Core Types

The core module provides pure data types that are shared across all layers of Aleph. It has no business logic, no platform-specific code, and only one external dependency: serde.

Design Philosophy

The core module follows the principle of high cohesion and zero coupling. Every type in this module is a pure data container with derived implementations for Debug, Clone, Serialize, and Deserialize. There are no methods beyond simple field access — business logic lives in the modules that consume these types.

This design ensures that:

• Types can be shared between the server, desktop bridge, and client interfaces without pulling in heavy dependencies
• Serialization contracts are stable and versioned via serde
• The module compiles quickly and never blocks other modules

MediaAttachment

Multimodal content support for images, documents, and video from clipboard or file uploads.

pub struct MediaAttachment {
    pub media_type: String,       // "image", "document", "video", "file"
    pub mime_type: String,        // "image/png", "application/pdf"
    pub data: String,             // Content (base64 or utf8)
    pub encoding: String,         // "base64" | "utf8"
    pub filename: Option<String>, // Original filename
    pub size_bytes: u64,          // Original size
}

The encoding field determines how to interpret data:

• "base64" — Binary content encoded as Base64 (images, PDFs)
• "utf8" — Plain text content (markdown, extracted text)

Usage: Used by CapturedContext to pass clipboard content from the Desktop Bridge to the Thinker, and by tool outputs that generate or receive files.

CapturedContext

Context from the user's active application, bridged from the native desktop layer.

pub struct CapturedContext {
    pub window_title: Option<String>,
    pub attachments: Option<Vec<MediaAttachment>>,
    pub session_id: Option<String>,
}

Fields:

• window_title — Title of the active window (e.g., "Notes.app", "VS Code")
• attachments — Clipboard content (images, files) captured at the moment of interaction
• session_id — Session identifier for multi-turn conversations

Usage: Injected into the prompt context when the user interacts via the Desktop Bridge, giving the AI awareness of what the user is currently working on.

CompressionStats

Statistics about the memory compression state, used by the Settings UI to display memory health.

pub struct CompressionStats {
    pub total_raw_memories: u64,
    pub total_facts: u64,
    pub valid_facts: u64,
    pub facts_by_type: HashMap<String, u64>,
}

Fields:

• total_raw_memories — Layer 1: raw user messages stored before compression
• total_facts — Layer 2: compressed facts extracted from memories
• valid_facts — Non-invalidated facts still considered active
• facts_by_type — Breakdown by type (preference, plan, learning, etc.)

Usage: Queried by the Gateway's config handlers to populate the memory dashboard in the Panel UI.

MemoryEntry

A memory record returned by the Memory system's API, suitable for serialization to clients.

pub struct MemoryEntry {
    pub id: String,
    pub window_title: String,
    pub user_input: String,
    pub ai_output: String,
    pub timestamp: i64,
    pub similarity_score: Option<f32>,
}

Usage: Returned by memory search queries. The similarity_score field is populated when the entry is retrieved via vector search, indicating relevance to the query.

Where These Types Are Used

Code Location

• src/core/mod.rs — Module entry point, re-exports
• src/core/types.rs — Type definitions

See Also

• Memory System — How memories are stored and compressed
• Desktop Bridge — How CapturedContext is captured from native apps
• Thinker — How context is injected into prompts