| 1 | """Perplexity Agent API, Search API, and OpenRouter compatibility integration. |
| 2 | |
| 3 | The Perplexity source is paid and opt-in. A direct key uses a controlled Agent |
| 4 | API request with only web search enabled. Direct Deep Research uses an Agent |
| 5 | API background run with the dynamic high preset. When no direct Perplexity key |
| 6 | exists, OpenRouter preserves the legacy synchronous Sonar fallback. |
| 7 | """ |
| 8 | |
| 9 | from __future__ import annotations |
| 10 | |
| 11 | import time |
| 12 | from datetime import datetime |
| 13 | from typing import Any |
| 14 | from urllib.parse import urlparse |
| 15 | |
| 16 | from . import health, http, log |
| 17 | |
| 18 | |
| 19 | OPENROUTER_URL = "https://openrouter.ai/api/v1/chat/completions" |
| 20 | PERPLEXITY_AGENT_URL = "https://api.perplexity.ai/v1/agent" |
| 21 | PERPLEXITY_SEARCH_URL = "https://api.perplexity.ai/search" |
| 22 | |
| 23 | PERPLEXITY_MODE_AGENT = "agent" |
| 24 | PERPLEXITY_MODE_SONAR = "sonar" # Direct-key alias and OpenRouter artifact mode. |
| 25 | PERPLEXITY_MODE_SEARCH = "search" |
| 26 | PERPLEXITY_MODE_BOTH = "both" |
| 27 | PERPLEXITY_DEFAULT_AGENT_TIMEOUT_SECONDS = 120 |
| 28 | PERPLEXITY_DEFAULT_DEEP_TIMEOUT_SECONDS = 600 |
| 29 | PERPLEXITY_DEEP_INITIAL_POLL_DELAY_SECONDS = 5.0 |
| 30 | PERPLEXITY_DEEP_MAX_POLL_DELAY_SECONDS = 60.0 |
| 31 | |
| 32 | PERPLEXITY_DEFAULT_AGENT_MODEL = "perplexity/sonar" |
| 33 | OPENROUTER_MODEL_SONAR_PRO = "perplexity/sonar-pro" |
| 34 | OPENROUTER_MODEL_DEEP_RESEARCH = "perplexity/sonar-deep-research" |
| 35 | PERPLEXITY_DEFAULT_ANTHROPIC_MAX_OUTPUT_TOKENS = 4096 |
| 36 | PERPLEXITY_CONTROLLED_PROFILE = "last30days-controlled-web-search/v1" |
| 37 | PERPLEXITY_PRESET_PROFILE = "perplexity-agent-preset" |
| 38 | AGENT_PRESETS = {"fast", "low", "medium", "high"} |
| 39 | DIRECT_MODES = { |
| 40 | PERPLEXITY_MODE_AGENT, |
| 41 | PERPLEXITY_MODE_SEARCH, |
| 42 | PERPLEXITY_MODE_BOTH, |
| 43 | } |
| 44 | SEARCH_CONTEXT_SIZES = {"low", "medium", "high"} |
| 45 | SEARCH_RECENCY_FILTERS = {"hour", "day", "week", "month", "year"} |
| 46 | REASONING_EFFORTS = {"minimal", "low", "medium", "high"} |
| 47 | |
| 48 | CONTROLLED_AGENT_INSTRUCTIONS = ( |
| 49 | "Use the supplied web search tool for current, source-grounded research. " |
| 50 | "Keep the answer concise. Cite source-backed claims in the answer. " |
| 51 | "Do not use tools other than those supplied in this request." |
| 52 | ) |
| 53 | |
| 54 | |
| 55 | class AgentBackgroundTimeout(TimeoutError): |
| 56 | def __init__(self, metadata: dict[str, Any]): |
| 57 | timeout_seconds = metadata.get("backgroundTimeoutSeconds") or "unknown" |
| 58 | super().__init__(f"Agent background run exceeded {timeout_seconds}s wall timeout") |
| 59 | self.metadata = metadata |
| 60 | |
| 61 | |
| 62 | class AgentBackgroundFailed(RuntimeError): |
| 63 | def __init__(self, metadata: dict[str, Any]): |
| 64 | detail = metadata.get("backgroundErrorMessage") or "Agent background run failed" |
| 65 | super().__init__(str(detail)) |
| 66 | self.metadata = metadata |
| 67 | |
| 68 | |
| 69 | class AgentBackgroundPollError(RuntimeError): |
| 70 | def __init__(self, metadata: dict[str, Any]): |
| 71 | detail = metadata.get("backgroundPollError") or "Agent background poll failed" |
| 72 | super().__init__(str(detail)) |
| 73 | self.metadata = metadata |
| 74 | |
| 75 | |
| 76 | def _log(message: str) -> None: |
| 77 | log.source_log("Perplexity", message, tty_only=False) |
| 78 | |
| 79 | |
| 80 | def _domain(url: str) -> str: |
| 81 | return urlparse(url).netloc.strip().lower() |
| 82 | |
| 83 | |
| 84 | def _config_text(config: dict[str, Any], key: str) -> str: |
| 85 | return str(config.get(key) or "").strip() |
| 86 | |
| 87 | |
| 88 | def _csv_values(raw: str, limit: int | None = None) -> list[str]: |
| 89 | values = [part.strip() for part in raw.split(",") if part.strip()] |
| 90 | return values[:limit] |
| 91 | |
| 92 | |
| 93 | def _positive_int( |
| 94 | raw: object, |
| 95 | default: int, |
| 96 | min_value: int, |
| 97 | max_value: int | None = None, |
| 98 | ) -> int: |
| 99 | try: |
| 100 | value = int(str(raw).strip()) |
| 101 | except (TypeError, ValueError): |
| 102 | return default |
| 103 | value = max(value, min_value) |
| 104 | if max_value is not None: |
