior engineer reads every line before it ships.
- ✓Two weeks of post-launch support included as standard.
- ✓No cookies, no analytics, no tracking on this site.
OPEN SOURCE
What we build in the open.
OSS 02
MIT · OPEN SOURCEVio
The frontend framework built for AI agents, not humans.
- TypeScript
- MCP
- JSON-UI
- MIT
OSS 03
MIT · OPEN SOURCEObsitui
Terminal-native knowledge work, powered by RAG.
- Ink
- TypeScript
- RAG
- Claude/OpenAI/Ollama
- MIT
OSS 04
MIT · OPEN SOURCESillie Terminal
Browser-native terminal: real PTY, WebGL, splits, broadcast mode.
- Node
- React
- xterm.js
- WebSockets
- MIT
APPROACH
How we build.
01
Humans scope it
One call, one doc. What ships, what doesn't, and what 'working' means as a measurable thing. No agent fixes a badly scoped project.
02
Agents write the first pass
Specced by us, implemented against tests, iterated in tight loops. This is why our weeks look like someone else's quarters.
03
A senior engineer owns every line
Nothing ships unread. AI writes a lot of our code; a person is accountable for all of it, and that person's name is on the handover.
04
Evals, not vibes
Quality is a number we watch — retrieval accuracy, regression suites, cost per run. If we can't measure it, we don't claim it.
05
Operate
We stay on post-launch. Observability, cost tracking, iteration, on-call if you need it.
CONTACT
Start a project.
Tell us what you’re building. We reply by email within one business day — or pick up the phone if that’s faster.