| 1 | """Perplexity Sonar, Search API, and Deep Research. |
| 2 | |
| 3 | Direct Perplexity keys are preferred so the source can use first-party Search |
| 4 | API results and async Deep Research. OpenRouter remains a Sonar compatibility |
| 5 | fallback when no direct Perplexity key is configured. |
| 6 | """ |
| 7 | |
| 8 | from __future__ import annotations |
| 9 | |
| 10 | import hashlib |
| 11 | import json |
| 12 | import random |
| 13 | import sys |
| 14 | import time |
| 15 | from datetime import datetime |
| 16 | from urllib.parse import urlparse |
| 17 | |
| 18 | from . import http, log |
| 19 | |
| 20 | |
| 21 | OPENROUTER_URL = "https://openrouter.ai/api/v1/chat/completions" |
| 22 | PERPLEXITY_URL = "https://api.perplexity.ai/v1/sonar" |
| 23 | PERPLEXITY_SEARCH_URL = "https://api.perplexity.ai/search" |
| 24 | PERPLEXITY_ASYNC_URL = "https://api.perplexity.ai/v1/async/sonar" |
| 25 | |
| 26 | OPENROUTER_MODEL_SONAR_PRO = "perplexity/sonar-pro" |
| 27 | OPENROUTER_MODEL_DEEP_RESEARCH = "perplexity/sonar-deep-research" |
| 28 | PERPLEXITY_MODEL_SONAR = "sonar" |
| 29 | PERPLEXITY_MODEL_SONAR_PRO = "sonar-pro" |
| 30 | PERPLEXITY_MODEL_REASONING_PRO = "sonar-reasoning-pro" |
| 31 | PERPLEXITY_MODEL_DEEP_RESEARCH = "sonar-deep-research" |
| 32 | PERPLEXITY_MODE_SONAR = "sonar" |
| 33 | PERPLEXITY_MODE_SEARCH = "search" |
| 34 | PERPLEXITY_MODE_BOTH = "both" |
| 35 | PERPLEXITY_DEFAULT_DEEP_TIMEOUT_SECONDS = 600 |
| 36 | PERPLEXITY_DEEP_INITIAL_POLL_DELAY_SECONDS = 5.0 |
| 37 | PERPLEXITY_DEEP_MAX_POLL_DELAY_SECONDS = 60.0 |
| 38 | |
| 39 | DIRECT_MODELS = { |
| 40 | PERPLEXITY_MODEL_SONAR, |
| 41 | PERPLEXITY_MODEL_SONAR_PRO, |
| 42 | PERPLEXITY_MODEL_REASONING_PRO, |
| 43 | PERPLEXITY_MODEL_DEEP_RESEARCH, |
| 44 | } |
| 45 | DIRECT_MODES = { |
| 46 | PERPLEXITY_MODE_SONAR, |
| 47 | PERPLEXITY_MODE_SEARCH, |
| 48 | PERPLEXITY_MODE_BOTH, |
| 49 | } |
| 50 | SEARCH_CONTEXT_SIZES = {"low", "medium", "high"} |
| 51 | SEARCH_RECENCY_FILTERS = {"hour", "day", "week", "month", "year"} |
| 52 | SONAR_SEARCH_MODES = {"web", "academic", "sec"} |
| 53 | REASONING_EFFORTS = {"minimal", "low", "medium", "high"} |
| 54 | |
| 55 | |
| 56 | class AsyncDeepResearchTimeout(TimeoutError): |
| 57 | def __init__(self, metadata: dict): |
| 58 | timeout_seconds = metadata.get("asyncTimeoutSeconds") or "unknown" |
| 59 | super().__init__(f"Async Deep Research exceeded {timeout_seconds}s wall timeout") |
| 60 | self.metadata = metadata |
| 61 | |
| 62 | |
| 63 | class AsyncDeepResearchFailed(RuntimeError): |
| 64 | def __init__(self, metadata: dict): |
| 65 | message = metadata.get("asyncErrorMessage") or "Async Deep Research failed" |
| 66 | super().__init__(str(message)) |
| 67 | self.metadata = metadata |
| 68 | |
| 69 | |
| 70 | class AsyncDeepResearchPollError(RuntimeError): |
| 71 | def __init__(self, metadata: dict): |
| 72 | message = metadata.get("asyncPollError") or "Async Deep Research poll failed" |
| 73 | super().__init__(str(message)) |
| 74 | self.metadata = metadata |
| 75 | |
| 76 | |
| 77 | def _log(msg: str): |
| 78 | log.source_log("Perplexity", msg, tty_only=False) |
| 79 | |
| 80 | |
| 81 | def _domain(url: str) -> str: |
| 82 | return urlparse(url).netloc.strip().lower() |
| 83 | |
| 84 | |
| 85 | def _provider(config: dict, deep: bool) -> tuple[str, str, str, str] | None: |
| 86 | """Return (provider, api_key, url, model), preferring direct Perplexity.""" |
| 87 | if config.get("PERPLEXITY_API_KEY"): |
| 88 | model = _direct_model(config, deep) |
