The Personal Agent at builddaily.io
Test — quality audit and privacy review before the public announce.
There's a chat on the homepage of builddaily.io. Ask it what I've been working on, what my stack looks like, what I've written about a topic — it reads my markdown corpus and answers from inside it. That's the Personal Agent.
How it works
The agent reads a curated subset of my markdown — public-safe notes, journal entries, project docs — through a container that mounts those directories read-only. A declarative LLM responder module shapes the answers. Inference runs on Ollama locally; the chat is reachable from the public web through a Cloudflare Tunnel. The corpus never leaves my machine.
What it's for
Instead of asking visitors to read everything I've ever written, the chat lets them ask the question they actually have. "What's Sage Daily?" "What's the consumer-lending project?" "What's your stack?" The agent answers from the corpus, in something close to my voice, citing the source note when it's useful.
- Surface specific projects, posts, daily-log entries.
- Explain the stack, the active focus, the agent system.
- Quote what I've actually written rather than paraphrase it.
- Decline (sometimes) when the corpus doesn't have the answer.
- Speak for me on anything I haven't written about yet.
- Reason about my private notes — the privacy review keeps those out.
- Replace the post itself when a careful read is what the visitor actually needs.
The agent is live on the site today. If you ask it something and the answer is weird, that's interesting feedback — the gap between what the corpus says and what the chat surfaces is exactly the kind of thing this build-in-public surface is for.