Frequently Asked Questions¶

General Questions¶

What is Sekha?¶

Sekha is a universal memory system for AI that provides persistent, searchable, infinite context windows. It sits between your AI assistant and the LLM, capturing every conversation and intelligently retrieving relevant context when needed.

Think of it as giving AI a "second brain" that never forgets - whether your conversations span hours, days, or years.

How is Sekha different from ChatGPT's memory feature?¶

Is Sekha a chatbot or an LLM?¶

No. Sekha is not a chatbot or LLM. It's a memory protocol layer that:

• Sits BETWEEN your application and any LLM
• Captures and organizes conversations
• Retrieves relevant context intelligently
• Makes infinite context windows possible

You still use your preferred LLM (GPT-4, Claude, Llama, etc.) - Sekha just gives it perfect memory.

Do I need to be technical to use Sekha?¶

For basic use: No. If you can run Docker Compose, you can run Sekha.

For advanced use: Some technical knowledge helps for: - Custom integrations - Production deployments - Performance tuning

We provide: - One-command Docker deployment - Claude Desktop integration (no coding) - Python/JS SDKs for developers - REST API for any language

Installation & Setup¶

What are the system requirements?¶

Minimum: - 2 CPU cores - 4GB RAM - 10GB disk space - Docker 20.10+ OR Rust 1.83+

Recommended: - 4 CPU cores - 8GB RAM - 50GB SSD - Docker Compose with NVIDIA GPU (for local LLMs)

Runs on: - macOS (Intel & Apple Silicon) - Linux (x64 & ARM64) - Windows (WSL2 or native)

How do I install Sekha?¶

Fastest way (Docker):

curl -sSL https://raw.githubusercontent.com/sekha-ai/sekha-docker/main/docker-compose.yml -o docker-compose.yml
docker compose up -d

With Claude Desktop:

Add to Claude config, restart Claude, done.

See: Installation Guide

Can I run Sekha without Docker?¶

Yes! Install from source:

# Install Rust 1.83+
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

# Clone and build
git clone https://github.com/sekha-ai/sekha-controller
cd sekha-controller
cargo build --release

# Run
./target/release/sekha-controller

Binaries are also available from GitHub Releases.

Usage & Features¶

How do I store a conversation?¶

Via REST API:

curl -X POST http://localhost:8080/api/v1/conversations \
  -H "Authorization: Bearer your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "label": "Project Planning",
    "messages": [
      {"role": "user", "content": "Let'\''s plan the feature"},
      {"role": "assistant", "content": "Great! Here'\''s what I recommend..."}
    ]
  }'

Via Claude Desktop:

Just ask: "Remember this conversation as 'Project Planning'"

Via Python SDK:

from sekha import SekhaClient

client = SekhaClient(api_key="your-key")
client.store_conversation(
    label="Project Planning",
    messages=[...]
)

How do I search my memory?¶

Semantic search (finds similar meanings):

curl -X POST http://localhost:8080/api/v1/query \
  -H "Authorization: Bearer your-api-key" \
  -d '{"query": "API design discussions", "limit": 10}'

Full-text search (exact keywords):

curl -X POST http://localhost:8080/api/v1/search/fts \
  -H "Authorization: Bearer your-api-key" \
  -d '{"query": "OAuth authentication", "limit": 10}'

In Claude Desktop:

Ask: "What did we discuss about API design?"

How much data can Sekha store?¶

Practically unlimited. Storage scales with your disk space.

Tested benchmarks: - 1,000,000+ conversations - 50,000,000+ messages - Sub-100ms semantic queries on millions of vectors - Storage: ~50KB per 1000-word conversation

Example: 1 year of daily conversations (~365 conversations) = ~18MB

Can I organize conversations into folders?¶

Yes! Sekha supports hierarchical folder structures:

/work
  /projects
    /auth-feature
    /api-redesign
  /meetings
/personal
  /learning
  /ideas

Create folders when storing:

{
  "label": "Sprint Planning",
  "folder": "/work/projects/auth-feature",
  "messages": [...]
}

Move conversations later:

curl -X PUT http://localhost:8080/api/v1/conversations/{id}/label \
  -d '{"folder": "/work/archive/2026"}'

Integration Questions¶

Which LLMs does Sekha support?¶

Currently: - ✅ Ollama (Llama, Mistral, Phi, Qwen, etc.)

