A research agent with a wallet.
It answers from free sources first. When the missing piece sits behind a paywall, it can buy one article straight from the writer, inside a budget you set. Four ways in, from a two-minute replay to the full system map and the build log.
Inside ResearchAgent
Replays a real run from its event log: free search, a purchase, a proof that fails, an on-chain refund, then a second buy that answers the question.
Slide deck
The AI Tinkerers deck. Boxes on the architecture slide open the project map.
→ next · ← back · O overview · N notes · F fullscreen
Project map
Three clickable levels, from the reader's agent, the writers and the ledger down to the primitives. Every arrow is numbered and explained.
How it evolved
39 days in charts: commits, pull requests, coding-agent prompts and tokens, replayed from git, GitHub and the agents' own logs.