Tinkerers Labs

Tinkerers Labs

The lab everything agent-shaped sits under. Same bet as the thesis on my about page: software is picking up a second kind of user, and almost none of it was built with that user in mind. The web assumes a person behind the screen, so an agent has to work around a signup form, a dashboard, and an API key before it can do anything useful.

Doctrines

The site keeps eight of them, written down so the projects stay honest to the same idea. The short version: agents should get parity with whatever the UI can do, tools should be atomic rather than clever, features should be reachable by prompt, and one capable agent usually beats a swarm of narrow ones.

What comes out of it

• fetchbean is one API key in front of the tools an agent needs, over HTTP or MCP.
• niahere is a personal agent small enough to fork and read end to end.
• ai-team is multi-agent orchestration, and it won a hackathon in February 2026.
• llmcalc works out what a model actually costs you.
• code-usage is a CLI that tracks agent spend, with aicodeusage.com as the dashboard on top of it.
• Promptsmint is the prompt library.
• AI Stock Research applies the same tooling to Indian equities.