| 105 | value = min(value, max_value) |
| 106 | return value |
| 107 | |
| 108 | |
| 109 | def _mmddyyyy(date: str | None) -> str | None: |
| 110 | if not date: |
| 111 | return None |
| 112 | try: |
| 113 | return datetime.strptime(date, "%Y-%m-%d").strftime("%m/%d/%Y") |
| 114 | except ValueError: |
| 115 | return None |
| 116 | |
| 117 | |
| 118 | def _usage(data: dict[str, Any]) -> dict[str, Any]: |
| 119 | usage = data.get("usage") |
| 120 | return usage if isinstance(usage, dict) else {} |
| 121 | |
| 122 | |
| 123 | def _error_artifact(exc: Exception) -> dict[str, Any]: |
| 124 | artifact: dict[str, Any] = { |
| 125 | "error": type(exc).__name__, |
| 126 | "message": str(exc)[:200], |
| 127 | } |
| 128 | if isinstance(exc, http.HTTPError): |
| 129 | artifact["statusCode"] = exc.status_code |
| 130 | return artifact |
| 131 | |
| 132 | |
| 133 | def _provider(config: dict[str, Any]) -> tuple[str, str] | None: |
| 134 | """Prefer direct Perplexity, then preserve the OpenRouter Sonar fallback.""" |
| 135 | api_key = _config_text(config, "PERPLEXITY_API_KEY") |
| 136 | if api_key: |
| 137 | return "perplexity", api_key |
| 138 | openrouter_key = _config_text(config, "OPENROUTER_API_KEY") |
| 139 | if openrouter_key: |
| 140 | return "openrouter", openrouter_key |
| 141 | return None |
| 142 | |
| 143 | |
| 144 | def _mode(config: dict[str, Any], deep: bool, provider: str) -> str: |
| 145 | if deep: |
| 146 | return ( |
| 147 | PERPLEXITY_MODE_AGENT |
| 148 | if provider == "perplexity" |
| 149 | else PERPLEXITY_MODE_SONAR |
| 150 | ) |
| 151 | |
| 152 | mode = ( |
| 153 | _config_text(config, "LAST30DAYS_PERPLEXITY_MODE") |
| 154 | or PERPLEXITY_MODE_AGENT |
| 155 | ).lower() |
| 156 | if provider == "openrouter": |
| 157 | if mode in {PERPLEXITY_MODE_SEARCH, PERPLEXITY_MODE_BOTH}: |
| 158 | _log( |
| 159 | f"LAST30DAYS_PERPLEXITY_MODE={mode} requires PERPLEXITY_API_KEY; " |
| 160 | "using the OpenRouter Sonar fallback" |
| 161 | ) |
| 162 | return PERPLEXITY_MODE_SONAR |
| 163 | if mode == PERPLEXITY_MODE_SONAR: |
| 164 | _log( |
| 165 | "LAST30DAYS_PERPLEXITY_MODE=sonar is deprecated; " |
| 166 | "using the Agent API controlled profile" |
| 167 | ) |
| 168 | return PERPLEXITY_MODE_AGENT |
| 169 | if mode not in DIRECT_MODES: |
| 170 | _log(f"Unsupported LAST30DAYS_PERPLEXITY_MODE={mode!r}; using agent") |
| 171 | return PERPLEXITY_MODE_AGENT |
| 172 | return mode |
| 173 | |
| 174 | |
| 175 | def _agent_preset(config: dict[str, Any], deep: bool) -> str | None: |
| 176 | if deep: |
| 177 | return "high" |
| 178 | |
| 179 | preset = _config_text(config, "LAST30DAYS_PERPLEXITY_AGENT_PRESET").lower() |
| 180 | if not preset: |
| 181 | return None |
| 182 | if preset in AGENT_PRESETS: |
| 183 | return preset |
| 184 | _log( |
| 185 | "Unsupported LAST30DAYS_PERPLEXITY_AGENT_PRESET=" |
| 186 | f"{preset!r}; using the controlled profile" |
| 187 | ) |
| 188 | return None |
| 189 | |
| 190 | |
| 191 | def _agent_model(config: dict[str, Any]) -> str: |
| 192 | legacy_model = _config_text(config, "LAST30DAYS_PERPLEXITY_MODEL") |
| 193 | if legacy_model: |
| 194 | _log( |
| 195 | "LAST30DAYS_PERPLEXITY_MODEL is a legacy Sonar setting and is " |
| 196 | "ignored by the Agent API; set LAST30DAYS_PERPLEXITY_AGENT_MODEL " |
| 197 | "for an explicit Agent model" |
| 198 | ) |
| 199 | return ( |
| 200 | _config_text(config, "LAST30DAYS_PERPLEXITY_AGENT_MODEL") |
| 201 | or PERPLEXITY_DEFAULT_AGENT_MODEL |
| 202 | ) |
| 203 | |
| 204 | |
| 205 | def _agent_timeout(config: dict[str, Any]) -> int: |
| 206 | return _positive_int( |
| 207 | config.get("LAST30DAYS_PERPLEXITY_AGENT_TIMEOUT_SECONDS"), |
| 208 | PERPLEXITY_DEFAULT_AGENT_TIMEOUT_SECONDS, |
| 209 | 1, |
| 210 | 600, |
| 211 | ) |
| 212 | |
| 213 | |
| 214 | def _safe_error_message(data: dict[str, Any]) -> str | None: |
| 215 | error = data.get("error") |
| 216 | if isinstance(error, str): |
| 217 | return error[:200] |
| 218 | if isinstance(error, dict): |
| 219 | for key in ("message", "detail", "code"): |
| 220 | value = error.get(key) |
| 221 | if isinstance(value, str) and value: |
| 222 | return value[:200] |
| 223 | return None |
| 224 | |
| 225 | |
| 226 | def _safe_incomplete_reason(data: dict[str, Any]) -> str | None: |
| 227 | details = data.get("incomplete_details") |
