Catches up ag2's shared-state-read-write and subagents demos (added in
PR #4359, post smalls batch 1) to parity with langgraph-python.
- shared-state-read-write: nested use-agent/use-agent-read and
set-state/use-agent-write on page.tsx; notes-card-render and
preferences-card-render on the card components (6 regions total)
- subagents: delegation-log-frontend on the log component;
subagent-setup + supervisor-delegation-tools on src/agents/subagents.py
wrapping AG2/AutoGen ConversableAgent definitions and the
@tool-decorated supervisor delegation functions (3 regions total)
Restores @region[provider-setup] + @region[configure-suggestions] on
agentic-chat/page.tsx and @region[frontend-tool-registration] +
@region[frontend-tool-handler] on frontend-tools/page.tsx that were
inadvertently dropped during the shared-tools symlink refactor.
- claude-sdk-typescript manifest: G3's headless-complete and a2ui-fixed-schema entries pointed at -agent.ts paths but the actual files are -prompt.ts. Bundle-demo-content blew up on the missing files.
- generate-catalog test: update reference and crewai-crews wired-count assertions for the post-blitz state. Multiple integrations now tie at 39 wired; auto-detection picks the alphabetical winner (langgraph-fastapi) over langgraph-python. crewai wired count moved to 37.
agno's package.json adds @copilotkit/shared, @copilotkit/voice, and openai for the new voice/multimodal/byoc demos using the same 'next' / '^5.9.0' pins langgraph-python uses (those identical pins are already in the baseline). Bumping the baseline per the validator's explicit suggestion.
- generate-registry: cross-validate not_supported_features against features list
- catalog-types: widen manifestation union to include 'starter'
- claude-sdk-python: drop invalid 'declarative-schema' from generative_ui enum
- crewai-crews: remove duplicate animated_preview_url key in mcp-apps demo
- built-in-agent + pydantic-ai: REASONING_MODEL env-var fallback so deployers can swap if gpt-5.2/gpt-5 isn't available
Wave-3 fill for the Claude Agent SDK (TypeScript) showcase.
- hitl-in-chat / hitl-in-chat-booking: frontend-tool useHumanInTheLoop
flow with an inline TimePickerCard. Pass-through agent — Claude sees
the book_call tool definition forwarded via AG-UI and calls it; the
frontend renders the picker and returns the user's selection.
- a2ui-fixed-schema: dedicated /a2ui-fixed-schema endpoint on the
Express agent ships flight_schema.json and a display_flight tool that
emits an a2ui_operations container. New runtime route
copilotkit-a2ui-fixed-schema runs the A2UI middleware with
injectA2UITool: false because the backend owns the rendering tool.
- headless-complete: dedicated /headless-complete endpoint exposes
get_weather + get_stock_price backend tools, frontend ports the
fully-headless useAgent surface and per-tool renderers from
claude-sdk-python (highlight_note remains a frontend-only useComponent
tool). New runtime route copilotkit-headless-complete proxies to it.
Adds a dedicated A2UI dynamic-schema cell for the Agno integration:
new `a2ui_dynamic_agent.py` owns its own `generate_a2ui` tool that
reads the runtime-injected client catalog from
`run_context.session_state["copilotkit"]["context"]` and feeds it
to a secondary OpenAI client bound to `render_a2ui`. Result is an
`a2ui_operations` container the runtime A2UI middleware forwards
to the registered frontend catalog renderer.
Unlike the main agent's existing `generate_a2ui` (hardcoded
internal catalog), this dedicated agent stays in sync with whatever
catalog the frontend registers via `<CopilotKit a2ui={{ catalog }}>`.
Frontend ports `page.tsx` + Zod definitions / renderers / catalog
from the langgraph-python reference. Dedicated runtime route at
`/api/copilotkit-declarative-gen-ui` runs with
`injectA2UITool: false` so the agent owns the tool itself.