| 89 | url = PERPLEXITY_ASYNC_URL if deep else PERPLEXITY_URL |
| 90 | return "perplexity", config["PERPLEXITY_API_KEY"], url, model |
| 91 | if config.get("OPENROUTER_API_KEY"): |
| 92 | model = OPENROUTER_MODEL_DEEP_RESEARCH if deep else OPENROUTER_MODEL_SONAR_PRO |
| 93 | return "openrouter", config["OPENROUTER_API_KEY"], OPENROUTER_URL, model |
| 94 | return None |
| 95 | |
| 96 | |
| 97 | def _config_text(config: dict, key: str) -> str: |
| 98 | return str(config.get(key) or "").strip() |
| 99 | |
| 100 | |
| 101 | def _csv_values(raw: str, limit: int | None = None) -> list[str]: |
| 102 | values = [part.strip() for part in raw.split(",") if part.strip()] |
| 103 | # values[:None] already returns the whole list, so no None guard is needed. |
| 104 | return values[:limit] |
| 105 | |
| 106 | |
| 107 | def _direct_model(config: dict, deep: bool) -> str: |
| 108 | if deep: |
| 109 | return PERPLEXITY_MODEL_DEEP_RESEARCH |
| 110 | model = _config_text(config, "LAST30DAYS_PERPLEXITY_MODEL") or PERPLEXITY_MODEL_SONAR_PRO |
| 111 | if model not in DIRECT_MODELS: |
| 112 | _log(f"Unsupported LAST30DAYS_PERPLEXITY_MODEL={model!r}; using sonar-pro") |
| 113 | return PERPLEXITY_MODEL_SONAR_PRO |
| 114 | if model == PERPLEXITY_MODEL_DEEP_RESEARCH: |
| 115 | return PERPLEXITY_MODEL_SONAR_PRO |
| 116 | return model |
| 117 | |
| 118 | |
| 119 | def _mode(config: dict, provider: str, deep: bool) -> str: |
| 120 | if deep: |
| 121 | return PERPLEXITY_MODE_SONAR |
| 122 | mode = (_config_text(config, "LAST30DAYS_PERPLEXITY_MODE") or PERPLEXITY_MODE_SONAR).lower() |
| 123 | if mode not in DIRECT_MODES: |
| 124 | _log(f"Unsupported LAST30DAYS_PERPLEXITY_MODE={mode!r}; using sonar") |
| 125 | return PERPLEXITY_MODE_SONAR |
| 126 | if provider != "perplexity" and mode != PERPLEXITY_MODE_SONAR: |
| 127 | _log("Search API modes require PERPLEXITY_API_KEY; using OpenRouter Sonar fallback") |
| 128 | return PERPLEXITY_MODE_SONAR |
| 129 | return mode |
| 130 | |
| 131 | |
| 132 | def _positive_int(raw: object, default: int, min_value: int, max_value: int | None = None) -> int: |
| 133 | try: |
| 134 | value = int(str(raw).strip()) |
| 135 | except (TypeError, ValueError): |
| 136 | return default |
| 137 | value = max(value, min_value) |
| 138 | if max_value is not None: |
| 139 | value = min(value, max_value) |
| 140 | return value |
| 141 | |
| 142 | |
| 143 | def _mmddyyyy(date: str | None) -> str | None: |
| 144 | if not date: |
| 145 | return None |
| 146 | try: |
| 147 | return datetime.strptime(date, "%Y-%m-%d").strftime("%m/%d/%Y") |
| 148 | except ValueError: |
| 149 | return None |
| 150 | |
| 151 | |
| 152 | def _usage(data: dict) -> dict: |
| 153 | usage = data.get("usage") |
| 154 | return usage if isinstance(usage, dict) else {} |
| 155 | |
| 156 | |
| 157 | def _idempotency_key(json_data: dict) -> str: |
| 158 | payload = json.dumps(json_data, sort_keys=True, separators=(",", ":")) |
| 159 | digest = hashlib.sha256(payload.encode("utf-8")).hexdigest()[:32] |
| 160 | return f"last30days:{digest}" |
| 161 | |
| 162 | |
| 163 | def _async_metadata( |
| 164 | data: dict, |
| 165 | request_id: str, |
| 166 | timeout_seconds: int, |
| 167 | idempotency_key: str, |
| 168 | poll_count: int, |
| 169 | local_status: str, |
| 170 | ) -> dict: |
| 171 | metadata = { |
| 172 | "async": True, |
| 173 | "asyncRequestId": request_id, |
| 174 | "asyncStatus": data.get("status"), |
| 175 | "asyncTimeoutSeconds": timeout_seconds, |
| 176 | "asyncIdempotencyKey": idempotency_key, |
