Services
This section documents DIAL services and deployment charts: what each is for, how it is configured, and how it is deployed. It is for platform engineers who deploy, configure, and upgrade DIAL. It assumes you know how the services fit together; see Architecture highlights.
Services in this section
| Service | Purpose |
|---|---|
| DIAL Core | The API gateway: the Unified API and the other supported AI APIs, access control, usage limits, load balancing, request processing, and resource storage |
| DIAL Chat | The web interface for users, which can also be embedded in other applications through Chat Overlay |
| DIAL Chat Themes | The static service that holds the color themes, logos, and icons of DIAL Chat |
| DIAL Model Provider | Runs the OpenAI, Vertex AI, Bedrock, and DIAL-to-DIAL adapters in one service, each under its own /compat/<adapter> path; replaces the standalone adapters |
| DIAL OpenAI Adapter | Connects models on Azure OpenAI, the OpenAI Platform, Azure AI Foundry, Bedrock OpenAI models, Mistral, Alibaba Cloud, vLLM, and Anthropic Messages providers. Will be deprecated; use DIAL Model Provider |
| DIAL Bedrock Adapter | Connects the chat, image, and embedding models of Amazon Bedrock, and Claude through the Anthropic API. Will be deprecated; use DIAL Model Provider |
| DIAL Vertex AI Adapter | Connects Gemini, Claude, Imagen, Veo, Mistral, and the embedding models of Google Cloud Vertex AI. Will be deprecated; use DIAL Model Provider |
| DIAL Quick Apps | Runs Quick Apps, applications composed from tools and an LLM orchestrator, and their visual editor; includes ChatHub |
| DIAL Admin | The administrator web interface, connected directly to Core, with optional deployment management, evaluation, and analytics services |
| DIAL Evaluations | Evaluation Framework Backend and Evaluation Metrics, which run DIAL Admin's test suites against models, applications, and MCP tools, and score the answers |
| DIAL Deployment Manager | Builds images and operates Kubernetes workloads for DIAL Admin, including MCP servers and model servings; includes the stdio MCP proxy images |
| DIAL Helm | Kubernetes deployment charts for the platform assembly, Core, container-based extensions, and Admin |
| DIAL OpenAPI to MCP | An application type that turns a REST API's OpenAPI 3.x document into MCP tools |
| DIAL RAG | An application that answers questions about the documents attached to a request, indexing them on first use |
| DIAL Generic RAG | An application type that answers questions over documents uploaded in advance, with a document screen in DIAL Admin |
| DIAL App Controller | Builds Code App images, deploys them as Knative services, and manages interpreter session pods for Core |
| DIAL App Builder Python | Build-time helper that downloads and validates Code App sources and prepares a Dockerfile for the image builder |
| DIAL Code Interpreter | Executes Python in Jupyter sessions for calculations, file analysis, and visualizations, including Quick Apps tools |
| DIAL Deep Research | Plans research over Generic RAG documents or StatGPT datasets, reviews findings, and produces reports with citations |
| DIAL Analytics | Ingests usage logs into ClickHouse and serves governed queries, tables, dashboards, and aggregate pipelines |
| DIAL Analytics Enrichment Runner | Executes SQL and LLM enrichment pipelines defined in Analytics, with persistent scheduling, retries, and spend tracking |
| DIAL Log Parser | Converts stored Core request logs into deployment/date-partitioned Parquet datasets for offline analysis and evaluation |
| DIAL Scheduler | Runs model and HTTP targets on recurring or one-time schedules with per-user delegated credentials and persistent run history |
Configuration for the other services is still in Operating DIAL > Configuration and moves into this section service by service.
What each service covers
- Overview — what the service does, what it depends on, and how it is deployed
- Configuration — every setting, with its default, grouped by the entity or subsystem it controls
- Upstream links — the service repository, GitHub releases, and published image tags or charts, linked from its overview