Adds a dedicated A2UI fixed-schema cell for the Agno integration:
backend `a2ui_fixed_agent.py` ships `flight_schema.json` +
`booked_schema.json` and a single `display_flight` tool that emits
an `a2ui_operations` container directly (no secondary LLM).
Frontend ports `page.tsx` + Zod definitions / renderers / catalog
from the langgraph-python reference. Dedicated runtime route at
`/api/copilotkit-a2ui-fixed-schema` runs the A2UI middleware with
`injectA2UITool: false` so the agent owns its rendering tool.
Port the A2UI dynamic-schema demo to the built-in-agent showcase. The
in-process tanstack agent owns a generate_a2ui tool that fires a
secondary LLM call (forced JSON output) to design the surface tree from
the registered client catalog. The dedicated runtime route runs the
A2UI middleware with injectA2UITool: false so the runtime does not
duplicate the tool slot; the middleware still serialises the catalog
into input.context and detects the a2ui_operations container in tool
results.
Manifest: register both new demos under features and demos.
Port the A2UI fixed-schema flight-card demo to built-in-agent. The
in-process tanstack agent owns a display_flight tool that emits an
a2ui_operations container directly; the dedicated runtime route runs the
A2UI middleware with injectA2UITool: false so the runtime does not
duplicate the slot.
Ports the langgraph-python hitl-in-chat demo to PydanticAI using the frontend-tool pattern: a Zod-typed book_call tool defined via useHumanInTheLoop renders a TimePickerCard inline in chat. Backend agent is a chat-only PydanticAI Agent with no tools — the AG-UI bridge surfaces the frontend-registered tool to the model on each run. Adds hitl-in-chat-booking as a manifest alias pointing at the same route.
Adds not_supported_features to the manifest for gen-ui-interrupt,
interrupt-headless, and hitl (the LangGraph graph-interrupt variant —
distinct from the supported hitl-in-chat / hitl-in-app / hitl-in-chat-booking).
All three depend on a graph-interrupt primitive that LangGraph
provides via interrupt() nodes. The built-in-agent integration uses
TanStack AI's chat-completions adapter, which has no equivalent
node-level pause/resume — so the underlying primitive simply isn't
available.
Adds placeholder pages and READMEs for gen-ui-interrupt and
interrupt-headless explaining the limitation and pointing at
langgraph-python for a working reference. Adds a disambiguation
README in the existing hitl/ directory (which hosts the supported
hitl-in-chat demo) to clarify the distinction with the unsupported
hitl feature ID.
Adds page.tsx + README stubs for gen-ui-interrupt and interrupt-headless.
Both rely on LangGraph's graph-level interrupt() / Command(resume=...)
primitive, which the Claude Agent SDK's Anthropic Messages stream does
not expose. Stubs explain the gap and link users to the hitl /
hitl-in-app cells, which model approval gating with regular tool calls.
Implements agentic-chat-reasoning, reasoning-default-render, and
tool-rendering-reasoning-chain for the built-in-agent integration.
The default tanstack factory uses gpt-4o (non-reasoning), so
REASONING_* events never flow. This commit adds a dedicated
reasoning-factory wrapping the v2 runtime with a reasoning-capable
OpenAI model (gpt-5.2) and reasoning_effort: low. The runtime's
tanstack converter already translates upstream
REASONING_START/REASONING_MESSAGE_CONTENT/REASONING_END events into
AG-UI reasoning events, so once the model emits reasoning the chain
surfaces in the chat with no extra plumbing.
A single shared route /api/copilotkit-reasoning registers all three
agents by ID, keeping bundle and cold-start cost out of unrelated
demos.
A polished brand-themed chat surface over the shared Claude agent.
Mirrors the simplified spring-ai port: gradient background, brand
fonts, seeded suggestion pills, plus a small decorative side panel
(static SVG sparkline + intent breakdown). No A2UI / declarative
generative UI infrastructure required — pure frontend dressing on top
of the existing /api/copilotkit endpoint.