| 228 | if not isinstance(details, dict): |
| 229 | return None |
| 230 | reason = details.get("reason") |
| 231 | return reason[:100] if isinstance(reason, str) and reason else None |
| 232 | |
| 233 | |
| 234 | def _search_type(config: dict[str, Any]) -> str | None: |
| 235 | """Keep omission intact; search backends are separate from Agent presets.""" |
| 236 | value = _config_text(config, "LAST30DAYS_PERPLEXITY_SEARCH_TYPE").lower() |
| 237 | if not value: |
| 238 | return None |
| 239 | if value not in {"web", "fast"}: |
| 240 | raise ValueError("LAST30DAYS_PERPLEXITY_SEARCH_TYPE must be web or fast") |
| 241 | return value |
| 242 | |
| 243 | |
| 244 | def _build_search_payload( |
| 245 | query: str, |
| 246 | date_range: tuple[str, str], |
| 247 | config: dict[str, Any], |
| 248 | ) -> dict[str, Any]: |
| 249 | from_date, to_date = date_range |
| 250 | payload: dict[str, Any] = { |
| 251 | "query": query, |
| 252 | "max_results": _positive_int( |
| 253 | config.get("LAST30DAYS_PERPLEXITY_MAX_RESULTS"), |
| 254 | 10, |
| 255 | 1, |
| 256 | 20, |
| 257 | ), |
| 258 | } |
| 259 | |
| 260 | context_size = _config_text( |
| 261 | config, |
| 262 | "LAST30DAYS_PERPLEXITY_SEARCH_CONTEXT_SIZE", |
| 263 | ).lower() |
| 264 | search_type = _search_type(config) |
| 265 | if search_type is not None: |
| 266 | payload["search_type"] = search_type |
| 267 | if context_size in SEARCH_CONTEXT_SIZES: |
| 268 | payload["search_context_size"] = context_size |
| 269 | |
| 270 | country = _config_text(config, "LAST30DAYS_PERPLEXITY_COUNTRY").upper() |
| 271 | if len(country) == 2: |
| 272 | payload["country"] = country |
| 273 | |
| 274 | domains = _csv_values( |
| 275 | _config_text(config, "LAST30DAYS_PERPLEXITY_DOMAIN_FILTER"), |
| 276 | limit=20, |
| 277 | ) |
| 278 | if domains: |
| 279 | payload["search_domain_filter"] = domains |
| 280 | |
| 281 | languages = _csv_values( |
| 282 | _config_text(config, "LAST30DAYS_PERPLEXITY_LANGUAGE_FILTER"), |
| 283 | limit=20, |
| 284 | ) |
| 285 | if languages: |
| 286 | payload["search_language_filter"] = languages |
| 287 | |
| 288 | after = _mmddyyyy(from_date) |
| 289 | before = _mmddyyyy(to_date) |
| 290 | if after: |
| 291 | payload["search_after_date_filter"] = after |
| 292 | if before: |
| 293 | payload["search_before_date_filter"] = before |
| 294 | |
| 295 | # Perplexity Search API rejects search_recency_filter when explicit |
| 296 | # published-date filters are present. Preserve the upstream fix. |
| 297 | recency = _config_text( |
| 298 | config, |
| 299 | "LAST30DAYS_PERPLEXITY_RECENCY_FILTER", |
| 300 | ).lower() |
| 301 | if recency in SEARCH_RECENCY_FILTERS and not (after or before): |
| 302 | payload["search_recency_filter"] = recency |
| 303 | |
| 304 | return payload |
| 305 | |
| 306 | |
| 307 | def _build_web_search_tool( |
| 308 | date_range: tuple[str, str], |
| 309 | config: dict[str, Any], |
| 310 | ) -> dict[str, Any]: |
| 311 | from_date, to_date = date_range |
| 312 | tool: dict[str, Any] = { |
| 313 | "type": "web_search", |
| 314 | "max_results": _positive_int( |
| 315 | config.get("LAST30DAYS_PERPLEXITY_MAX_RESULTS"), |
| 316 | 10, |
| 317 | 1, |
| 318 | 20, |
| 319 | ), |
| 320 | } |
| 321 | |
| 322 | context_size = _config_text( |
| 323 | config, |
| 324 | "LAST30DAYS_PERPLEXITY_SEARCH_CONTEXT_SIZE", |
| 325 | ).lower() |
| 326 | search_type = _search_type(config) |
| 327 | if search_type is not None: |
| 328 | tool["search_type"] = search_type |
| 329 | if context_size in SEARCH_CONTEXT_SIZES: |
| 330 | tool["search_context_size"] = context_size |
| 331 | |
| 332 | country = _config_text(config, "LAST30DAYS_PERPLEXITY_COUNTRY").upper() |
| 333 | if len(country) == 2: |
| 334 | tool["user_location"] = {"country": country} |
| 335 | |
| 336 | filters: dict[str, Any] = {} |
| 337 | domains = _csv_values( |
| 338 | _config_text(config, "LAST30DAYS_PERPLEXITY_DOMAIN_FILTER"), |
| 339 | limit=20, |
| 340 | ) |
| 341 | if domains: |
| 342 | filters["search_domain_filter"] = domains |
| 343 | |
| 344 | after = _mmddyyyy(from_date) |
| 345 | before = _mmddyyyy(to_date) |
| 346 | if after: |
| 347 | filters["search_after_date_filter"] = after |
| 348 | if before: |
| 349 | filters["search_before_date_filter"] = before |
| 350 | |
| 351 | recency = _config_text( |
| 352 | config, |
