---
title: Perception
description: Define how an agent sees — attention filters and contextual focus.
---

**Perception** defines how an agent perceives and filters information. It controls what the agent pays attention to and how much detail it processes.

## Structure

```go
type Model struct {
    AttentionFilters  []AttentionFilter
    ContextWindow     float64 // 0.0 = minimal context, 1.0 = full context
    DetailOrientation float64 // 0.0 = high-level, 1.0 = fine-grained
}
```

## Attention filters

Attention filters direct the agent's focus to specific aspects of input:

```go
type AttentionFilter struct {
    Name     string   // filter name
    Keywords []string // keywords to watch for
    Patterns []string // regex patterns to match
    Prompt   string   // prompt fragment when filter activates
}
```

### Example: security-focused perception

```go
perception.Model{
    AttentionFilters: []perception.AttentionFilter{
        {
            Name:     "security",
            Keywords: []string{"password", "token", "auth", "injection", "XSS"},
            Prompt:   "Pay special attention to security implications in this context.",
        },
        {
            Name:     "performance",
            Keywords: []string{"latency", "throughput", "memory", "CPU", "bottleneck"},
            Prompt:   "Note any performance concerns.",
        },
    },
    ContextWindow:     0.8,
    DetailOrientation: 0.9,
}
```

## Fields

| Field | Type | Range | Description |
|-------|------|-------|-------------|
| `AttentionFilters` | `[]AttentionFilter` | — | Keyword and pattern-based focus directives |
| `ContextWindow` | `float64` | 0.0–1.0 | How much surrounding context to consider |
| `DetailOrientation` | `float64` | 0.0–1.0 | High-level overview vs fine-grained analysis |

## Embedding in a persona

Perception is a value object embedded directly in the persona:

```go
persona := &persona.Persona{
    Perception: perception.Model{
        ContextWindow:     0.8,
        DetailOrientation: 0.7,
    },
}
```