LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows
Python ai-agentsdata-engineeringdocument-aigenerative-aiidpjson-extractionllmmcp-serverocrpdf-extractionprompt-engineeringstructured-output
import {
generateApiRunStatusId,
generateUUID,
PROMPT_RUN_API_STATUSES,
PROMPT_RUN_TYPES,
} from "../helpers/GetStaticData";
import { useAlertStore } from "../store/alert-store";
import { useCustomToolStore } from "../store/custom-tool-store";
import { usePromptRunQueueStore } from "../store/prompt-run-queue-store";
import { usePromptRunStatusStore } from "../store/prompt-run-status-store";
import { useSessionStore } from "../store/session-store";
import { useAxiosPrivate } from "./useAxiosPrivate";
import { useExceptionHandler } from "./useExceptionHandler";
import usePromptOutput from "./usePromptOutput";
// Tracks the latest run nonce per (promptId, statusKey) so stale timeouts
// from a previous run don't falsely cancel a newer run of the same combo.
const runNonceMap = new Map();
// Must exceed server LLM adapter timeout (default 900s) so UI doesn't bail first.
const SOCKET_TIMEOUT_MS = 16 * 60 * 1000; // 16 minutes
const usePromptRun = () => {
const { pushPromptRunApi, freeActiveApi } = usePromptRunQueueStore();
const { generatePromptOutputKey } = usePromptOutput();
const { addPromptStatus, removePromptStatus } = usePromptRunStatusStore();
const { details, llmProfiles, listOfDocs } = useCustomToolStore();
const { sessionDetails } = useSessionStore();
const axiosPrivate = useAxiosPrivate();
const { setAlertDetails } = useAlertStore();
const handleException = useExceptionHandler();
const makeApiRequest = (requestOptions) => axiosPrivate(requestOptions);
const runPromptApi = (api) => {
const [promptId, docId, profileId] = api.split("__");
const runId = generateUUID();
const body = {
id: promptId,
document_id: docId,
profile_manager: profileId,
run_id: runId,
};
const requestOptions = {
method: "POST",
url: `/api/v1/unstract/${sessionDetails?.orgId}/prompt-studio/fetch_response/${details?.tool_id}`,
headers: {
"X-CSRFToken": sessionDetails?.csrfToken,
"Content-Type": "application/json",
},
data: body,
};
// Fire-and-forget: POST dispatches the Celery task, socket delivers result.
const statusKey = generateApiRunStatusId(docId, profileId);
const nonceKey = `${promptId}__${statusKey}`;
const nonce = generateUUID();
runNonceMap.set(nonceKey, nonce);
makeApiRequest(requestOptions)
.then((res) => {
// Handle pending-indexing response: clear running status immediately
if (res?.data?.status === "pending") {
removePromptStatus(promptId, statusKey);
setAlertDetails({
type: "info",
content:
res?.data?.message || "Document is being indexed. Please wait.",
});
return;
}
// Timeout safety net: clear stale status if socket event never arrives.
// Only clears if this is still the latest run for this combo.
setTimeout(() => {
if (runNonceMap.get(nonceKey) !== nonce) {
return;
}
const current = usePromptRunStatusStore.getState().promptRunStatus;
if (
current?.[promptId]?.[statusKey] === PROMPT_RUN_API_STATUSES.RUNNING
) {
removePromptStatus(promptId, statusKey);
setAlertDetails({
type: "warning",
content: "Prompt execution timed out. Please try again.",
});
}
runNonceMap.delete(nonceKey);
}, SOCKET_TIMEOUT_MS);
})
.catch((err) => {
setAlertDetails(
handleException(err, "Failed to generate prompt output"),
);
removePromptStatus(promptId, statusKey);
runNonceMap.delete(nonceKey);
})
.finally(() => {
freeActiveApi();
});
};
const runBulkPromptApi = (promptIds, docId, profileId) => {
const runId = generateUUID();
const body = {
prompt_ids: promptIds,
document_id: docId,
profile_manager: profileId,
run_id: runId,
};
const requestOptions = {
method: "POST",
url: `/api/v1/unstract/${sessionDetails?.orgId}/prompt-studio/bulk_fetch_response/${details?.tool_id}`,
headers: {
"X-CSRFToken": sessionDetails?.csrfToken,
"Content-Type": "application/json",
},
data: body,
};
const statusKey = generateApiRunStatusId(docId, profileId);
const nonces = {};
promptIds.forEach((promptId) => {
const nonceKey = `${promptId}__${statusKey}`;
const nonce = generateUUID();
nonces[promptId] = nonce;
runNonceMap.set(nonceKey, nonce);
});
// Timeout safety net: clear stale status if socket event never arrives.