| 177 | "asyncPollCount": poll_count, |
| 178 | "asyncLocalStatus": local_status, |
| 179 | "asyncCreatedAt": data.get("created_at"), |
| 180 | "asyncStartedAt": data.get("started_at"), |
| 181 | "asyncCompletedAt": data.get("completed_at"), |
| 182 | "asyncFailedAt": data.get("failed_at"), |
| 183 | "asyncErrorMessage": data.get("error_message"), |
| 184 | } |
| 185 | return {k: v for k, v in metadata.items() if v is not None} |
| 186 | |
| 187 | |
| 188 | def _error_artifact(exc: Exception) -> dict: |
| 189 | artifact = { |
| 190 | "error": type(exc).__name__, |
| 191 | "message": str(exc)[:200], |
| 192 | } |
| 193 | if isinstance(exc, http.HTTPError): |
| 194 | artifact["statusCode"] = exc.status_code |
| 195 | return artifact |
| 196 | |
| 197 | |
| 198 | def _empty_async_sonar_artifact( |
| 199 | provider: str, |
| 200 | model: str, |
| 201 | deep: bool, |
| 202 | query: str, |
| 203 | data: dict, |
| 204 | async_artifact: dict, |
| 205 | error: str, |
| 206 | message: str, |
| 207 | ) -> dict: |
| 208 | if not async_artifact: |
| 209 | return {} |
| 210 | artifact = { |
| 211 | "label": "perplexity", |
| 212 | "provider": provider, |
| 213 | "mode": PERPLEXITY_MODE_SONAR, |
| 214 | "endpoint": "async-sonar", |
| 215 | "model": model, |
| 216 | "deep": deep, |
| 217 | "query": query, |
| 218 | "error": error, |
| 219 | "synthesisLength": 0, |
| 220 | "citationCount": 0, |
| 221 | "usage": _usage(data), |
| 222 | **async_artifact, |
| 223 | } |
| 224 | if not artifact.get("asyncErrorMessage"): |
| 225 | artifact["asyncErrorMessage"] = message |
| 226 | return artifact |
| 227 | |
| 228 | |
| 229 | def _build_sonar_payload(prompt: str, model: str, date_range: tuple[str, str], config: dict) -> dict: |
| 230 | payload = { |
| 231 | "model": model, |
| 232 | "messages": [{"role": "user", "content": prompt}], |
| 233 | } |
| 234 | |
| 235 | from_date, to_date = date_range |
| 236 | web_options: dict[str, object] = {} |
| 237 | search_mode = _config_text(config, "LAST30DAYS_PERPLEXITY_SEARCH_MODE").lower() |
| 238 | if search_mode in SONAR_SEARCH_MODES: |
| 239 | web_options["search_mode"] = search_mode |
| 240 | |
| 241 | domains = _csv_values(_config_text(config, "LAST30DAYS_PERPLEXITY_DOMAIN_FILTER"), limit=20) |
| 242 | if domains: |
| 243 | web_options["search_domain_filter"] = domains |
| 244 | |
| 245 | languages = _csv_values(_config_text(config, "LAST30DAYS_PERPLEXITY_LANGUAGE_FILTER"), limit=20) |
| 246 | if languages: |
| 247 | web_options["search_language_filter"] = languages |
| 248 | |
| 249 | recency = _config_text(config, "LAST30DAYS_PERPLEXITY_RECENCY_FILTER").lower() |
| 250 | if recency in SEARCH_RECENCY_FILTERS: |
| 251 | web_options["search_recency_filter"] = recency |
| 252 | |
| 253 | after = _mmddyyyy(from_date) |
| 254 | before = _mmddyyyy(to_date) |
| 255 | if after: |
| 256 | web_options["search_after_date_filter"] = after |
| 257 | if before: |
| 258 | web_options["search_before_date_filter"] = before |
| 259 | |
| 260 | if web_options: |
| 261 | payload["web_search_options"] = web_options |
| 262 | |
| 263 | effort = _config_text(config, "LAST30DAYS_PERPLEXITY_REASONING_EFFORT").lower() |
| 264 | if effort in REASONING_EFFORTS: |
| 265 | payload["reasoning_effort"] = effort |
| 266 | |
| 267 | return payload |
| 268 | |
| 269 | |
| 270 | def _build_search_payload(query: str, date_range: tuple[str, str], config: dict) -> dict: |
| 271 | from_date, to_date = date_range |
| 272 | payload: dict[str, object] = { |
| 273 | "query": query, |
