Did you know?

Short stories, numbers and terms from the world of AI agents. Five minutes of reading that make the rest of the site easier to follow.

Neimar in numbers

Everything here is already on the platform.

11

agent types

Chat, voice, vision, workflow, email, analyst, security, portal, document, website and the Orchestrator.

5

trust levels

From L0 Observe to L4 Autonomous. An agent earns more freedom over time.

40+

prompt injection patterns

Known attack patterns the security layer recognizes in incoming messages.

1

script tag

That is all it takes to put the chat widget on any website.

4

messaging channels

WhatsApp, Telegram, Viber and SMS, on top of web chat, email and phone.

20+

languages

Agents detect the language of the conversation or stick to the one you choose.

A short history of talking machines

The idea of a machine you can talk to is older than the computer.

  1. 1920

    The word robot is born

    Czech writer Karel Čapek used it in his play R.U.R. The word was suggested by his brother Josef, from robota, meaning forced labor.

  2. 1950

    The imitation game

    Alan Turing proposed a test: if you cannot tell in conversation whether you are talking to a person or a machine, the machine passes. We still call it the Turing test.

  3. 1956

    Artificial intelligence gets its name

    John McCarthy coined the term for a summer workshop at Dartmouth College, where the field was born.

  4. 1966

    ELIZA, the first chatbot

    Joseph Weizenbaum's program at MIT imitated a psychotherapist by rephrasing what you typed. Many users confided in it, to the author's surprise.

  5. 1997

    A computer beats the world chess champion

    IBM's Deep Blue won a six game match against Garry Kasparov.

  6. 2011

    Machines answer in plain language

    IBM Watson won the quiz show Jeopardy! against two of its best champions. The same year Siri shipped on the iPhone.

  7. 2017

    The Transformer

    A Google research paper titled Attention Is All You Need introduced the architecture behind every modern language model, including the ones Neimar agents run on.

  8. 2022

    Everyone talks to an AI

    ChatGPT opened to the public in November and brought conversational AI into everyday life.

Models read tokens, not words

A language model splits text into tokens: short pieces of words. In English a token is roughly four characters, so a hundred tokens is about seventy five words. Languages with fewer training texts, such as Serbian, usually need more tokens per word. That is why plans count tokens and not messages.

Neimar builds agents that talk to your customers.

Illustration. Real tokenizers split text differently from model to model.

Small things inside Neimar

Details you notice once you start building.

One conversation builds the whole system

Describe your business to the Orchestrator and it creates the agents, workflows, CRM and knowledge bases for you.

One knowledge base, every channel

Agents in a project share knowledge. Change an answer in one place and web chat, phone and email all give the new one.

Conversation in, data out

A conversation can end in structured data. An order taken in chat arrives at your system as JSON, ready for the workflow.

Agents earn trust

Every agent starts at a lower trust level and is promoted as it proves itself. Acting on its own is a privilege, not a default.

Addresses the way people say them

The voice agent handles descriptions like 'across from the bakery, second entrance' thanks to a knowledge base of local landmarks.

Fits around what you already run

Workflows receive webhooks and call external APIs, so Neimar wraps around the software you already use.

A small glossary

Terms you will meet in the documentation.

Agent
A program that holds a goal, reads context and decides what to do next, within limits you set.
LLM
Large language model. The engine that reads and writes text behind every agent.
Token
The unit a model reads and writes. Roughly a short word or part of a longer one.
Prompt
The instructions and context given to a model. An agent's personality lives in its prompt.
Knowledge base
Your documents, searched at answer time so the agent answers from your facts, not from memory.
Hallucination
A confident answer that is not true. Knowledge bases and trust levels exist to keep it rare.
Prompt injection
Text in a message or document that tries to override the agent's instructions. Neimar screens for known patterns.
Workflow
A chain of triggers and actions: when this happens, do that, then call that system.
Webhook
An HTTP call one system makes to another when something happens.
Trust level
How much an agent may do on its own, from observing (L0) to fully autonomous (L4).

See it in action

Describe your business and we will show you what the agents can do with it.

Talk to us