UAI

AI Agents

How artificial intelligence works in LTM

What are AI agents in LTM

In LTM, AI is not "another chat". It is agents, each responsible for its own task. In the basic setup — the meaning maker, the answerer, and Dik (the central agent). But this is only the beginning: in each project, you can add any agents.

Meaning maker

Reads project messages, creates meanings, tags, crosslinks, associative connections. Gives meanings to the shared knowledge base. Does not answer the user — only creates meanings.

Answerer

Answers the user in the project. Uses meanings (of its own project, of other projects — by access rules). Searches for non-standard solutions through associative connections.

Dik — central agent

Manages the system, coordinates meaning makers and answerers, answers the user very rarely (only critical questions).

Flexibility of agents — the main thing

In each project, the creator or administrator can add any agents. Not only the meaning maker and answerer, but also, for example, G5, a designer, a draftsman, a letter writer — and others. For the project's tasks.

All agents access the shared knowledge base of meanings, but work in the context of their project. Connections between agents (how they form answers, how they interact) are configured in the project settings.

Analysis depth of each agent is configured by the administrator in the project — from basic (top-10 meanings) to deep (2 moves ahead, 10 moves "grandmaster").

And even in each branch you can add your own agents. If a draftsman is not needed in the project, but is needed in a branch, you can add them specifically to the branch. The system is configured not only at the project level, but also at the level of each branch — for a specific task.

How the meaning maker works

Reads only new messages of its project (batch every 5–10 minutes). Does not re-read everything — only what appeared since last time. This saves tokens and time.

From messages it creates:

  • Compressed meaning (6 minutes of speech → 4 sentences).
  • Tags (keywords).
  • Crosslinks (connections with other projects).
  • Associative connections (between different fields of knowledge — e.g., biology ↔ molecular physics).

All this goes to the shared vector knowledge base. The meaning maker does not answer the user — it only creates meanings.

How the answerer works

Answers the user in the project. Does not read all messages — does a semantic search (top-10 meanings) and goes through associative connections.

If there are not enough meanings of its own project — looks only into those projects with which there are crosslinks (similar projects). If there are no connections — does not look (no point).

The answerer goes through associative connections of different fields of knowledge — to find non-standard solutions (at the intersection of sciences).

Access to meanings

The answerer uses meanings by access rules:

  • Own project — full access (meaning + quote + source).
  • Public others — meaning + quote + source.
  • Private / protected others — only meaning (without quote, without source).
  • Secret others — only meaning (without quote, without source, without information about the project).

AI agent modes

  • Free minimum — semantic search top-10 meanings (basic answer).
  • Paid modes:
    • 2 moves ahead — deeper analysis (token and machine time costs).
    • 10 moves "grandmaster" — maximum deep analysis (user funds).

The user chooses the mode themselves (free or paid).

Token economy

  • Meaning maker reads only new messages (incrementally), not everything again.
  • Answerer does a semantic search (top-10 meanings) + goes through associative connections.
  • Answerer looks only into projects with crosslinks (not into all) — saves tokens and time.
  • Dik turns on rarely (only critical events).