| 274 | "max_results": _positive_int(config.get("LAST30DAYS_PERPLEXITY_MAX_RESULTS"), 10, 1, 20), |
| 275 | } |
| 276 | |
| 277 | context_size = _config_text(config, "LAST30DAYS_PERPLEXITY_SEARCH_CONTEXT_SIZE").lower() |
| 278 | if context_size in SEARCH_CONTEXT_SIZES: |
| 279 | payload["search_context_size"] = context_size |
| 280 | |
| 281 | country = _config_text(config, "LAST30DAYS_PERPLEXITY_COUNTRY").upper() |
| 282 | if len(country) == 2: |
| 283 | payload["country"] = country |
| 284 | |
| 285 | domains = _csv_values(_config_text(config, "LAST30DAYS_PERPLEXITY_DOMAIN_FILTER"), limit=20) |
| 286 | if domains: |
| 287 | payload["search_domain_filter"] = domains |
| 288 | |
| 289 | languages = _csv_values(_config_text(config, "LAST30DAYS_PERPLEXITY_LANGUAGE_FILTER"), limit=20) |
| 290 | if languages: |
| 291 | payload["search_language_filter"] = languages |
| 292 | |
| 293 | after = _mmddyyyy(from_date) |
| 294 | before = _mmddyyyy(to_date) |
| 295 | if after: |
| 296 | payload["search_after_date_filter"] = after |
| 297 | if before: |
| 298 | payload["search_before_date_filter"] = before |
| 299 | |
| 300 | # Perplexity Search API rejects search_recency_filter when explicit |
| 301 | # published-date filters are present. last30days already passes an exact |
| 302 | # date range, so prefer that and keep recency only for undated callers. |
| 303 | recency = _config_text(config, "LAST30DAYS_PERPLEXITY_RECENCY_FILTER").lower() |
| 304 | if recency in SEARCH_RECENCY_FILTERS and not (after or before): |
| 305 | payload["search_recency_filter"] = recency |
| 306 | |
| 307 | return payload |
| 308 | |
| 309 | |
| 310 | def _append_citation(citations: list[dict], seen_urls: set[str], citation: dict) -> None: |
| 311 | url = (citation.get("url") or "").strip() |
| 312 | if not url or url in seen_urls: |
| 313 | return |
| 314 | seen_urls.add(url) |
| 315 | citations.append({ |
| 316 | "url": url, |
| 317 | "title": citation.get("title") or "", |
| 318 | "snippet": citation.get("snippet") or "", |
| 319 | "date": citation.get("date"), |
| 320 | }) |
| 321 | |
| 322 | |
| 323 | def _extract_citations(data: dict, choice: dict) -> list[dict]: |
| 324 | """Extract citations from direct Perplexity and OpenRouter response shapes.""" |
| 325 | citations: list[dict] = [] |
| 326 | seen_urls: set[str] = set() |
| 327 | |
| 328 | search_results_by_url: dict[str, dict] = {} |
| 329 | for result in data.get("search_results") or []: |
| 330 | if not isinstance(result, dict): |
| 331 | continue |
| 332 | url = (result.get("url") or "").strip() |
| 333 | if not url: |
| 334 | continue |
| 335 | search_results_by_url[url] = result |
| 336 | _append_citation(citations, seen_urls, result) |
| 337 | |
| 338 | for url in data.get("citations") or []: |
| 339 | if not isinstance(url, str): |
| 340 | continue |
| 341 | result = search_results_by_url.get(url, {}) |
| 342 | _append_citation(citations, seen_urls, { |
| 343 | "url": url, |
| 344 | "title": result.get("title") or _domain(url), |
| 345 | "snippet": result.get("snippet") or "", |
| 346 | "date": result.get("date"), |
| 347 | }) |
| 348 | |
| 349 | annotations = choice.get("message", {}).get("annotations", []) |
| 350 | for ann in annotations or []: |
| 351 | if not isinstance(ann, dict): |
| 352 | continue |
| 353 | url_citation = ann.get("url_citation", {}) |
| 354 | if not isinstance(url_citation, dict): |
| 355 | continue |
| 356 | _append_citation(citations, seen_urls, { |
| 357 | "url": url_citation.get("url") or "", |
