Skip to content

Architecture

Understand how Sekha works under the hood.

System Components

Sekha consists of three main components:

1. Controller (Rust)

The core memory engine handling all operations:

  • REST API server (Axum)
  • Database layer (SeaORM + SQLite)
  • Memory orchestration logic
  • Context assembly algorithms
  • MCP protocol server

Learn more →

2. LLM Bridge (Python)

Handles all LLM-related operations:

  • Embedding generation (Ollama/OpenAI/Anthropic)
  • Conversation summarization
  • Label suggestions
  • Async queue processing

Learn more →

3. Storage Layer

Dual storage approach:

  • SQLite - Structured data (conversations, metadata)
  • ChromaDB - Vector embeddings for semantic search

Learn more →

Key Concepts

Memory Orchestration

How Sekha intelligently manages and retrieves memories:

Memory Orchestration →

  • Semantic search with embeddings
  • Context budgeting and prioritization
  • Importance scoring
  • Folder-based organization

Data Flow

End-to-end flow of data through the system:

  1. Store → Messages saved to SQLite
  2. Embed → LLM Bridge generates embeddings
  3. Index → ChromaDB stores vectors
  4. Query → Semantic search retrieves relevant context
  5. Assemble → Optimal context built for LLM

Architecture Diagrams

High-Level Overview

graph TB
    Client[Client Application] --> API[REST API / MCP Server]
    API --> Controller[Controller Core]
    Controller --> DB[(SQLite)]
    Controller --> Vector[(ChromaDB)]
    Controller --> Bridge[LLM Bridge]
    Bridge --> Ollama[Ollama]

Request Flow

sequenceDiagram
    participant C as Client
    participant A as API
    participant Ctrl as Controller
    participant DB as SQLite
    participant Vec as ChromaDB
    participant LLM as LLM Bridge

    C->>A: POST /conversations
    A->>Ctrl: Create conversation
    Ctrl->>DB: Store messages
    Ctrl->>LLM: Generate embeddings
    LLM->>Vec: Store vectors
    Vec-->>C: 201 Created

Next Steps

Built with MkDocs and Material for MkDocs