// Only clears if this is still the latest run for each prompt combo.
const clearStaleStatuses = () => {
promptIds.forEach((promptId) => {
const nonceKey = `${promptId}__${statusKey}`;
if (runNonceMap.get(nonceKey) !== nonces[promptId]) {
return;
}
const current = usePromptRunStatusStore.getState().promptRunStatus;
if (
current?.[promptId]?.[statusKey] === PROMPT_RUN_API_STATUSES.RUNNING
) {
removePromptStatus(promptId, statusKey);
}
runNonceMap.delete(nonceKey);
});
};
makeApiRequest(requestOptions)
.then((res) => {
if (res?.data?.status === "pending") {
promptIds.forEach((promptId) => {
removePromptStatus(promptId, statusKey);
});
setAlertDetails({
type: "info",
content:
res?.data?.message || "Document is being indexed. Please wait.",
});
return;
}
setTimeout(clearStaleStatuses, SOCKET_TIMEOUT_MS);
})
.catch((err) => {
setAlertDetails(
handleException(err, "Failed to generate prompt output"),
);
promptIds.forEach((promptId) => {
const nonceKey = `${promptId}__${statusKey}`;
removePromptStatus(promptId, statusKey);
runNonceMap.delete(nonceKey);
});
})
.finally(() => {
freeActiveApi();
});
};
const runPrompt = (listOfApis) => {
if (!listOfApis?.length) {
return;
}
listOfApis.forEach(runPromptApi);
};
const prepareApiRequests = (promptIds, profileIds, docIds) => {
const apiRequestsToQueue = [];
const promptRunApiStatus = [];
const combinations = [];
for (const promptId of promptIds) {
for (const profileId of profileIds) {
for (const docId of docIds) {
combinations.push({ promptId, profileId, docId });
}
}
}
combinations.forEach(({ promptId, profileId, docId }) => {
if (!promptRunApiStatus[promptId]) {
promptRunApiStatus[promptId] = {};
}
const key = generatePromptOutputKey(
promptId,
docId,
profileId,
null,
false,
);
const statusKey = generateApiRunStatusId(docId, profileId);
promptRunApiStatus[promptId][statusKey] = PROMPT_RUN_API_STATUSES.RUNNING;
apiRequestsToQueue.push(key);
});
return { apiRequestsToQueue, promptRunApiStatus };
};
const syncPromptRunApisAndStatus = (promptApis) => {
const promptRunApiStatus = {};
promptApis.forEach((apiDetails) => {
const [promptId, docId, profileId] = apiDetails.split("__");
const statusKey = generateApiRunStatusId(docId, profileId);
if (!promptRunApiStatus[promptId]) {
promptRunApiStatus[promptId] = {};
}
promptRunApiStatus[promptId][statusKey] = PROMPT_RUN_API_STATUSES.RUNNING;
});
addPromptStatus(promptRunApiStatus);
pushPromptRunApi(promptApis);
};
const handlePromptRunRequest = (
promptRunType,
promptId = null,
profileId = null,
docId = null,
) => {
const promptIds = promptId
? [promptId]
: details?.prompts.map((p) => p.prompt_id) || [];
const profileIds = profileId
? [profileId]
: llmProfiles.map((p) => p.profile_id) || [];
const docIds = docId ? [docId] : listOfDocs.map((d) => d.document_id) || [];
let apiRequestsToQueue = [];
let promptRunApiStatus = {};
const paramsMap = {
[PROMPT_RUN_TYPES.RUN_ONE_PROMPT_ONE_LLM_ONE_DOC]: {
requiredParams: ["promptId", "profileId", "docId"],
prompts: [promptId],
profiles: [profileId],
docs: [docId],
},
[PROMPT_RUN_TYPES.RUN_ONE_PROMPT_ONE_LLM_ALL_DOCS]: {
requiredParams: ["promptId", "profileId"],
prompts: [promptId],
profiles: [profileId],
docs: docIds,
},
[PROMPT_RUN_TYPES.RUN_ONE_PROMPT_ALL_LLMS_ONE_DOC]: {
requiredParams: ["promptId", "docId"],
prompts: [promptId],
profiles: profileIds,
docs: [docId],
},
[PROMPT_RUN_TYPES.RUN_ONE_PROMPT_ALL_LLMS_ALL_DOCS]: {
requiredParams: ["promptId"],
prompts: [promptId],
profiles: profileIds,
docs: docIds,
},
[PROMPT_RUN_TYPES.RUN_ALL_PROMPTS_ALL_LLMS_ONE_DOC]: {
requiredParams: ["docId"],
prompts: promptIds,
profiles: profileIds,
docs: [docId],
},
[PROMPT_RUN_TYPES.RUN_ALL_PROMPTS_ALL_LLMS_ALL_DOCS]: {
requiredParams: [],
prompts: promptIds,
profiles: profileIds,
docs: docIds,
},
};
const params = paramsMap[promptRunType];
if (!params) {
return;
}
const paramValues = { promptId, profileId, docId };
const missingParams = params.requiredParams.filter(
(param) => !paramValues[param],
);
if (missingParams.length > 0) {
return;
}
({ apiRequestsToQueue, promptRunApiStatus } = prepareApiRequests(
params.prompts,
params.profiles,
params.docs,
));
addPromptStatus(promptRunApiStatus);
// Use bulk API when multiple prompts target the same (profile, doc) to
// prevent the "Document being indexed" race condition.
const isBulk = params.prompts.length > 1;
if (isBulk) {
for (const pId of params.profiles) {
for (const dId of params.docs) {
runBulkPromptApi(params.prompts, dId, pId);
}
}
} else {
pushPromptRunApi(apiRequestsToQueue);
}
};
return {
runPrompt,
handlePromptRunRequest,
syncPromptRunApisAndStatus,
};
};
export default usePromptRun;