Overview
AI-awareDRYL.Components.Agents is a separate, experimental companion package
(0.6.0) that bridges the
Microsoft Agent Framework
to DRYL's AI vocabulary. You bring your own AIAgent; DRYL wires the run to the UI —
automatic AiState, structured streaming, a live tool-call trace, interactive
canvas artifacts and ready-made human-in-the-loop dialogs. The core library stays
dependency-free; the LLM SDK lives only here.
AIAgent to the very same calls. Tool calls and dialogs
are real: the Agent Framework invokes them for you, so the DRYL dialogs pop up live.
Install
A NuGet-only companion package, registered alongside the core.
dotnet add package DRYL.Components.Agents// Program.cs — alongside the core registration
builder.Services.AddDrylComponents().AddDrylAgents();
// One provider in your root layout (required by the tool dialogs):
// <DrylDialogProvider />Explore the package
Each capability has its own page — with what it is for, how it works, where the innovation lies, and a live demo.
Agent Run
Wrap a run: automatic AiState + a live tool-call trace, zero manual state.
Structured Generation
DrylAiGenerate<T> — JSON streams straight into live Blazor components.
AI Field
DrylAiField — make any input AI-capable in one line, without touching it.
Display Tools
Hand the agent charts, stats & timelines — it answers with live components.
Human-in-the-Loop
Ready-made dialog AIFunctions — the agent asks the user and waits.
AI Canvas
DrylAiCanvas — the AI answers with an interactive artifact bound to your data and commands, and the user can take it apart by hand.
Canvas Dock
Floating prompt dock — one input, one line of live status, a context chip for the selected element, the transcript on demand, Actions and Suggestions slots for the host's own controls and prompt chips, and a microphone that turns the dock into a voice panel.
Voice
Speak to your agent — a WebRTC realtime session with the same tools, the same conversation and no audio through your server.
Multi-Agent Handoff
Chain agents into one run — a living handoff timeline with a baton.
Collaborative Build
StartBuild — agent and user refine a typed artifact round by round.
Run Health
Error surfacing with retry + token-usage badges from the stream.