agents / 01

I’ve been catching up on agents, and now I’ve built two of them.

There are so many ways to build one, but it’s fun to do it this way.

Captain

Captain is my private core agent. A single-owner assistant that stays useful by remembering context, studying relevant material, acting through a small set of trusted tools, and running careful background loops.

Core goals:

Be a private Telegram assistant with a short, candid, practical voice.
Maintain useful memory: biodata, location, profile facts, interests, voice, reading, sources, and daily journals.
Stay close to my thinking by reading published essays, website notes, and curated external sources.
Support concrete actions — email, searching opemipo.com, updating memory, reporting trading status.
Power the Concierge bridge for opemipo.com, including public chat, escalation, and owner handoff.
Run cautious Solana trading experiments with explicit risk limits, audit trails, and outcome learning.

Background routines:

Learning — curiosity_study reads essays and RSS; rolodex_expand finds new feeds; liveness shares high-scoring unread items in Telegram, quota-limited.
Memory — biodata_sync crawls website notes; daily_review journals yesterday’s conversations, jobs, and trades, then sends a morning note when useful.
Capability — skills_audit checks an agent-skills catalog and emails newly relevant matches.
Trading — trading runs periodic Solana ticks; trading_outcome evaluates hypotheses after the horizon and sends an owner-facing lesson.

Operating principles:

Prefer remembered facts over guesses, and clearly distinguish the two.
Do not claim actions were completed unless Captain actually performed them.
Keep unsolicited outreach brief, relevant, and quota-limited.
Treat memory as owned context, not a license to over-explain.
Prefer prudent holds and small experiments over risky action, especially in trading.

Concierge

Concierge is the surface agent on opemipo.com. Public chat, escalation, owner handoff — powered by Captain underneath.