openray/eko repository overview

⁠🛡️ Ekō

Open-source, context-aware security proxy that prevents leaking sensitive data to AI services — with first-class detectors for African regulated domains.

Ekō sits between your application and an AI provider, detects sensitive data in prompts, and either redacts or tokenizes it before it leaves your environment.

🎮 Try the live playground →⁠  ·  📦 Source & full docs on GitHub →⁠

⁠🚀 Quick start

docker run --rm -p 8080:8080 openray/eko:main-latest

Verify it is up:

curl http://localhost:8080/health
curl http://localhost:8080/ready

Sanitize a prompt:

curl -X POST http://localhost:8080/v1/sanitize \
  -H "Content-Type: application/json" \
  -d '{"prompt":"My BVN is 12345678901 and email is [email protected]"}'

⁠🏷️ Supported tags & architectures

Architectures: multi-arch manifest — linux/amd64 and linux/arm64 (runs natively on Apple Silicon / ARM servers and x86).

⁠🔌 OpenAI-compatible proxy

Point the OpenAI SDK at Ekō — only the base URL changes:

export OPENAI_BASE_URL="http://localhost:8080/v1"

Routes: POST /v1/chat/completions, POST /v1/responses, plus the core POST /v1/sanitize. Ekō forwards the sanitized request to the configured upstream.

⁠🎨 Run with your own configuration

docker run -d \
  -p 8080:8080 \
  -v $(pwd)/configs/config.yaml:/app/configs/config.yaml:ro \
  -v $(pwd)/patterns:/app/patterns \
  openray/eko:main-latest

Start from configs/config.example.yaml in the repo. For a Redis-backed token vault and the optional contextual SLM sidecar, use Docker Compose — see the GitHub README⁠.

⁠🌍 What gets detected

• Credentials & secrets — API keys (OpenAI, Anthropic, Google, AWS, Azure), database connection strings, JWT/OAuth, SSH keys, env-style secrets (API_KEY=, PASSWORD=)
• PII — Nigerian (BVN, NIN, NUBAN, +234), Kenyan (M-Pesa, ID, +254), South African (ID, +27), Ghanaian (Mobile Money, +233), plus emails, IBAN, SWIFT/BIC
• Financial — credit card numbers, bank accounts, transaction references
• Custom business patterns — your own regex rules under patterns/custom

Optional contextual detection of person names and addresses via an opt-in Small Language Model sidecar.

⁠📊 Monitoring

Prometheus-compatible metrics at GET /metrics; liveness and readiness at /health and /ready. A container HEALTHCHECK is built in.

⁠🔗 Links

• 📦 GitHub: https://github.com/Openray-ai/eko⁠
• 🎮 Playground: https://ek-playground.openray.workers.dev/⁠
• 🌍 OpenRay: https://openray.ai⁠
• 📜 License: MIT

_{Ekō (Yoruba): "to guard, to protect" — built with ❤️ in Africa, for the world.}