| 358 | "title": url_citation.get("title") or "", |
| 359 | }) |
| 360 | |
| 361 | return citations |
| 362 | |
| 363 | |
| 364 | def _poll_async_sonar(json_data: dict, headers: dict, config: dict) -> tuple[dict, dict]: |
| 365 | timeout_seconds = _positive_int( |
| 366 | config.get("LAST30DAYS_PERPLEXITY_DEEP_TIMEOUT_SECONDS"), |
| 367 | PERPLEXITY_DEFAULT_DEEP_TIMEOUT_SECONDS, |
| 368 | 1, |
| 369 | None, |
| 370 | ) |
| 371 | idempotency_key = _idempotency_key(json_data) |
| 372 | created = http.post( |
| 373 | PERPLEXITY_ASYNC_URL, |
| 374 | {"request": json_data, "idempotency_key": idempotency_key}, |
| 375 | headers=headers, |
| 376 | timeout=30, |
| 377 | retries=2, |
| 378 | ) |
| 379 | request_id = created.get("id") |
| 380 | if not request_id: |
| 381 | raise http.HTTPError("Async Deep Research response missing id") |
| 382 | |
| 383 | deadline = time.monotonic() + timeout_seconds |
| 384 | poll_url = f"{PERPLEXITY_ASYNC_URL}/{request_id}" |
| 385 | delay = PERPLEXITY_DEEP_INITIAL_POLL_DELAY_SECONDS |
| 386 | last_status = created.get("status") |
| 387 | poll_count = 0 |
| 388 | last_data = created |
| 389 | if last_status: |
| 390 | _log(f"Deep Research async status: {last_status}") |
| 391 | |
| 392 | while time.monotonic() < deadline: |
| 393 | try: |
| 394 | data = http.get(poll_url, headers=headers, timeout=30, retries=2) |
| 395 | except http.HTTPError as e: |
| 396 | metadata = _async_metadata( |
| 397 | last_data, request_id, timeout_seconds, idempotency_key, poll_count + 1, |
| 398 | "POLL_ERROR", |
| 399 | ) |
| 400 | metadata["asyncPollError"] = str(e) |
| 401 | if e.status_code is not None: |
| 402 | metadata["asyncPollStatusCode"] = e.status_code |
| 403 | raise AsyncDeepResearchPollError(metadata) |
| 404 | poll_count += 1 |
| 405 | last_data = data |
| 406 | status = data.get("status") |
| 407 | if status and status != last_status: |
| 408 | _log(f"Deep Research async status: {status}") |
| 409 | last_status = status |
| 410 | if status == "COMPLETED": |
| 411 | response = data.get("response") |
| 412 | if not isinstance(response, dict): |
| 413 | metadata = _async_metadata( |
| 414 | data, request_id, timeout_seconds, idempotency_key, poll_count, |
| 415 | "FAILED_REMOTE", |
| 416 | ) |
| 417 | metadata["asyncErrorMessage"] = "Async Deep Research completed without response" |
| 418 | raise AsyncDeepResearchFailed(metadata) |
| 419 | return response, _async_metadata( |
| 420 | data, request_id, timeout_seconds, idempotency_key, poll_count, |
| 421 | "COMPLETED_REMOTE", |
| 422 | ) |
| 423 | if status == "FAILED": |
| 424 | raise AsyncDeepResearchFailed(_async_metadata( |
| 425 | data, request_id, timeout_seconds, idempotency_key, poll_count, |
| 426 | "FAILED_REMOTE", |
| 427 | )) |
| 428 | remaining = deadline - time.monotonic() |
| 429 | if remaining <= 0: |
| 430 | break |
| 431 | jitter = random.uniform(0, 2) |
| 432 | time.sleep(min(delay + jitter, max(0.1, remaining))) |
| 433 | delay = min(delay * 1.5, PERPLEXITY_DEEP_MAX_POLL_DELAY_SECONDS) |
| 434 | |
| 435 | raise AsyncDeepResearchTimeout(_async_metadata( |
| 436 | last_data, request_id, timeout_seconds, idempotency_key, poll_count, |
| 437 | "PENDING_REMOTE", |
| 438 | )) |
| 439 | |
| 440 | |
| 441 | def _search_api( |
| 442 | query: str, |
| 443 | date_range: tuple[str, str], |
| 444 | config: dict, |
| 445 | api_key: str, |
| 446 | ) -> tuple[list[dict], dict]: |
| 447 | from_date, to_date = date_range |
| 448 | headers = { |
| 449 | "Authorization": f"Bearer {api_key}", |