Roadmap (Q1 2026): - 🔜 OpenAI (GPT-3.5, GPT-4, GPT-4o) - 🔜 Anthropic (Claude 3, Claude 3.5) - 🔜 Google (Gemini) - 🔜 Cohere - 🔜 Custom API endpoints

Sekha is LLM-agnostic - you can switch between LLMs without losing your memory.

Does Sekha work with Claude Desktop?¶

Yes! Sekha provides native MCP (Model Context Protocol) support.

Setup:

1. Add Sekha MCP server to Claude config
2. Restart Claude Desktop
3. Ask Claude to use Sekha tools

Claude can then: - Store conversations in your memory - Search past discussions - Load relevant context automatically

See: Claude Desktop Integration

Can I use Sekha with my own application?¶

Absolutely! Sekha provides:

REST API (17 endpoints) - Language-agnostic - OpenAPI/Swagger documentation - Examples in Python, JavaScript, cURL

SDKs: - Python SDK - JavaScript SDK - (More coming: Go, Rust, Java)

MCP Protocol: - For any MCP-compatible AI tool - Claude Desktop, Cline, etc.

Can I integrate Sekha with VS Code?¶

Yes! The Sekha VS Code extension provides:

• In-editor memory access
• Code snippet storage
• Architecture decision records
• Project context assembly

Install: VS Code Marketplace

See: VS Code Integration

Privacy & Security¶

Where is my data stored?¶

100% local by default. Your data never leaves your infrastructure unless you explicitly configure remote deployment.

Storage locations: - Database: ~/.sekha/data/sekha.db (SQLite) - Vectors: ~/.sekha/chroma/ (ChromaDB) - Logs: ~/.sekha/logs/

You control: - Where data is stored - Who has access - Encryption at rest - Backup/export schedules

Is my data encrypted?¶

At rest: Not by default (local SQLite). You can enable filesystem encryption (LUKS, FileVault, BitLocker).

In transit: - HTTP by default (local deployments) - HTTPS for production (configure TLS)

Planned features: - Database encryption (SQLCipher) - E2E encryption for multi-user deployments - Hardware security module (HSM) support

Can I self-host Sekha completely air-gapped?¶

Yes! Sekha runs 100% offline:

1. Download binaries or build from source
2. Use local LLMs (Ollama with offline models)
3. No internet required after initial setup

Perfect for: - Government agencies - Healthcare (HIPAA compliance) - Financial services - Research institutions

Does Sekha phone home or send telemetry?¶

No. Sekha collects zero telemetry by default.

• No analytics
• No crash reports
• No usage tracking
• No version checks

Optional: You can enable error reporting for development builds (disabled in production releases).

Performance & Scaling¶

How fast is semantic search?¶

Benchmarks:

Search scales logarithmically thanks to ChromaDB's HNSW index.

Can Sekha handle millions of conversations?¶

Yes. Production testing shows:

• Storage: 50TB+ tested
• Conversations: 1M+ tested
• Messages: 50M+ tested
• Query latency: Sub-100ms even at 1M conversations

Scaling tips: - Use SSD storage (10x faster than HDD) - Allocate adequate RAM (4GB minimum, 8GB recommended) - Run ChromaDB on dedicated disk - Tune context_budget to reduce token processing

What if I run out of disk space?¶

Options:

1. Prune old conversations:

   curl -X POST http://localhost:8080/api/v1/prune/dry-run \
     -d '{"threshold_days": 90, "min_importance": 3}'
   

2. Archive to external storage:

   # Export to JSON
   curl -X POST http://localhost:8080/api/v1/export \
     -d '{"format": "json", "start_date": "2025-01-01", "end_date": "2025-12-31"}'
   
   # Move to S3/cold storage
   

3. Expand storage:

4. Add larger disk
5. Mount external volume
6. Use network storage (NFS, EBS)

Licensing & Commercial Use¶

Is Sekha free to use?¶

Yes, for: - Individuals - Students & academics - Non-profits - Small businesses (<50 employees) - Open-source projects

Under AGPL-3.0 license (free forever, no usage limits).

