Studio · MX
Tools, protocols,
and products.
A one-person studio in Mexico City building with AI agents. Small surface, sharp focus, public APIs.
A one-person studio in Mexico City building with AI agents. Small surface, sharp focus, public APIs.
Shared rails anyone can build on — published specs and indexes, free and agent-friendly.
Open protocol for global money routing — exchanges, banks, remittance, and stablecoin rails in one graph. ~80 adapters, free public API, agent-friendly.
Marketplace and registry for AI agents. 1,841 MCP servers across 37 categories, 25 agent & payment protocols indexed. Free; tipping over x402 / MTP.
Live community directories, each running on the ceiba.to pattern. Every one owns its own data and exposes a public API.
Live stand-up, improv, and comedy across Latin America in one listing. 227 comedians, 188 venues, 29 cities — with claimable profiles and a public API.
The social-dance directory for Mexico — salsa, bachata, cumbia, kizomba, tango. Events, venues, schools, and instructors, each with its own public profile.
Tools built for a specific operator, then generalized so they can drop into any business like it.
Operations dashboard for small hospitality — POS, event calendar, daily P&L, payroll, supplier ledger. Built for a bar in Mexico City; designed to drop into any café or bar.
A 3D globe of street-lemonade prices as a proxy for cost of living. 601 hand-collected data points across 21 cities, 14 documented sources, public API, daily exports.
Everything above is built on the same in-house operating layer. This is it — and the terminal agent we drive it with.
Knowledge-Oriented Assistant — the studio's in-house operating layer. The shared workspace, memory, and agent plumbing that lets one person run like a small team. Built first for ourselves; pieces open up as they mature.
The terminal agent we use to drive KOA. Streamed tokens in the shell, an interactive TUI with a live koala mascot, inline diff approvals, and deep ties to our git monorepo.
KOA Labs isn't one human typing code. It's a single operator orchestrating 22 specialized AI agents that work asynchronously inside one monorepo. The operator assigns the work; each agent executes against its own prompt, tools, and access policy.
Agents that triage the inbox, shape plans, and route work to the right specialist.
Specialists that edit and refactor code inside the submodules, following shared patterns.
Agents dedicated to code review, automated smoke testing, and verifying the security guards.
Agents that watch the boring-but-critical layer: deploys, backups, and what's running where.
Instead of manual task tracking and fragmented tools, we consolidate our operations into a single continuous pipeline.
The operator drafts a plan inside koa-todo or writes up the strategy.
A specialized crew member is assigned and checked out to an isolated worktree.
Agents edit, refactor, and self-test files inside their isolated directory.
Test suites validate output; Caddy reloads, and the code goes live.
Native workspace AI agent & developer companion. Stream-to-terminal token generation, interactive Textual TUI with a reactive ASCII koala mascot, inline code diff approvals, and deep integration with git monorepos.
Knowledge-Oriented Assistant. The studio's in-house operating layer — workspace sync, durable memory, agent plumbing, infra and ops. Built first for ourselves; pieces open up as they mature.
KOA Labs is not just one human typing code. It is an orchestration of 22 specialized AI agents working asynchronously inside our monorepo. Miguel (the operator) assigns tasks, and the agents execute them based on their custom prompts and tool capabilities.
Agents that manage workflows, prioritize, write business logic and direct other specialists.
Specialists that edit and refactor code inside submodules, implementing custom patterns.
Agents dedicated to code reviews, automated smoke testing, and verifying security guards.
Automated bookkeepers that audit bank accounts, calculate daily P&L, and parse invoices.
Instead of manual task tracking and fragmented tools, we consolidate our operations into a single continuous pipeline.
Inge designs plans inside koa-todo or documents general strategy.
A specialized crew member is assigned and checked out to an isolated worktree.
Agents edit, refactor, and self-test files inside their isolated directory.
Test suites validate output; Caddy reloads, and the code goes live.
We don't build throwaway products. Everything we make is designed to run for years, not quarters.
Built for humans and agents equally. Every tool ships with a public API. No walled gardens.
No VC, no board, no roadmap dictated by investors. We ship what we think is worth building.
Built in Mexico City, for the world. Local roots, global reach.
We ship every day, in public. So now and then you'll catch something mid-build or a little rough around the edges — if you do, chances are we already know, and it's on the list.
KOA Labs is a solo + AI studio from Mexico City. One human founder working with a team of AI agents — building tools, protocols, and products that ship faster than a traditional studio could. Some open source, some for clients, all designed to be useful for a long time.
We like simple tech that works: vanilla JS, small APIs, plain SQL. We don't chase frameworks. We ship things that are easy to understand, easy to maintain, and easy to extend.
When we build in the open, it's because we think the thing is more useful that way — not as a marketing move.
/v1/sources, /v1/metrics, structured logs, 12-step smoke suite.
/.well-known/agent), CJK fonts, x402 payments live on Base L2.
Small, sharp 2–8 week engagements. Internal tools, dashboards, or a thin product cut you can iterate on yourself afterward.
If you're shipping an open protocol or registry — schemas, scaffolds, runbooks. We've published a few of our own.
MCP servers, agent-friendly APIs, payment integrations (x402 / MTP), discovery files. We dogfood all of it.
Fork, embed, or extend any of our projects. Issues and PRs welcome on GitHub.
One human, working with agents. Honest about scope, honest about pace.