Capabilities
This section explains what DIAL provides, one capability per page. Each page covers the problem the capability solves, what DIAL offers for it, and what to check before you rely on it, then links to the architecture, configuration, and how-to pages. It is for architects and evaluators who have read What is DIAL.
In this section
Reach models
- Model access and AI APIs — one governed entry point to models from many providers, through the Unified API, OpenAI Responses, or Anthropic Messages, with aliases and routers in front of them.
- Multimodality — text, images, audio, and video in one conversation or application, carried as attachments.
- Load balancing and failover — spread a model's traffic across endpoints and move it elsewhere when an endpoint fails.
Protect and observe
- Guardrails — interceptors that inspect and change requests and responses for personal data, content rules, and audit.
- Analytics and observability — how AI is used and what it costs, and whether the platform is healthy.
Connect and build
- Tools and enterprise data — MCP Toolsets, REST APIs, custom APIs through routes, and Generic RAG.
- Agentic platform — models and applications as agents that compose into multi-agent applications and run on a schedule.
- Coding agents and harnesses — govern the coding agents developers already run.
Improve and share
- Evaluations — measure quality on your own test cases and catch regressions after a change.
- Collaboration and sharing — share resources with specific people or publish them to the organization after review.
To find a capability by the need it meets, see the table in What DIAL provides.