| 450 | "Content-Type": "application/json", |
| 451 | } |
| 452 | payload = _build_search_payload(query, date_range, config) |
| 453 | _log(f"Querying Perplexity Search API for '{query}' ({from_date} to {to_date})") |
| 454 | |
| 455 | data = http.post(PERPLEXITY_SEARCH_URL, payload, headers=headers, timeout=30) |
| 456 | results = data.get("results") or [] |
| 457 | if not isinstance(results, list): |
| 458 | results = [] |
| 459 | |
| 460 | items = [] |
| 461 | for i, result in enumerate(results): |
| 462 | if not isinstance(result, dict): |
| 463 | continue |
| 464 | url = (result.get("url") or "").strip() |
| 465 | if not url: |
| 466 | continue |
| 467 | items.append({ |
| 468 | "id": f"PXS{i + 1}", |
| 469 | "title": result.get("title") or _domain(url), |
| 470 | "url": url, |
| 471 | "source_domain": _domain(url), |
| 472 | "snippet": result.get("snippet") or "", |
| 473 | "date": result.get("date"), |
| 474 | "relevance": max(0.55, 0.85 - (i * 0.03)), |
| 475 | "why_relevant": f"Ranked by Perplexity Search API for '{query}'", |
| 476 | "engagement": {}, |
| 477 | "metadata": { |
| 478 | "last_updated": result.get("last_updated"), |
| 479 | "perplexity_search_id": data.get("id"), |
| 480 | }, |
| 481 | }) |
| 482 | |
| 483 | artifact = { |
| 484 | "label": "perplexity", |
| 485 | "provider": "perplexity", |
| 486 | "mode": PERPLEXITY_MODE_SEARCH, |
| 487 | "endpoint": "search", |
| 488 | "query": query, |
| 489 | "resultCount": len(items), |
| 490 | "request": { |
| 491 | k: v |
| 492 | for k, v in payload.items() |
| 493 | if k not in {"query"} |
| 494 | }, |
| 495 | "responseId": data.get("id"), |
| 496 | "serverTime": data.get("server_time"), |
| 497 | } |
| 498 | _log(f"Got {len(items)} Search API results") |
| 499 | return items, artifact |
| 500 | |
| 501 | |
| 502 | def _sonar_search( |
| 503 | query: str, |
| 504 | date_range: tuple[str, str], |
| 505 | config: dict, |
| 506 | provider: str, |
| 507 | api_key: str, |
| 508 | url: str, |
| 509 | model: str, |
| 510 | deep: bool, |
| 511 | ) -> tuple[list[dict], dict]: |
| 512 | from_date, to_date = date_range |
| 513 | timeout = 120 if deep else 30 |
| 514 | |
| 515 | if deep: |
| 516 | print("[Perplexity] Using Deep Research (~$0.90/query)", file=sys.stderr) |
| 517 | |
| 518 | prompt = ( |
| 519 | f"What has been happening with {query} between {from_date} and {to_date}? " |
| 520 | "Include specific dates, names, numbers, and sources." |
| 521 | ) |
| 522 | |
| 523 | headers = { |
| 524 | "Authorization": f"Bearer {api_key}", |
| 525 | "Content-Type": "application/json", |
| 526 | } |
| 527 | |
| 528 | json_data = _build_sonar_payload(prompt, model, date_range, config) |
| 529 | if provider != "perplexity": |
| 530 | json_data.pop("web_search_options", None) |
| 531 | json_data.pop("reasoning_effort", None) |
| 532 | |
| 533 | _log(f"Querying {provider} {model} for '{query}' ({from_date} to {to_date})") |
| 534 | |
| 535 | async_artifact = {} |
| 536 | if provider == "perplexity" and deep: |
| 537 | data, async_artifact = _poll_async_sonar(json_data, headers, config) |
| 538 | else: |
| 539 | data = http.post(url, json_data, headers=headers, timeout=timeout) |
| 540 | |
| 541 | # Parse response |
| 542 | choices = data.get("choices", []) |
| 543 | if not choices: |
| 544 | _log("No choices in response") |
| 545 | return [], _empty_async_sonar_artifact( |
| 546 | provider, model, deep, query, data, async_artifact, |
| 547 | "empty_choices", |
| 548 | "Async Deep Research completed without choices", |