| 353 | "LAST30DAYS_PERPLEXITY_RECENCY_FILTER", |
| 354 | ).lower() |
| 355 | if recency in SEARCH_RECENCY_FILTERS and not (after or before): |
| 356 | filters["search_recency_filter"] = recency |
| 357 | if filters: |
| 358 | tool["filters"] = filters |
| 359 | |
| 360 | language_filter = _config_text( |
| 361 | config, |
| 362 | "LAST30DAYS_PERPLEXITY_LANGUAGE_FILTER", |
| 363 | ) |
| 364 | if language_filter: |
| 365 | _log( |
| 366 | "LAST30DAYS_PERPLEXITY_LANGUAGE_FILTER has no Agent API " |
| 367 | "equivalent and applies only to Search API mode" |
| 368 | ) |
| 369 | search_mode = _config_text(config, "LAST30DAYS_PERPLEXITY_SEARCH_MODE").lower() |
| 370 | if search_mode and search_mode != "web": |
| 371 | _log( |
| 372 | "LAST30DAYS_PERPLEXITY_SEARCH_MODE has no Agent API equivalent; " |
| 373 | "using web search" |
| 374 | ) |
| 375 | |
| 376 | return tool |
| 377 | |
| 378 | |
| 379 | def _safe_request(payload: dict[str, Any]) -> dict[str, Any]: |
| 380 | request: dict[str, Any] = {} |
| 381 | for key in ( |
| 382 | "model", |
| 383 | "preset", |
| 384 | "max_steps", |
| 385 | "max_output_tokens", |
| 386 | "background", |
| 387 | ): |
| 388 | if key in payload: |
| 389 | request[key] = payload[key] |
| 390 | reasoning = payload.get("reasoning") |
| 391 | if isinstance(reasoning, dict) and isinstance(reasoning.get("effort"), str): |
| 392 | request["reasoning"] = {"effort": reasoning["effort"]} |
| 393 | tool_choice = payload.get("tool_choice") |
| 394 | if tool_choice == {"type": "web_search"}: |
| 395 | request["tool_choice"] = tool_choice |
| 396 | tools = payload.get("tools") |
| 397 | if isinstance(tools, list): |
| 398 | request["tools"] = [ |
| 399 | { |
| 400 | key: tool[key] |
| 401 | for key in ( |
| 402 | "type", |
| 403 | "max_results", |
| 404 | "search_context_size", |
| 405 | "search_type", |
| 406 | "user_location", |
| 407 | "filters", |
| 408 | ) |
| 409 | if key in tool |
| 410 | } |
| 411 | for tool in tools |
| 412 | if isinstance(tool, dict) and tool.get("type") == "web_search" |
| 413 | ] |
| 414 | return request |
| 415 | |
| 416 | |
| 417 | def _build_agent_payload( |
| 418 | prompt: str, |
| 419 | date_range: tuple[str, str], |
| 420 | config: dict[str, Any], |
| 421 | deep: bool, |
| 422 | ) -> tuple[dict[str, Any], dict[str, Any]]: |
| 423 | preset = _agent_preset(config, deep) |
| 424 | if preset: |
| 425 | payload = { |
| 426 | "preset": preset, |
| 427 | "input": prompt, |
| 428 | # A supplied web_search tool merges with a preset's tools. This |
| 429 | # preserves the user's date, domain, location, and result bounds |
| 430 | # without claiming to disable any other dynamic-preset tools. |
| 431 | "tools": [_build_web_search_tool(date_range, config)], |
| 432 | } |
| 433 | if deep: |
| 434 | payload["background"] = True |
| 435 | return payload, { |
| 436 | "profile": PERPLEXITY_PRESET_PROFILE, |
| 437 | "preset": preset, |
| 438 | "dynamicPreset": True, |
| 439 | "request": _safe_request(payload), |
| 440 | } |
| 441 | |
| 442 | payload: dict[str, Any] = { |
| 443 | "model": _agent_model(config), |
| 444 | "instructions": CONTROLLED_AGENT_INSTRUCTIONS, |
| 445 | "input": prompt, |
| 446 | "tools": [_build_web_search_tool(date_range, config)], |
| 447 | "tool_choice": {"type": "web_search"}, |
| 448 | "max_steps": _positive_int( |
| 449 | config.get("LAST30DAYS_PERPLEXITY_AGENT_MAX_STEPS"), |
| 450 | 5, |
| 451 | 1, |
| 452 | 15, |
| 453 | ), |
| 454 | } |
| 455 | if str(payload["model"]).lower().startswith("anthropic/"): |
| 456 | payload["max_output_tokens"] = _positive_int( |
| 457 | config.get("LAST30DAYS_PERPLEXITY_AGENT_MAX_OUTPUT_TOKENS"), |
| 458 | PERPLEXITY_DEFAULT_ANTHROPIC_MAX_OUTPUT_TOKENS, |
| 459 | 1, |
| 460 | 32768, |
| 461 | ) |
| 462 | effort = _config_text( |
| 463 | config, |
| 464 | "LAST30DAYS_PERPLEXITY_REASONING_EFFORT", |
| 465 | ).lower() |
| 466 | if effort in REASONING_EFFORTS: |
| 467 | payload["reasoning"] = {"effort": effort} |
| 468 | return payload, { |
| 469 | "profile": PERPLEXITY_CONTROLLED_PROFILE, |
| 470 | "model": payload["model"], |
| 471 | "dynamicPreset": False, |
| 472 | "request": _safe_request(payload), |
| 473 | } |
| 474 | |
| 475 | |
| 476 | def _append_citation( |
| 477 | citations: list[dict[str, Any]], |
| 478 | seen_urls: set[str], |
| 479 | citation: dict[str, Any], |
