Shared State
Read and write agent state with Angular signals, and choose between shared state and agent context.
Shared state and agent context solve two different data flows.
| Data flow | Use |
|---|---|
| Agent and application both read and write the value | injectAgentStore and agent.setState |
| The application owns the value and the agent only reads it | connectAgentContext or CopilotKitAgentContext |
This page covers shared state. For values the agent only reads, see Agent Read-Only Context.
Read agent state
injectAgentStore returns a signal that resolves one agent. The store exposes
messages, state, and run status as nested signals.
import { Component, computed } from "@angular/core";
import { injectAgentStore } from "@copilotkit/angular";
type WorkspaceState = {
notes: string[];
priority: "low" | "normal" | "high";
};
const EMPTY_STATE: WorkspaceState = {
notes: [],
priority: "normal",
};
@Component({
selector: "app-workspace",
template: `
<p>Priority: {{ state().priority }}</p>
<ul>
@for (note of state().notes; track note) {
<li>{{ note }}</li>
}
</ul>
<button type="button" (click)="setPriority('high')">
Mark high priority
</button>
`,
})
export class WorkspaceComponent {
readonly store = injectAgentStore("default");
readonly state = computed(
() => (this.store().state() as WorkspaceState | undefined) ?? EMPTY_STATE,
);
protected setPriority(priority: WorkspaceState["priority"]): void {
const agent = this.store().agent;
const current = (agent.state as WorkspaceState | undefined) ?? EMPTY_STATE;
agent.setState({ ...current, priority });
}
}Read through store().state() so Angular tracks changes. Write through the
plain AG-UI agent at store().agent. Replace the object instead of mutating it
in place.
Click Mark high priority, then ask the agent for the current priority. It answers "high".
The same store gives you:
store().messages()for the current conversationstore().isRunning()for loading controlsstore().agentforsetState,addMessage, andabortRun
To start a run, call inject(CopilotKit).core.runAgent({ agent }), as in the
headless chat.
That call sends your registered frontend tools and agent context with the run.
Keep ownership clear
- Use shared state for data that the agent may change.
- Use context for application-owned facts.
- Keep durable user preferences in your own data store; publish only the part the current agent needs.
- Scope
injectAgentStoreto the same agent id as the chat or workflow that reads the state.
Next steps
How is this guide?