# How Zild Works

Zild turns an incoming event into a governed operational decision: respond, use a tool, create work, or hand off to a person.

Source: https://zild.ai/en-US/docs/getting-started/how-it-works

Zild turns an incoming event into a governed operational decision: respond, use a tool, create work, or hand off to a person.

## Execution flow

- A channel receives a message, call, file, or system event.
- Zild identifies the tenant, app, agent, user, and available context.
- The agent applies instructions, knowledge, permissions, and guardrails.
- The platform decides whether to answer, call a tool, create a task, or escalate.
- The result is recorded for monitoring, quality review, and improvement.

## Platform layers



<table class="api-params"><thead><tr><th>Layer</th><th>Purpose</th></tr></thead><tbody><tr><td>Context</td><td>Conversation history, CRM data, documents, transcripts, and approved knowledge.</td></tr><tr><td>Decision</td><td>Agent reasoning limited by instructions, confidence, policy, and permissions.</td></tr><tr><td>Action</td><td>Messages, tool calls, tasks, transfers, CRM updates, or workflow triggers.</td></tr><tr><td>Evidence</td><td>Conversation records, transcripts, evaluations, metrics, and audit history.</td></tr></tbody></table>



## Operational control

The important rule is simple: the AI should act only when it has enough context, evidence, permission, and instruction. Otherwise it should ask for more information, limit the answer, or transfer to a human.