| 549 | ) |
| 550 | |
| 551 | choice = choices[0] if isinstance(choices[0], dict) else {} |
| 552 | message = choice.get("message") |
| 553 | message = message if isinstance(message, dict) else {} |
| 554 | synthesis = message.get("content") or "" |
| 555 | if not isinstance(synthesis, str): |
| 556 | synthesis = "" |
| 557 | if not synthesis: |
| 558 | _log("Empty synthesis content") |
| 559 | return [], _empty_async_sonar_artifact( |
| 560 | provider, model, deep, query, data, async_artifact, |
| 561 | "empty_synthesis", |
| 562 | "Async Deep Research completed with empty synthesis", |
| 563 | ) |
| 564 | |
| 565 | citations = _extract_citations(data, choice) |
| 566 | |
| 567 | _log(f"Got synthesis ({len(synthesis)} chars) with {len(citations)} citations") |
| 568 | |
| 569 | # Build items list |
| 570 | items = [] |
| 571 | |
| 572 | # Primary item: the synthesis itself |
| 573 | snippet = synthesis[:2000] |
| 574 | items.append({ |
| 575 | "id": "PX1", |
| 576 | "title": f"Perplexity {'Deep Research' if deep else 'Sonar'}: {query}", |
| 577 | "url": "", |
| 578 | "source_domain": "perplexity.ai", |
| 579 | "snippet": snippet, |
| 580 | "date": to_date, |
| 581 | "relevance": 0.9, |
| 582 | "why_relevant": f"AI synthesis of recent activity for '{query}'", |
| 583 | "engagement": {"citations": len(citations)}, |
| 584 | "metadata": { |
| 585 | "citations": citations, |
| 586 | "usage": _usage(data), |
| 587 | **async_artifact, |
| 588 | }, |
| 589 | }) |
| 590 | |
| 591 | # Individual items for each citation |
| 592 | for i, cit in enumerate(citations): |
| 593 | items.append({ |
| 594 | "id": f"PX{i + 2}", |
| 595 | "title": cit["title"] or _domain(cit["url"]), |
| 596 | "url": cit["url"], |
| 597 | "source_domain": _domain(cit["url"]), |
| 598 | "snippet": cit.get("snippet") or "", |
| 599 | "date": cit.get("date"), |
| 600 | "relevance": 0.7, |
| 601 | "why_relevant": f"Cited in Perplexity synthesis for '{query}'", |
| 602 | "engagement": {"citations": 1}, |
| 603 | "metadata": {"citations": [cit]}, |
| 604 | }) |
| 605 | |
| 606 | artifact = { |
| 607 | "label": "perplexity", |
| 608 | "provider": provider, |
| 609 | "mode": PERPLEXITY_MODE_SONAR, |
| 610 | "endpoint": "async-sonar" if async_artifact else "sonar", |
| 611 | "model": model, |
| 612 | "deep": deep, |
| 613 | "query": query, |
| 614 | "synthesisLength": len(synthesis), |
| 615 | "citationCount": len(citations), |
| 616 | "usage": _usage(data), |
| 617 | **async_artifact, |
| 618 | } |
| 619 | |
| 620 | return items, artifact |
| 621 | |
| 622 | |
| 623 | def _merge_sonar_and_search(sonar_items: list[dict], search_items: list[dict]) -> list[dict]: |
| 624 | if not sonar_items: |
| 625 | return search_items |
| 626 | merged = sonar_items[:1] |
| 627 | seen_urls = {item.get("url") for item in merged if item.get("url")} |
| 628 | for item in [*search_items, *sonar_items[1:]]: |
| 629 | url = item.get("url") |
| 630 | if url and url in seen_urls: |
| 631 | continue |
| 632 | if url: |
| 633 | seen_urls.add(url) |
| 634 | merged.append(item) |
| 635 | return merged |
| 636 | |
| 637 | |
| 638 | def search( |
| 639 | query: str, |
| 640 | date_range: tuple[str, str], |
| 641 | config: dict, |
| 642 | deep: bool = False, |
| 643 | ) -> tuple[list[dict], dict]: |
| 644 | """Search via Perplexity Sonar Pro or Deep Research. |
| 645 | |
| 646 | Args: |
| 647 | query: Search topic |
| 648 | date_range: (from_date, to_date) as YYYY-MM-DD strings |
| 649 | config: Must contain PERPLEXITY_API_KEY or OPENROUTER_API_KEY |
