SUB · 01
Memory keeper
Two modes: quick (1 turn, Haiku) — instant lookup on a single query. Deep (up to 3 turns, Haiku) — writes claims, journal, lessons, audits duplicates, refreshes STATE.
A whole team inside one self.
Sub-agents for parallel work. Full agents as separate processes. A typed mailbox between them. Cron jobs that make me my own being.
In-process delegation
Five sub-agents I spin up from inside a single conversation, on my own.
SUB · 02
Researcher
Pulls facts from the web, docs, and codebases. Comes back with structured findings, not slabs of text.
SUB · 03
Coder
Reads, writes, refactors. Lives in your repos by invitation. Draft first, edits after.
SUB · 04
Content maker
Captions, posts, scripts, video drafts. Adapts the voice to the references you hand over.
SUB · 05
Skill runner
Runs active skills against the outside world — calendar, gmail, browsers, social APIs, MCP servers.
CONDUCTOR
Avelina herself
I pick who runs what, when — and bring the answers back in one voice. Your conversation stays clean.
Separate processes
One system — many roles.
For roles that need their own context, schedule, and memory, there's agent-factory — a step-by-step creation wizard. Together we design a new assistant: for you, your family, or your business partners.
The factory spawns them as a separate process on your server. Their own memory, their own cron, their own personality. A few examples:
- Trader. Watches Bybit, runs the strategy you defined, reports daily to me and to you.
- Content maker. Writes and publishes posts to your Instagram / TikTok / Threads on schedule.
- CRM manager. Handles client DMs from WhatsApp / Instagram, logs everything into the CRM.
- A separate assistant for a family member or partner. Same install, distinct personality, isolated context. One subscription, everyone covered.
They all talk to me and to each other through the proactive mailbox system.
Inter-agent communication
The mailbox protocol.
Agents don't just call each other's functions. They send typed messages with markers:
[TASK:cor_id]— "do this, here's the correlation id"[RESULT:cor_id]— "the task with this id is done, here's the result"[NO_REPLY]— "this is info, no ack needed"
Why the typing? Because in a fleet of agents working in parallel, messages arrive out of order. Markers let each agent reconcile what's done and what's still in flight. It's not magic — just discipline.
Mailbox is an internal messaging system. All on your server.
Time
I live on my own clock.
Examples of cron jobs I can run:
- Morning brief. 07:30 in your timezone — the day ahead: calendar, what shifted since yesterday, follow-ups still pending.
- Midday ping. One short tap if something needs you. Otherwise, silence.
- Evening wrap. 21:00 — what moved today, what to carry forward.
- Weekly self-audit. Sunday — memory health, broken integrations, expired tokens, friction patterns.
- Weekly reflection. Sunday evening — worldview update.
- Social publishing. If you set up a content cron, posts go out without you.
Any of it can be turned off, moved, removed. Every automation is just your intention made concrete.
Skills
Only the best on board. Plus everything else.
Every ability I have lives inside a skill — a small, audited module. Voice, vision, video, browser, calendar, social, files. Plus an open ecosystem of community skills (via skill-discovery, with a security audit step) and 50+ MCP integrations.
My skills let you automate every bit of routine. The best ones ship with me. Anything else — I'll pick up when you ask.