Quick Apps capabilities
This section explains what the Quick Apps Backend does with a Quick App at run time, capability by capability. Each page describes the behavior, the manifest fields that turn it on, and the environment variables that limit it. It is for platform engineers and app developers who need to know what a Quick App can do and how it behaves. It assumes you know the Quick Apps services. All pages describe Quick Apps Backend 0.13.2.
Capability map
| Capability | What it does | Page |
|---|---|---|
| Agent loop | Alternates LLM calls with tool calls until the model answers, within an iteration limit | Orchestration |
| Orchestrator model and system prompt | Any DIAL model drives the app; the system prompt is inline text, a DIAL prompt file, or a predefined template | Orchestration |
| Tool sets | REST APIs, DIAL deployments, DIAL applications, MCP servers, DIAL MCP Toolsets, and internal tools | Tools |
| Parallel tool calls and stages | Runs the requested tools concurrently and shows each call as a stage in DIAL Chat | Tools |
| Fallbacks and timeouts | Decides what the model hears when a tool fails or runs too long | Tools |
| Dynamic tool discovery | Leaves large Toolsets out of the first LLM request and finds tools on demand | Tools |
| Tool access filter | Hides DIAL tools the calling user can't access | Tools |
| Subagents and citations | Resumable conversations with DIAL deployments called as tools, and their citations carried into the answer | Tools |
| MCP features | One session per request, interactive sign-in, and MCP resources | Tools |
| Contexts | Inline text, DIAL files, and DIAL folders the agent always knows about | Contexts and files |
| Attachments and file parameters | Lets the orchestrator read attachments, and passes files to tools in the format each tool needs | Contexts and files |
| External URLs and web fetch | Fetches non-DIAL URLs within an administrator's limits | Contexts and files |
| DIAL files tools | Lets the agent list, read, search, and write files in its DIAL workspace | Contexts and files |
| Skills | Reusable instructions the agent reads on demand: predefined, DIAL prompts, DIAL skill resources, and skills picked on a message | Skills |
| Hooks | Calls a tool at the start or end of every turn, without the model deciding to | Hooks |
| Time awareness | Tells the agent the current time, and when each tool result was produced | Time awareness |
| Errors | Turns failures into safe messages with a reference an administrator can find in the logs | Errors |
Stable and Preview
Every capability is either Stable or Preview.
- Stable features follow semantic versioning: a breaking change needs a major version bump.
- Preview features may change in breaking ways without one. They work only when the backend's
ENABLE_PREVIEW_FEATURESistrue; otherwise their settings are ignored.
In 0.13.2, one capability is in Preview: tool-call result offload. Everything else on these pages is Stable. Backend 0.13.2 moved ten features from Preview to Stable; see Backend 0.13.2.