Documentation for Kamiwaza 0.13.0

This documentation outlines the key components and systems of the Kamiwaza platform, which is no longer actively maintained. For the current GA release, see 1.0.1.

Model & Data Handling

Vector Databases

• What it is: A specialized database for storing and retrieving high-dimensional vector data, such as embeddings generated from text or images. Kamiwaza can integrate with vector backends such as Milvus and Qdrant.
• Why it matters: Vector databases are the engine behind similarity search, Retrieval-Augmented Generation (RAG), and other semantic retrieval workflows. Kamiwaza's abstraction layer lets deployments adopt supported backends without changing user-facing workflows.

Embeddings Management

• What it is: The process of generating, storing, and managing vector embeddings for your data. Kamiwaza provides platform services that automate embedding generation using a dedicated, standalone OpenAI-compatible llama.cpp embedding service deployed via Helm (replacing legacy in-process fallbacks).
• Why it matters: Consistent and efficient embedding management is fundamental to vector search quality and retrieval performance. A dedicated embedding service reduces container footprint and improves service decoupling compared to in-process fallbacks.

Data Catalog

• What it is: A centralized inventory of data assets, containers, and secret metadata. Kamiwaza integrates with DataHub-backed catalog workflows and uses URN-based references across the platform.
• Why it matters: As AI systems grow, so does the data they consume. A data catalog improves discoverability, governance, and access control across models, retrieval flows, connectors, applications, and tools.

Orchestration & Serving

Orchestration Engine

• What it is: The control and execution layer that manages model launches, runtime coordination, and distributed work. Kamiwaza uses Ray and related serving runtimes to handle that workload.
• Why it matters: This layer keeps inference, background work, and runtime routing coordinated across cluster resources.

Model Serving

• What it is: The process of taking a model and making it available for real-time inference via a managed API route. Kamiwaza supports multiple serving engines such as vLLM, llama.cpp, and MLX, depending on environment and model type.
• Why it matters: Different models have different hardware and runtime requirements. Multi-engine support helps the platform balance compatibility, performance, and cost.

API Gateway

• What it is: A single, unified entry point for platform APIs. Kamiwaza uses FastAPI plus ingress routing to expose those services through a consistent customer-facing interface.
• Why it matters: A gateway simplifies development by providing one access surface for models, retrieval, apps, tools, logging, and administration. It is also the natural place to enforce authentication, rate limiting, and logging.

Security & Operations

Identity & Access Management

• What it is: The system that handles user authentication (who you are) and authorization (what you're allowed to do).
• Why it matters: Robust security is non-negotiable in an enterprise setting. Kamiwaza's IAM services ensure that only authorized users and applications can access sensitive data and models.

Caching

• What it is: A high-speed storage layer used for session state, temporary runtime coordination, and frequently accessed data.
• Why it matters: Caching dramatically improves the performance and responsiveness of AI applications, especially those with high request volumes, leading to a better user experience and lower operational costs.