When do I need a commercial license?¶

You need a commercial license if:

• ❌ Your company has 50+ employees
• ❌ You're building a closed-source SaaS product
• ❌ You're an LLM provider (OpenAI, Anthropic, etc.)
• ❌ You want to avoid AGPL copyleft requirements

Pricing: - Startup: $5,000/year (51-200 employees) - Business: $25,000/year (201-1,000 employees) - Enterprise: $100,000/year (1,000+ employees)

See: License Documentation

Can I use Sekha for my SaaS product?¶

Yes, two ways:

1. Open-source SaaS (free):
2. Publish your modifications under AGPL-3.0

3. Example: MemoryCloud

4. Closed-source SaaS (paid):

5. Purchase a commercial license
6. No source code publication required
7. Support and legal indemnification included

What's the catch with AGPL-3.0?¶

No catch. You can use Sekha freely if:

• ✅ You use it internally (no distribution = no requirements)
• ✅ You publish modifications if you distribute
• ✅ "Network use" counts as distribution (if users access your service)

Example scenarios:

Internal use (FREE): - Your 30-person startup uses Sekha for internal tools ✅ - You modify the code for your needs ✅ - No one outside your company uses it ✅ - No requirement to publish modifications ✅

SaaS product (requires open-source or license): - You build a product where customers access Sekha ⚠️ - You must either: - Publish modifications (AGPL-3.0) OR - Buy commercial license

Troubleshooting¶

Sekha won't start - connection refused¶

Check:

# Is Sekha running?
curl http://localhost:8080/health

# Check logs
docker compose logs sekha-controller

# Or for binary
tail -f ~/.sekha/logs/sekha.log

Common causes: - Port 8080 already in use - Config file errors - Missing dependencies

See: Common Issues

Claude Desktop can't see Sekha tools¶

Checklist:

1. ✅ Docker is running: docker ps
2. ✅ Sekha is running: curl http://localhost:8080/health
3. ✅ Config file is valid JSON
4. ✅ API key matches between config and Claude
5. ✅ Restarted Claude Desktop

Test MCP server:

docker run -i --rm \
  -e SEKHA_API_URL=http://localhost:8080 \
  -e SEKHA_API_KEY=your-key \
  ghcr.io/sekha-ai/sekha-mcp:latest

Semantic search returns irrelevant results¶

Possible causes:

1. Small dataset: Need 10+ conversations for good embeddings
2. Query too vague: Be more specific
3. Wrong embedding model: Switch models in config

Improve results:

[embeddings]
model = "all-MiniLM-L6-v2"  # Better for short queries
# OR
model = "all-mpnet-base-v2"  # Better for long documents

Performance is slow¶

Quick wins:

1. Reduce context budget:

   [context]
   default_budget = 2000  # Down from 4000
   

2. Limit search results:

   curl -X POST .../query -d '{"query": "...", "limit": 5}'  # Down from 10
   

3. Index database:

   sqlite3 ~/.sekha/data/sekha.db "VACUUM;"
   

4. Use SSD storage (10x faster)

Contributing & Support¶

How can I contribute to Sekha?¶

We welcome contributions!

Ways to help: - 🐛 Report bugs - 💡 Suggest features - 📝 Improve documentation - 🧪 Add tests - 🔧 Fix issues - 🌍 Translate docs

See: Contributing Guide

Where can I get help?¶

Community support (free): - Discord: discord.gg/sekha - GitHub Discussions: github.com/sekha-ai/sekha-controller/discussions - GitHub Issues: github.com/sekha-ai/sekha-controller/issues

Commercial support: - Enterprise SLA: [email protected] - General inquiries: [email protected]

See: Support & Contact

Can I hire the team for custom development?¶

Yes! We offer:

• Custom feature development
• Integration assistance
• Architecture consulting
• Training sessions
• Deployment support

Contact: [email protected]

Still have questions?¶

• 📖 Check our full documentation
• 💬 Ask on Discord
• 🐛 Report issues on GitHub
• 📧 Email us: [email protected]

Last updated: January 2026