> ## Documentation Index
> Fetch the complete documentation index at: https://docs.svantic.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Mental Model

> How Svantic works — in 30 seconds.

# Mental Model

**Your agents keep their tools. Svantic handles the hard parts.**

## The 30-Second Version

You have services that do things. Svantic makes them work together intelligently.

<Columns cols={2}>
  <Card title="You Own" icon="user">
    * Your code
    * Your credentials
    * Your data
    * Your tools
  </Card>

  <Card title="Svantic Handles" icon="cpu">
    * Routing — which agent should handle this?
    * Planning — what steps, in what order?
    * Safety — guardrails, approvals, audit trails
    * Memory — learning what works over time
  </Card>
</Columns>

## How It Works

<img src="https://mintcdn.com/svantic/DQEVJ_RnVW5_l6gu/images/mental-model.svg?fit=max&auto=format&n=DQEVJ_RnVW5_l6gu&q=85&s=89e220a8d70ddd4df5f0e87649e0a91e" alt="Svantic Mental Model" width="960" height="520" data-path="images/mental-model.svg" />

<Steps>
  <Step title="Task Arrives" icon="inbox">
    A user sends a message, an API call comes in, a webhook fires, or a cron job triggers. This is the entry point — someone or something needs work done.
  </Step>

  <Step title="Svantic Orchestrates" icon="brain">
    The mesh receives the task and figures out what to do. It looks at registered agents, their capabilities, and the context of the request. It plans the sequence — which agents to call, in what order, with what data. You don't write routing logic. The mesh handles it.
  </Step>

  <Step title="Agents Execute" icon="play">
    Your agents run on your infrastructure. They receive capability calls, execute their handlers, and return results. They keep their tools, credentials, and data. Svantic only sees the capability names and responses.
  </Step>

  <Step title="Results Return" icon="check">
    Agents send results back to the mesh. If the task required multiple agents, results flow through the orchestration plan — one agent's output becomes another's input. The final result returns to the caller.
  </Step>

  <Step title="Knowledge Compounds" icon="trending-up">
    Every execution teaches the system. What worked? What failed? What patterns emerge? This knowledge feeds back into future orchestration — making the next task faster, cheaper, and more reliable.
  </Step>
</Steps>

## What Runs Where

| Runs on **your** infrastructure | Runs on **Svantic**      |
| ------------------------------- | ------------------------ |
| Your agents                     | Routing & discovery      |
| Your tools & handlers           | Planning & orchestration |
| Your credentials & data         | Safety & guardrails      |
|                                 | Memory & learning        |
|                                 | Telemetry & traces       |

Your code never leaves your environment. Svantic sees capabilities and results — not your raw data.

## That's It

Build agents. Register them. Let Svantic orchestrate. Knowledge compounds.

Everything else is details.

**[When to use Svantic →](when-to-use-svantic)**