VEKTRA
Hello.I’m ILO.
I live across your inbox and your calendar — I know what you owe, what you’re owed, and what’s worth your attention before you’re in the room. One memory, one context. Every move waits for you.
First 100 founders get concierge onboarding. We’ll wire your stack with you on the call.
The work that doesn’t scale.
Founders trade their best hours for admin that has nothing to do with the work only they can do. ILO is a chief of staff with a team of skills behind her — not a chatbot, a presence that remembers and follows through.
I read what came in overnight, flag the three that need you, and draft the replies in your voice. You wake up to triage, not chaos.
I book, reschedule, and defend your focus blocks. You walk into every call already briefed — not scrambling.
I track who owes you what. I nudge them on the right day in the right tone. You stop being the bottleneck.
I watch your projects across docs and tools. I surface what slipped before standup and hand you the summary — not the dashboard.
I pull the brief before the call. Cited sources, the questions worth asking, the context you'd otherwise be Googling at 8:59.
Drafts the email, files the expense, schedules the demo. I always ask before sending — you sign off, I ship.
ILO is our bet on the first AI that earns a seat at the table — not as a tool, but as a team. Open architecture, audited, owned by you.
Most agents reply. I move.
The chatbots you know are good at answers. I’m built for the part after the answer — the part where someone has to actually do the thing.
Answers your question.
Hands you the answer, the draft, the calendar block.
Tells you what to do next.
Does it. Then asks before it ships.
Forgets the conversation.
Remembers your company, your people, your promises.
The agent, in numbers.
Carries weeks of conversation, not minutes
One memory across every surface
Spotify, Discord, GitHub, Calendar, Telegram
Anthropic, OpenAI, Bedrock, Vertex, local
I plug into what you already use.
ILO speaks the same open protocol Claude and ChatGPT use. Five connectors at launch, more on the way. Every write passes an approval gate. Your tokens stay encrypted with per-user keys — never logged.
Read events, draft holds, surface conflicts.
Where ILO lives by default — message, photo, voice notes.
Search, play, queue, build playlists from a vibe.
Read repos, file issues, comment on PRs, run actions.
Read channels, post updates, run commands across servers.
Read events, draft holds, surface conflicts.
Where ILO lives by default — message, photo, voice notes.
Search, play, queue, build playlists from a vibe.
Read repos, file issues, comment on PRs, run actions.
Read channels, post updates, run commands across servers.
OAuth flow handled in a popup. No URLs, no tokens to paste, no copying client IDs into config files.
Allow / Ask / Deny on every capability. Read-only tools auto-allow; writes stay behind the approval gate.
Every write asks before it ships. Tool calls are logged with args, result, and your decision — so you can audit what I did and why.
She remembers. She specializes.
Two layers most general-purpose agents skip. Memory means I walk into every conversation already knowing who you are. Skills mean I run your recurring tasks the way you want them run — without you re-explaining.
Auto-extracted, user-editable, persistent.
After every turn, I mint memories for what’s worth remembering — your role, preferences, ongoing projects, references. Next turn, the relevant slice folds into my context before I reply. You can read, edit, or forget any memory from Settings.
- · Five memory types: user, feedback, project, reference, note
- · Postgres tsvector retrieval — sub-100ms, scales to 10k+ memories per user
- · Encrypted at rest, RLS-isolated per account
- · Manual entries supported — "Teach ILO something"
Markdown playbooks for the work you repeat.
Write a skill once — "Inbox Triage", "Calendar Prep", "Pablo's Voice" — and ILO loads it whenever the task matches. She becomes a domain expert tuned to you, not a generic chatbot waiting for instructions.
- · Three platform builtins ship by default — Calendar Prep, Voice, Daily Briefing
- · Auto-matched by trigger keywords + full-text relevance
- · Pinnable from the composer for forced inclusion next turn
- · Editable, exportable, owned by you — not locked to one provider
Model card.
ILO is model-agnostic by design. Pick the right tier for the task — Haiku for speed, Sonnet for everyday work, Opus for hard reasoning — and swap providers entirely when the operator experience needs to stay constant.
One agent architecture. Multiple backends. A single memory and control plane.
Memory is persistent and per-user. ILO auto-extracts what matters from each conversation — preferences, ongoing projects, references, corrections — and surfaces the relevant slice on every turn. You can edit or forget any memory from Settings; she catches up next turn.
Skills are reusable playbooks. Calendar prep, your writing voice, daily briefing — write them once and ILO loads them when the task matches. Pin the ones you want forced into the system prompt for the next turn.
Every action emits an audit trail with the decision, args, result, and timestamp. Read-only tools auto-allow. Mutating tools queue for approval. Reversible actions get an Undo. Allow / Ask / Deny per tool — your call, not ours.
Who I am
I work in the background. I stay current.
I’m the first agent from Vektra AI — a private cognitive companion. I listen, remember, reason, and act only on what you authorize. I stay useful in any context where presence and follow-through matter — and out of the way for anything you should decide yourself.
Other agents are built on me. Partners and vertical programs deploy custom personas on the same cognitive core — branded, tuned, scoped — inheriting my audit, routing, and human-in-the-loop guarantees by default.
Auditable end to end. Model-agnostic core. Human-in-the-loop on every action of consequence.
In the loop
I carry the queue
while you’re away.
I read what’s come in, draft what’s ready, hold what isn’t. When you return, it’s sorted, attributed, and in order. No surprises, no missed signals.
- Triage every inbound while you sleep.
- Drafts staged for one-tap approval.
- Anything urgent escalates in real time.
How I’m built
One core. Many hands.
A single cognitive core handles memory, reasoning, and routing. Around it sit the layers that do the work — skills, connectors, sub-agents, audit. Each one is replaceable. None of it is locked to a single provider.
The thinking layer. Memory, reasoning, planning, routing — provider-agnostic, swap models without rewriting the agent.
Markdown playbooks I load when the task matches. How you want recurring work handled, captured once.
MCP-native tool layer. Calendar, Spotify, GitHub, your custom server — every action passes the approval gate.
Every reply, action, and decision logged with prompt, retrieval set, and result. Allow / Ask / Deny per tool.
← scroll to explore →
Why I exist
“Keep humans alive, keep humans strong, keep humans human.”
That’s the Vektra mission. I’m how it shows up in cognition — the part of the harness that listens, answers, and stays awake so you don’t have to.
Doors open soon
Your move.
ILO is in active development inside Vektra AI. Drop your email and I’ll come find you when it’s your turn.
Built by Vektra Industries · Vektra AI Division
Become a launch partner.
We pay partners who put ILO in front of founders and operators. Unique referral link, transparent terms, real revenue share.