Sekha: Universal AI Memory Controller¶

The Memory System That Never Forgets - Build AI That Remembers Everything

The Problem Sekha Solves¶

Every AI conversation today faces critical failures:

1. Most LLMs have 8k-128k token limits. One conversation can hit that.
2. ChatGPT, Claude, and others lose context after 30 days or fewer messages.
3. Agents need persistent memory to track multi-day workflows and learn from mistakes.
4. You re-explain your project, codebase, and preferences in every new session.
5. Your intellectual capital - research notes, decisions, insights - disappears.

The Solution¶

Sekha gives AI persistent, searchable, infinite memory - like a second brain that never forgets.

graph LR
    A[User/Agent] -->|Conversations| B[Sekha Controller]
    B -->|Stores| C[SQLite + Chroma]
    B -->|Via LLM Bridge| D[LiteLLM]
    D -->|Any LLM| E[OpenAI/Ollama/Claude/100+]
    C -->|Retrieves| B
    B -->|Context| A

    style B fill:#4051b5
    style C fill:#2d3748
    style D fill:#805ad5

Sekha sits between you and any LLM, capturing every interaction and intelligently retrieving relevant context when needed.

Key Features¶

♾️ Infinite Context - Conversations spanning millions of messages
🔍 Semantic Search - Find conversations by meaning, not just keywords
🧠 Smart Assembly - Auto-build perfect context from past interactions
📊 Hierarchical Summaries - Daily → Weekly → Monthly rollups
🏷️ Organization - Labels, folders, importance scoring
🔒 Sovereign - Self-hosted, local-first, your data never leaves
🔌 LLM Agnostic - Works with 100+ LLMs via LiteLLM
⚡ Production Ready - 80%+ test coverage, Docker deployment, sub-100ms queries

Use Cases¶

Quick Start¶

# Clone deployment repo
git clone https://github.com/sekha-ai/sekha-docker.git
cd sekha-docker/docker

# Start full stack
docker compose -f docker-compose.yml -f docker-compose.full.yml up -d

# Verify health
curl http://localhost:8080/health

Full Installation Guide Python Packages

Architecture¶

Sekha is built for production use with world-class engineering:

┌───────────────────────────────────────────────────────────┐
│  SEKHA CONTROLLER (Rust) - Memory Orchestration Engine    │
│  • REST API (19 endpoints)                                │
│  • MCP Server (7 tools for Claude Desktop)                │
│  • 4-Phase Context Assembly                               │
│  • SQLite (metadata) + ChromaDB (vectors)                 │
└────────────────────┬──────────────────────────────────────┘
                     │
                     ▼
┌───────────────────────────────────────────────────────────┐
│  LLM BRIDGE (Python) - REQUIRED                           │
│  • Embedding generation (nomic-embed-text)                │
│  • Summarization & entity extraction                      │
│  • LiteLLM gateway (100+ LLM providers)                   │
│  • Celery async task queue                                │
└────────────────────┬──────────────────────────────────────┘
                     │
          ┌──────────┼──────────┐
          ▼          ▼          ▼
     ┌────────┐ ┌──────────┐ ┌──────────┐
     │ Ollama │ │  OpenAI  │ │ Claude   │
     │ Local  │ │   GPT-4  │ │  Sonnet  │
     └────────┘ └──────────┘ └──────────┘
                 + 97 more LLM providers

┌───────────────────────────────────────────────────────────┐
│  PROXY (Python) - OPTIONAL                                │
│  • Transparent capture for generic LLM clients            │
│  • Auto-injects context from past conversations           │
│  • OpenAI-compatible API endpoint                         │
│  • Web UI dashboard                                       │
└───────────────────────────────────────────────────────────┘

Required Components:

1. Sekha Controller (Rust) - Core memory engine
2. LLM Bridge (Python) - Universal LLM adapter
3. ChromaDB - Vector similarity search
4. Redis - Async task queue broker

Optional Components:

• Proxy - For transparent capture
• Ollama - For local LLMs (or use cloud providers)

Architecture Deep Dive

Open Source & Sovereign¶

Dual License Model:

• AGPL-3.0 - Free forever for individuals, non-profits, academics, small businesses (<50 employees)
• Commercial License - Usage-based pricing for enterprises (contact for details)

Your data, your control:

✅ Self-hosted on your infrastructure
✅ No telemetry or phone-home
✅ Air-gapped deployment ready
✅ GDPR/HIPAA-compliant architecture
✅ Full data portability (export to JSON/Markdown)

License Details

Multi-Repository Ecosystem¶

Sekha is built as a modular system:

Now Available on PyPI:

• pip install sekha-llm-bridge - Universal LLM adapter
• pip install sekha-mcp - MCP protocol server for Claude Desktop

Python Installation Guide

Community & Support¶

Next Steps¶