The full canonical version (langgraph-python ExampleCanvas + A2UI
catalog) is documented as out-of-scope in the per-demo README.
Adds the MCP Apps demo to the claude-sdk-python integration. The
dedicated /mcp-apps FastAPI endpoint forwards AG-UI input_data.tools
(injected by the runtime's MCP Apps middleware) straight to the
Anthropic Messages API and streams tool_use blocks back as AG-UI
TOOL_CALL_* events. The runtime middleware then fetches each MCP UI
resource and emits the activity event the built-in
MCPAppsActivityRenderer paints in the chat.
Also extends the manifest schema with not_supported_features (a
pre-approved field for surfacing stub pages on features that cannot
be ported to a given backend).
Both demos are built on useLangGraphInterrupt, which hooks into the
LangGraph interrupt lifecycle. AWS Strands does not expose an equivalent
first-class interrupt primitive, and we don't yet have a Strands-side
adapter. Rather than skipping them silently, surface them under the
manifest's new not_supported_features list and ship per-demo stub pages
that explain the rationale and link to ergonomic replacements
(hitl-in-chat for gen-ui-interrupt; headless-complete + hitl-in-chat
for interrupt-headless).
- manifest.yaml: add not_supported_features list with both ids.
- src/app/demos/gen-ui-interrupt/page.tsx + README.md: stub.
- src/app/demos/interrupt-headless/page.tsx + README.md: stub.
- PARITY_NOTES.md: cross-reference the stubs from the Skipped section.
Simplified port of the langgraph-python flagship beautiful-chat. The
canonical reference combines THREE features (A2UI Dynamic + Open
Generative UI + MCP Apps); the AG2 port ships the first two together
in a single dedicated runtime so users can see them coexist. MCP is
left to the existing /demos/mcp-apps cell, and the canonical's app-mode
toggle / todos canvas is intentionally out of scope.
Components:
- src/agents/beautiful_chat.py — ConversableAgent that owns generate_a2ui
explicitly (mirrors a2ui_dynamic.py); system prompt steers the LLM toward
generate_a2ui for catalog-bound visuals and generateSandboxedUi for
free-form / educational visualisations.
- src/agent_server.py — mounted at /beautiful-chat.
- src/app/api/copilotkit-beautiful-chat/route.ts — dedicated runtime with
a2ui.injectA2UITool: false (agent owns the tool) and
openGenerativeUI.agents: ['beautiful-chat']. Both flags scoped to this
cell so they don't leak into the shared /api/copilotkit endpoint.
- src/app/demos/beautiful-chat/page.tsx — single CopilotKit provider with
both a2ui catalog and openGenerativeUI design skill; reuses the catalog
from /demos/declarative-gen-ui to avoid duplication.
- src/app/demos/beautiful-chat/README.md — per-demo doc.
Manifest: added beautiful-chat to features list and demos section.
The langgraph-python beautiful-chat ships a much larger surface
(ExampleCanvas, GenerativeUIExamples, declarative A2UI catalog,
dedicated runtime combining openGenerativeUI + a2ui + mcpApps). That
ecosystem depends on dozens of starter-level sub-components.
This Strands port mirrors the spring-ai sibling pattern: a polished
landing-style chat shell with brand theming and seeded suggestions,
sitting on top of the shared Strands agent. It is the 'show me a
great-looking starter chat' cell.
Cosmetic styling is entirely on the frontend; the backend agent and
the shared /api/copilotkit runtime are unchanged. Register the new
'beautiful-chat' agent name in the shared route so the cell resolves.
- src/app/demos/beautiful-chat/page.tsx + README.md: brand-themed
CopilotChat shell with useConfigureSuggestions for seeded prompts.
- src/app/api/copilotkit/route.ts: register 'beautiful-chat' agent name.
- manifest.yaml: surface beautiful-chat under features + demos.
- PARITY_NOTES.md: move beautiful-chat from Skipped to Shipped (simplified).
Sam Julien