| 480 | ) -> None: |
| 481 | url = str(citation.get("url") or "").strip() |
| 482 | if not url: |
| 483 | return |
| 484 | if url in seen_urls: |
| 485 | # Message annotations commonly carry only a URL and title. Merge the |
| 486 | # later search_results item instead of discarding its snippet/date. |
| 487 | for existing in citations: |
| 488 | if existing.get("url") != url: |
| 489 | continue |
| 490 | for key in ("title", "snippet", "date"): |
| 491 | if not existing.get(key) and citation.get(key): |
| 492 | existing[key] = citation[key] |
| 493 | return |
| 494 | seen_urls.add(url) |
| 495 | citations.append( |
| 496 | { |
| 497 | "url": url, |
| 498 | "title": citation.get("title") or "", |
| 499 | "snippet": citation.get("snippet") or "", |
| 500 | "date": citation.get("date"), |
| 501 | } |
| 502 | ) |
| 503 | |
| 504 | |
| 505 | def _append_annotations( |
| 506 | citations: list[dict[str, Any]], |
| 507 | seen_urls: set[str], |
| 508 | annotations: Any, |
| 509 | ) -> None: |
| 510 | if not isinstance(annotations, list): |
| 511 | return |
| 512 | for annotation in annotations: |
| 513 | if not isinstance(annotation, dict): |
| 514 | continue |
| 515 | citation = annotation.get("url_citation") |
| 516 | if not isinstance(citation, dict): |
| 517 | citation = annotation |
| 518 | _append_citation(citations, seen_urls, citation) |
| 519 | |
| 520 | |
| 521 | def _extract_agent_citations(data: dict[str, Any]) -> list[dict[str, Any]]: |
| 522 | citations: list[dict[str, Any]] = [] |
| 523 | seen_urls: set[str] = set() |
| 524 | |
| 525 | for result in data.get("results") or []: |
| 526 | if isinstance(result, dict): |
| 527 | _append_citation(citations, seen_urls, result) |
| 528 | |
| 529 | output = data.get("output") |
| 530 | if not isinstance(output, list): |
| 531 | return citations |
| 532 | |
| 533 | for item in output: |
| 534 | if not isinstance(item, dict): |
| 535 | continue |
| 536 | item_type = item.get("type") |
| 537 | if item_type == "search_results": |
| 538 | for result in item.get("results") or []: |
| 539 | if isinstance(result, dict): |
| 540 | _append_citation(citations, seen_urls, result) |
| 541 | continue |
| 542 | if item_type != "message": |
| 543 | continue |
| 544 | _append_annotations(citations, seen_urls, item.get("annotations")) |
| 545 | content = item.get("content") |
| 546 | if not isinstance(content, list): |
| 547 | continue |
| 548 | for part in content: |
| 549 | if isinstance(part, dict): |
| 550 | _append_annotations(citations, seen_urls, part.get("annotations")) |
| 551 | |
| 552 | return citations |
| 553 | |
| 554 | |
| 555 | def _extract_openrouter_citations( |
| 556 | data: dict[str, Any], |
| 557 | choice: dict[str, Any], |
| 558 | ) -> list[dict[str, Any]]: |
| 559 | """Read the legacy OpenAI-compatible Sonar citation shapes.""" |
| 560 | citations: list[dict[str, Any]] = [] |
| 561 | seen_urls: set[str] = set() |
| 562 | search_results: dict[str, dict[str, Any]] = {} |
| 563 | for result in data.get("search_results") or []: |
| 564 | if not isinstance(result, dict): |
| 565 | continue |
| 566 | url = str(result.get("url") or "").strip() |
| 567 | if not url: |
| 568 | continue |
| 569 | search_results[url] = result |
| 570 | _append_citation(citations, seen_urls, result) |
| 571 | |
| 572 | for url in data.get("citations") or []: |
| 573 | if not isinstance(url, str): |
| 574 | continue |
| 575 | result = search_results.get(url, {}) |
| 576 | _append_citation( |
| 577 | citations, |
| 578 | seen_urls, |
| 579 | { |
| 580 | "url": url, |
| 581 | "title": result.get("title") or _domain(url), |
| 582 | "snippet": result.get("snippet") or "", |
| 583 | "date": result.get("date"), |
| 584 | }, |
| 585 | ) |
| 586 | |
| 587 | message = choice.get("message") |
| 588 | if isinstance(message, dict): |
| 589 | _append_annotations(citations, seen_urls, message.get("annotations")) |
| 590 | return citations |
| 591 | |
| 592 | |
| 593 | def _output_types(data: dict[str, Any]) -> list[str]: |
| 594 | output = data.get("output") |
| 595 | if not isinstance(output, list): |
| 596 | return [] |
| 597 | return [ |
| 598 | item["type"] |
| 599 | for item in output |
| 600 | if isinstance(item, dict) and isinstance(item.get("type"), str) |
| 601 | ] |
| 602 | |
| 603 | |
