Build and run agents you can see, understand and trust.
Python agentchatbotlarge-language-modelsllmllm-agentmcpmulti-agentmulti-modalreact-agentrealtime-agentvoice-agentvoice-assistant
import { useQuery, useQueryClient } from '@tanstack/react-query';
import { useCallback } from 'react';
import { sessionApi } from '../api';
import type { SessionView, CreateSessionRequest, UpdateSessionRequest } from '../api';
/** Stable empty list, so `sessions` keeps its identity between renders. */
const NO_SESSIONS: SessionView[] = [];
/** Every agent's session list, so one mutation refreshes all of them. */
const SESSIONS_KEY = ['sessions'];
/**
* Manages session views for a given agent.
*
* Each entry is a `SessionView` (record + is_running + optional team
* detail) — the same shape the backend returns.
*
* Backed by the shared query cache, keyed on `agentId`. The chat page
* mounts this hook twice — once for the outer sidebar's list, once
* inside the viewport — and both now read the same entry, so a rename
* from the sidebar reaches the viewport's header without either side
* knowing about the other. Drilling into a team member gives the
* viewport a different `agentId`, hence its own entry, which is what
* that view wants.
*
* @param agentId - The agent whose sessions to load. Pass null to skip fetching.
* @returns Object with the loaded `sessions` array plus `loading` /
* `error` flags and `refetch` / `create` / `update` / `remove`
* helpers that all keep the list in sync.
*/
export function useSessions(agentId: string | null) {
const queryClient = useQueryClient();
const {
data,
isPending,
error,
refetch: runRefetch,
} = useQuery({
queryKey: [...SESSIONS_KEY, agentId],
queryFn: () => sessionApi.list(agentId as string).then((res) => res.sessions),
enabled: agentId !== null,
// This list carries `is_running` and team membership, both of
// which move underneath the page. Sharing one copy between the
// two mounts is the point here, not skipping the read — so a
// view that needs it still re-reads it when it mounts.
staleTime: 0,
});
/** Drop every agent's list, so both mounts re-read after a write. */
const invalidate = useCallback(
() => queryClient.invalidateQueries({ queryKey: SESSIONS_KEY }),
[queryClient],
);
/**
* Reload the session list.
*
* Also returns the fresh list, so a caller reacting to an event can
* act on it immediately — reading the `sessions` state right after
* awaiting would still see the pre-update value from its closure.
*
* @returns The reloaded views, or an empty array when there is no
* agent or the request failed.
*/
const refetch = useCallback(async (): Promise<SessionView[]> => {
const result = await runRefetch();
return result.data ?? NO_SESSIONS;
}, [runRefetch]);
/** Creates a new session and refreshes the list. */
const create = useCallback(
async (body: CreateSessionRequest) => {
const res = await sessionApi.create(body);
await invalidate();
return res;
},
[invalidate],
);
/** Updates a session's model config and refreshes the list. */
const update = useCallback(
async (sessionId: string, body: UpdateSessionRequest) => {
if (!agentId) throw new Error('No agent selected');
const res = await sessionApi.update(sessionId, agentId, body);
await invalidate();
return res;
},
[agentId, invalidate],
);
/** Deletes a session and refreshes the list. */
const remove = useCallback(
async (sessionId: string) => {
if (!agentId) throw new Error('No agent selected');
await sessionApi.delete(sessionId, agentId);
await invalidate();
},
[agentId, invalidate],
);
return {
sessions: data ?? NO_SESSIONS,
loading: agentId !== null && isPending,
error: error as Error | null,
refetch,
create,
update,
remove,
};
}