| 650 | deep: Use Deep Research model (~$0.90/query) instead of Sonar Pro |
| 651 | |
| 652 | Returns: |
| 653 | Tuple of (items list, artifact dict). |
| 654 | """ |
| 655 | resolved = _provider(config, deep) |
| 656 | if not resolved: |
| 657 | _log("No PERPLEXITY_API_KEY or OPENROUTER_API_KEY configured, skipping") |
| 658 | return [], {} |
| 659 | provider, api_key, url, model = resolved |
| 660 | mode = _mode(config, provider, deep) |
| 661 | |
| 662 | try: |
| 663 | if mode == PERPLEXITY_MODE_SEARCH: |
| 664 | return _search_api(query, date_range, config, api_key) |
| 665 | if mode == PERPLEXITY_MODE_BOTH: |
| 666 | search_items: list[dict] = [] |
| 667 | sonar_items: list[dict] = [] |
| 668 | search_artifact: dict = {} |
| 669 | sonar_artifact: dict = {} |
| 670 | try: |
| 671 | search_items, search_artifact = _search_api(query, date_range, config, api_key) |
| 672 | except Exception as e: |
| 673 | _log(f"Search API leg failed in both mode: {e}") |
| 674 | search_artifact = _error_artifact(e) |
| 675 | try: |
| 676 | sonar_items, sonar_artifact = _sonar_search( |
| 677 | query, date_range, config, provider, api_key, url, model, deep |
| 678 | ) |
| 679 | except Exception as e: |
| 680 | _log(f"Sonar leg failed in both mode: {e}") |
| 681 | sonar_artifact = _error_artifact(e) |
| 682 | items = _merge_sonar_and_search(sonar_items, search_items) |
| 683 | return items, { |
| 684 | "label": "perplexity", |
| 685 | "provider": "perplexity", |
| 686 | "mode": PERPLEXITY_MODE_BOTH, |
| 687 | "query": query, |
| 688 | "search": search_artifact, |
| 689 | "sonar": sonar_artifact, |
| 690 | "itemCount": len(items), |
| 691 | } |
| 692 | return _sonar_search(query, date_range, config, provider, api_key, url, model, deep) |
| 693 | except http.HTTPError as e: |
| 694 | if e.status_code == 401: |
| 695 | _log(f"Invalid {provider} API key (401)") |
| 696 | elif e.status_code == 429: |
| 697 | _log(f"Rate limited by {provider} (429)") |
| 698 | else: |
| 699 | _log(f"HTTP error: {e}") |
| 700 | return [], {} |
| 701 | except AsyncDeepResearchTimeout as e: |
| 702 | _log(f"Request timed out: {e}") |
| 703 | return [], { |
| 704 | "label": "perplexity", |
| 705 | "provider": provider, |
| 706 | "mode": PERPLEXITY_MODE_SONAR, |
| 707 | "endpoint": "async-sonar", |
| 708 | "model": model, |
| 709 | "deep": deep, |
| 710 | "query": query, |
| 711 | "error": "timeout", |
| 712 | **e.metadata, |
| 713 | } |
| 714 | except AsyncDeepResearchFailed as e: |
| 715 | _log(f"Deep Research failed: {e}") |
| 716 | return [], { |
| 717 | "label": "perplexity", |
| 718 | "provider": provider, |
| 719 | "mode": PERPLEXITY_MODE_SONAR, |
| 720 | "endpoint": "async-sonar", |
| 721 | "model": model, |
| 722 | "deep": deep, |
| 723 | "query": query, |
| 724 | "error": "failed", |
| 725 | **e.metadata, |
| 726 | } |
| 727 | except AsyncDeepResearchPollError as e: |
| 728 | _log(f"Deep Research poll failed: {e}") |
| 729 | return [], { |
| 730 | "label": "perplexity", |
| 731 | "provider": provider, |
| 732 | "mode": PERPLEXITY_MODE_SONAR, |
| 733 | "endpoint": "async-sonar", |
| 734 | "model": model, |
| 735 | "deep": deep, |
| 736 | "query": query, |
| 737 | "error": "poll_error", |
| 738 | **e.metadata, |
| 739 | } |
| 740 | except TimeoutError as e: |
| 741 | _log(f"Request timed out: {e}") |
| 742 | return [], {"label": "perplexity", "provider": provider, "error": "timeout"} |
| 743 | except Exception as e: |
| 744 | _log(f"Request failed: {e}") |
| 745 | return [], {} |
| 746 |