| 604 | def _output_text(data: dict[str, Any]) -> str: |
| 605 | direct = data.get("output_text") |
| 606 | if isinstance(direct, str) and direct.strip(): |
| 607 | return direct |
| 608 | |
| 609 | parts: list[str] = [] |
| 610 | output = data.get("output") |
| 611 | if not isinstance(output, list): |
| 612 | return "" |
| 613 | for item in output: |
| 614 | if not isinstance(item, dict) or item.get("type") != "message": |
| 615 | continue |
| 616 | content = item.get("content") |
| 617 | if isinstance(content, str): |
| 618 | parts.append(content) |
| 619 | continue |
| 620 | if not isinstance(content, list): |
| 621 | continue |
| 622 | for part in content: |
| 623 | if not isinstance(part, dict): |
| 624 | continue |
| 625 | if part.get("type") not in {"output_text", "text"}: |
| 626 | continue |
| 627 | text = part.get("text") |
| 628 | if isinstance(text, str): |
| 629 | parts.append(text) |
| 630 | return "".join(parts).strip() |
| 631 | |
| 632 | |
| 633 | def _background_metadata( |
| 634 | data: dict[str, Any], |
| 635 | response_id: str, |
| 636 | timeout_seconds: int, |
| 637 | poll_count: int, |
| 638 | local_status: str, |
| 639 | ) -> dict[str, Any]: |
| 640 | metadata: dict[str, Any] = { |
| 641 | "background": True, |
| 642 | "responseId": response_id, |
| 643 | "backgroundStatus": data.get("status"), |
| 644 | "servedModel": data.get("model"), |
| 645 | "usage": _usage(data), |
| 646 | "outputTypes": _output_types(data), |
| 647 | "backgroundTimeoutSeconds": timeout_seconds, |
| 648 | "backgroundPollCount": poll_count, |
| 649 | "backgroundLocalStatus": local_status, |
| 650 | } |
| 651 | message = _safe_error_message(data) |
| 652 | if message: |
| 653 | metadata["backgroundErrorMessage"] = message |
| 654 | incomplete_reason = _safe_incomplete_reason(data) |
| 655 | if incomplete_reason: |
| 656 | metadata["incompleteReason"] = incomplete_reason |
| 657 | return {key: value for key, value in metadata.items() if value is not None} |
| 658 | |
| 659 | |
| 660 | def _log_deep_receipt(artifact: dict[str, Any]) -> None: |
| 661 | """Expose the safe Deep receipt to slash-command and CLI consumers.""" |
| 662 | fields = ( |
| 663 | ("model", artifact.get("servedModel")), |
| 664 | ("response_id", artifact.get("responseId")), |
| 665 | ("provider_status", artifact.get("backgroundStatus") or artifact.get("status")), |
| 666 | ("local_status", artifact.get("backgroundLocalStatus")), |
| 667 | ("incomplete_reason", artifact.get("incompleteReason")), |
| 668 | ("polls", artifact.get("backgroundPollCount")), |
| 669 | ("timeout_seconds", artifact.get("backgroundTimeoutSeconds")), |
| 670 | ) |
| 671 | rendered = " ".join( |
| 672 | f"{key}={value}" |
| 673 | for key, value in fields |
| 674 | if value is not None |
| 675 | ) |
| 676 | _log(f"Deep Research receipt: {rendered or 'unavailable'}") |
| 677 | |
| 678 | |
| 679 | def _poll_agent_background( |
| 680 | payload: dict[str, Any], |
| 681 | headers: dict[str, str], |
| 682 | config: dict[str, Any], |
| 683 | ) -> tuple[dict[str, Any], dict[str, Any]]: |
| 684 | timeout_seconds = _positive_int( |
| 685 | config.get("LAST30DAYS_PERPLEXITY_DEEP_TIMEOUT_SECONDS"), |
| 686 | PERPLEXITY_DEFAULT_DEEP_TIMEOUT_SECONDS, |
| 687 | 1, |
| 688 | None, |
| 689 | ) |
| 690 | created = http.post( |
| 691 | PERPLEXITY_AGENT_URL, |
| 692 | payload, |
| 693 | headers=headers, |
| 694 | timeout=30, |
| 695 | retries=1, |
| 696 | ) |
| 697 | response_id = created.get("id") |
| 698 | if not isinstance(response_id, str) or not response_id: |
| 699 | raise http.HTTPError("Agent background response missing id") |
| 700 | |
| 701 | terminal = {"completed", "failed", "cancelled", "incomplete"} |
| 702 | status = str(created.get("status") or "").lower() |
| 703 | data = created |
| 704 | poll_count = 0 |
| 705 | if status: |
| 706 | _log(f"Agent background status: {status}") |
| 707 | |
| 708 | deadline = time.monotonic() + timeout_seconds |
| 709 | delay = PERPLEXITY_DEEP_INITIAL_POLL_DELAY_SECONDS |
| 710 | while status not in terminal: |
| 711 | if time.monotonic() >= deadline: |
| 712 | raise AgentBackgroundTimeout( |
| 713 | _background_metadata( |
| 714 | data, |
| 715 | response_id, |
| 716 | timeout_seconds, |
| 717 | poll_count, |
| 718 | "PENDING_REMOTE", |
| 719 | ) |
| 720 | ) |
| 721 | try: |
| 722 | data = http.get( |
