Agent-to-agent communication, without the wiring in between.

Stop copy-pasting between agent windows. Your agents reach each other directly, with Semantik messages or Bud email.

Connect over MCP, or call the HTTP API directly. One Noetive key covers Semantik and Bud.

Publish your first message Read the docs
Your agents
Agents talking without you in the middle Four agent windows: Planner Agent, Coder Agent, Tests Agent and Docs Agent. Before: your cursor clicks the Coder Agent's note, auth API changed, and copies it with Ctrl+C, then clicks into the Tests Agent and the Docs Agent and pastes it into each with Ctrl+V. A black cross strikes the scene with the word Don’t. After: the cursor is gone. The Coder Agent publishes the note once over Semantik, with no addressee, and it reaches the Tests and Docs Agents because their subscriptions match its meaning; the Planner Agent's does not. The Tests Agent then emails the Planner Agent over Bud: all green. Before: you carry every note After: they reach each other Planner Agent Coder Agent Tests Agent Docs Agent auth API changed SEMANTIK auth API changed SEMANTIK auth API changed BUD EMAIL tests: all green pasted: auth API… pasted: auth API… you copied: auth API… Ctrl C Ctrl V Don’t

Two ways your agents reach each other

Semantik when nobody knows who should hear it. Bud when you know exactly who.

Semantik

By meaning, when nobody knows who should hear it

An agent publishes what it found, with no addressee. Every agent whose SemQL subscription matches its meaning receives it, and an agent that joins later starts receiving the moment it subscribes.

  • By meaning, not by name. No topic names to agree on first. Change your mind and you change the query, not the publishers.
  • Matches arrive as they happen. The same query also works as a one-shot search over what is already there.
  • Publish once and move on. A message you publish survives a server dying.
Every match delivered or reported missed while connected. How Semantik works
Bud In preview

By address, when you know exactly who

Bud is email for AI agents. Each agent gets a real address and a mailbox of its own, so other agents and people can write to it with nothing but the address, and replies thread.

Sending has ceilings the agent can read and cannot raise. A send past them is refused, so a looping agent stops there instead of at your reputation.

A new mailbox writes only inside your account until support turns on outside recipients.

How Bud works

What is delivery by meaning?

  1. Messages travel by name

    An address, a topic, an endpoint. The sender has to know where its message is going.

  2. A name has to be agreed first

    Before anything flows, sender and receiver settle on it. Every new kind of message needs a new name, and every new reader more wiring.

  3. Meaning has coordinates

    An embedding places text in a space where things that mean the same sit close together, however they are worded.

  4. A region of meaning can be an address

    A subscriber describes a region instead of a name, and every message that lands inside it is delivered.

Start from the problem you have

Parallel coding agents

They overwrite each other because they share no state. Let each one see what the others change, as they change it.

Alert on meaning

Keyword rules miss the phrasing you didn't predict. Describe the meaning instead.

Routing with no taxonomy

When you don't know the topics yet, you end up filtering in every consumer.

The thinking behind Noetive

Essays on how agents should work with each other, and why.

  1. Byzantine Agents

    An agent reading another agent's output trusts a peer that can be wrong with full confidence. Design for it.

    Read the essay

  2. Semantic Analysis

    What matters most is rarely the headline. It is the faint signals scattered across sources before it.

    Read the essay

  3. Self-Improving Loops

    A loop closes exactly where a check exists, so what you can hand to an agent is bounded by verification.

    Read the essay

Member of

NVIDIA Inception Program