How We Built Myra, Olbrain’s First Agent

Every company has a front door. Investors write in about the round. Journalists ask for an interview. A possible customer wants to know whether agents could run their operations. A developer is stuck connecting Alchemist. A job applicant wants to know where their application stands. Each of them deserves the right answer, fast — and some of what they ask must never reach the wrong person.

At Olbrain, all of these people now talk to Myra, our First Agent. This is how we built her, and what she taught us about building agent systems an enterprise can trust.

Investors, journalists, possible customers, builders and job applicants all write to Myra, Olbrain's First Agent, on WhatsApp, email and website chat. Behind her, four specialist agents answer quietly: Investor Relations, Pre-Sales, Media Relations and Technical Support.

A new role, not a receptionist

Agentic enterprises open up roles that never existed for humans. The first of them is the role that meets everyone who reaches the company. We call it the First Agent.

A First Agent is not a receptionist who forwards calls. She knows the whole organisation, decides what each person needs, and answers on its behalf. That takes judgement, so we built Myra as a Superagent — Olbrain’s name for a multi-agent system — rather than as a single chatbot.

Built from a conversation, not from code

Myra was not written by an engineering team. Alok Gotam, Olbrain’s CEO, built her by asking Claude to connect with Alchemist, Olbrain’s Agentic-AI Developer. Claude brought the context: who Olbrain talks to, what we can say publicly, what must stay private. Alchemist turned that into a working agent system, tested it, and published it on the Agent OS, where it runs and is governed in Noesis.

One voice, five agents

Myra is a team of five agents. Visitors only ever meet one of them.

The Lead Agent reads every message, decides who the person is and what they need, and replies as Myra.
Investor Relations handles investor questions and checks who is asking.
Pre-Sales works out whether Olbrain fits a business and quotes only published prices.
Media Relations answers journalists from approved facts.
Technical Support answers setup and developer questions from our documentation.

When a question needs a specialist, the Lead Agent consults them quietly and answers in her own voice. The visitor never hears about hand-offs, routing or internal teams. It always feels like talking to one person.

Wherever people already are

Myra answers on WhatsApp, by email and in the chat on olbrain.com. Each channel follows its own manners: short plain messages on WhatsApp, full sentences by email, tap-to-open links in website chat. A website visitor can move to WhatsApp in one tap, with their first message already typed.

Trust is the hard part

Answering questions is easy. Answering them safely is the real work. Myra’s rules are written for the situations that would embarrass a company.

Investors. On WhatsApp and email she quietly checks the sender against our investor list. Typing a fund name or a title is never proof. Only verified investors receive investor material, and nobody can tell from her reply whether a check happened.
Journalists. She answers product questions, sends interview requests to press@olbrain.com, and never shares funding details with the press.
Honesty. She never invents a fact, a price or a link. When she doesn’t know, she says so and points to the short Discuss your use case form.
No false promises. She never claims to have flagged or passed anything on, because she can’t.
Focus. She politely declines homework, coding help and anything unrelated to Olbrain, however hard someone insists.

Tested before every release, governed after it

Every change to Myra is tested before it reaches anyone. The Lead Agent alone runs 50 test conversations, covering cases such as a journalist asking about funding, an investor who can’t be verified, and a visitor who sends the same question twice.

Every change is also a new version. Nothing reaches the public until we make it live, and any earlier version can be restored. We review Myra’s real conversations, and every slip becomes a new test, so the same mistake cannot quietly return. In her first weeks the Lead Agent went through more than 60 versions. That is what governed iteration looks like: fast, but never blind.

What we learned

Context is the starting point. The person building an agent system knows their business best, and Claude or Codex already hold that context. Alchemist turns it into agents.
Guardrails matter more than answers. Most of Myra’s rules are about what she must never say.
Tests turn judgement into proof. A rule is an instruction; only a passing test shows the agent follows it.
Governance makes speed safe. Versions, tests and review let us change Myra daily without risking what she says to an investor or a journalist.

Talk to Myra

Message Myra on WhatsApp at wa.me/919818691119, email myra@olbrain.com, or open the chat on olbrain.com.

Want to build your own agent system? Ask your Claude or Codex to connect with Alchemist, or tell us about your use case through the short form at olbrain.com/discuss.