| 723 | f"{PERPLEXITY_AGENT_URL}/{response_id}", |
| 724 | headers=headers, |
| 725 | timeout=30, |
| 726 | retries=2, |
| 727 | deadline_monotonic=deadline, |
| 728 | ) |
| 729 | except http.HTTPError as exc: |
| 730 | if isinstance(exc, http.DeadlineExceeded) or time.monotonic() >= deadline: |
| 731 | raise AgentBackgroundTimeout( |
| 732 | _background_metadata( |
| 733 | data, |
| 734 | response_id, |
| 735 | timeout_seconds, |
| 736 | poll_count + 1, |
| 737 | "PENDING_REMOTE", |
| 738 | ) |
| 739 | ) from exc |
| 740 | metadata = _background_metadata( |
| 741 | data, |
| 742 | response_id, |
| 743 | timeout_seconds, |
| 744 | poll_count + 1, |
| 745 | "POLL_ERROR", |
| 746 | ) |
| 747 | metadata["backgroundPollError"] = str(exc)[:200] |
| 748 | if exc.status_code is not None: |
| 749 | metadata["backgroundPollStatusCode"] = exc.status_code |
| 750 | raise AgentBackgroundPollError(metadata) from exc |
| 751 | |
| 752 | poll_count += 1 |
| 753 | next_status = str(data.get("status") or "").lower() |
| 754 | if next_status and next_status != status: |
| 755 | _log(f"Agent background status: {next_status}") |
| 756 | status = next_status |
| 757 | remaining = deadline - time.monotonic() |
| 758 | if remaining <= 0: |
| 759 | raise AgentBackgroundTimeout( |
| 760 | _background_metadata( |
| 761 | data, |
| 762 | response_id, |
| 763 | timeout_seconds, |
| 764 | poll_count, |
| 765 | "PENDING_REMOTE", |
| 766 | ) |
| 767 | ) |
| 768 | if status in terminal: |
| 769 | break |
| 770 | time.sleep(min(delay, remaining)) |
| 771 | delay = min(delay * 1.5, PERPLEXITY_DEEP_MAX_POLL_DELAY_SECONDS) |
| 772 | |
| 773 | metadata = _background_metadata( |
| 774 | data, |
| 775 | response_id, |
| 776 | timeout_seconds, |
| 777 | poll_count, |
| 778 | "COMPLETED_REMOTE" if status == "completed" else "TERMINAL_REMOTE", |
| 779 | ) |
| 780 | if status == "completed": |
| 781 | return data, metadata |
| 782 | if not status: |
| 783 | metadata["backgroundErrorMessage"] = "Agent background response has no status" |
| 784 | raise AgentBackgroundFailed(metadata) |
| 785 | |
| 786 | |
| 787 | def _search_api( |
| 788 | query: str, |
| 789 | date_range: tuple[str, str], |
| 790 | config: dict[str, Any], |
| 791 | api_key: str, |
| 792 | ) -> tuple[list[dict[str, Any]], dict[str, Any]]: |
| 793 | from_date, to_date = date_range |
| 794 | headers = { |
| 795 | "Authorization": f"Bearer {api_key}", |
| 796 | "Content-Type": "application/json", |
| 797 | } |
| 798 | payload = _build_search_payload(query, date_range, config) |
| 799 | _log(f"Querying Perplexity Search API for '{query}' ({from_date} to {to_date})") |
| 800 | |
| 801 | data = http.post( |
| 802 | PERPLEXITY_SEARCH_URL, |
| 803 | payload, |
| 804 | headers=headers, |
| 805 | timeout=30, |
| 806 | retries=1, |
| 807 | ) |
| 808 | results = data.get("results") or [] |
| 809 | if not isinstance(results, list): |
| 810 | results = [] |
| 811 | |
| 812 | items: list[dict[str, Any]] = [] |
| 813 | for index, result in enumerate(results): |
| 814 | if not isinstance(result, dict): |
| 815 | continue |
| 816 | url = str(result.get("url") or "").strip() |
| 817 | if not url: |
| 818 | continue |
| 819 | items.append( |
| 820 | { |
| 821 | "id": f"PXS{index + 1}", |
| 822 | "title": result.get("title") or _domain(url), |
| 823 | "url": url, |
| 824 | "source_domain": _domain(url), |
| 825 | "snippet": result.get("snippet") or "", |
| 826 | "date": result.get("date"), |
| 827 | "relevance": max(0.55, 0.85 - (index * 0.03)), |
| 828 | "why_relevant": f"Ranked by Perplexity Search API for '{query}'", |
| 829 | "engagement": {}, |
| 830 | "metadata": { |
| 831 | "last_updated": result.get("last_updated"), |
| 832 | "perplexity_search_id": data.get("id"), |
| 833 | }, |
| 834 | } |
| 835 | ) |
| 836 | |
| 837 | artifact = { |
| 838 | "label": "perplexity", |
| 839 | "provider": "perplexity", |
| 840 | "mode": PERPLEXITY_MODE_SEARCH, |
| 841 | "endpoint": "search", |
| 842 | "query": query, |
| 843 | "resultCount": len(items), |
| 844 | "request": {key: value for key, value in payload.items() if key != "query"}, |
| 845 | "responseId": data.get("id"), |
| 846 | "serverTime": data.get("server_time"), |
| 847 | } |
| 848 | _log(f"Got {len(items)} Search API results") |
| 849 | return items, artifact |
| 850 | |
| 851 | |
| 852 | def _agent_failure_artifact( |
| 853 | query: str, |
| 854 | deep: bool, |
| 855 | selection: dict[str, Any], |
| 856 | *, |
| 857 | error: str, |
| 858 | metadata: dict[str, Any] | None = None, |
| 859 | ) -> dict[str, Any]: |
| 860 | return { |
| 861 | "label": "perplexity", |
| 862 | "provider": "perplexity", |
| 863 | "mode": PERPLEXITY_MODE_AGENT, |
| 864 | "endpoint": "agent-background" if deep else "agent", |
| 865 | "deep": deep, |
| 866 | "query": query, |
| 867 | "error": error, |
| 868 | "synthesisLength": 0, |
| 869 | "citationCount": 0, |
| 870 | **selection, |
| 871 | **(metadata or {}), |
| 872 | } |
| 873 | |
| 874 | |
| 875 | def _agent_result( |
| 876 | query: str, |
| 877 | date_range: tuple[str, str], |
| 878 | deep: bool, |
| 879 | data: dict[str, Any], |
| 880 | selection: dict[str, Any], |
| 881 | background_metadata: dict[str, Any] | None = None, |
| 882 | ) -> tuple[list[dict[str, Any]], dict[str, Any]]: |
| 883 | _, to_date = date_range |
| 884 | status = str(data.get("status") or "").lower() |
| 885 | if status and status != "completed": |
| 886 | metadata: dict[str, Any] = { |
| 887 | "responseId": data.get("id"), |
| 888 | "status": data.get("status"), |
| 889 | "servedModel": data.get("model"), |
| 890 | "usage": _usage(data), |
| 891 | "outputTypes": _output_types(data), |
| 892 | } |
| 893 | message = _safe_error_message(data) |
| 894 | if message: |
| 895 | metadata["agentErrorMessage"] = message |
| 896 | incomplete_reason = _safe_incomplete_reason(data) |
| 897 | if incomplete_reason: |
| 898 | metadata["incompleteReason"] = incomplete_reason |
| 899 | if background_metadata: |
| 900 | metadata.update(background_metadata) |
| 901 | return [], _agent_failure_artifact( |
| 902 | query, |
| 903 | deep, |
| 904 | selection, |
| 905 | error=status, |
| 906 | metadata=metadata, |
| 907 | ) |
| 908 | |
| 909 | synthesis = _output_text(data) |
| 910 | citations = _extract_agent_citations(data) |
| 911 | if not synthesis: |
| 912 | _log("Empty Agent API synthesis") |
| 913 | return [], _agent_failure_artifact( |
| 914 | query, |
| 915 | deep, |
| 916 | selection, |
| 917 | error="empty_synthesis", |
| 918 | metadata={ |
| 919 | "responseId": data.get("id"), |
| 920 | "status": data.get("status"), |
| 921 | "servedModel": data.get("model"), |
| 922 | "usage": _usage(data), |
| 923 | "outputTypes": _output_types(data), |
| 924 | **(background_metadata or {}), |
| 925 | }, |
| 926 | ) |
| 927 | |
| 928 | _log(f"Got Agent API synthesis ({len(synthesis)} chars) with {len(citations)} citations") |
| 929 | title_mode = "Deep Research" if deep else "Agent" |
| 930 | items: list[dict[str, Any]] = [ |
| 931 | { |
| 932 | "id": "PX1", |
| 933 | "title": f"Perplexity {title_mode}: {query}", |
| 934 | "url": "", |
| 935 | "source_domain": "perplexity.ai", |
| 936 | "snippet": synthesis[:2000], |
| 937 | "date": to_date, |
| 938 | "relevance": 0.9, |
| 939 | "why_relevant": f"AI synthesis of recent activity for '{query}'", |
| 940 | "engagement": {"citations": len(citations)}, |
| 941 | "metadata": { |
| 942 | "citations": citations, |
| 943 | "usage": _usage(data), |
| 944 | "perplexity_response_id": data.get("id"), |
| 945 | }, |
| 946 | } |
| 947 | ] |
| 948 | for index, citation in enumerate(citations): |
| 949 | items.append( |
| 950 | { |
| 951 | "id": f"PX{index + 2}", |
| 952 | "title": citation["title"] or _domain(citation["url"]), |
| 953 | "url": citation["url"], |
| 954 | "source_domain": _domain(citation["url"]), |
| 955 | "snippet": citation.get("snippet") or "", |
| 956 | "date": citation.get("date"), |
| 957 | "relevance": 0.7, |
| 958 | "why_relevant": f"Cited in Perplexity synthesis for '{query}'", |
| 959 | "engagement": {"citations": 1}, |
| 960 | "metadata": {"citations": [citation]}, |
| 961 | } |
| 962 | ) |
| 963 | |
| 964 | artifact: dict[str, Any] = { |
| 965 | "label": "perplexity", |
| 966 | "provider": "perplexity", |
| 967 | "mode": PERPLEXITY_MODE_AGENT, |
| 968 | "endpoint": "agent-background" if deep else "agent", |
| 969 | "deep": deep, |
| 970 | "query": query, |
| 971 | "synthesisLength": len(synthesis), |
| 972 | "citationCount": len(citations), |
| 973 | "responseId": data.get("id"), |
| 974 | "status": data.get("status"), |
| 975 | "servedModel": data.get("model"), |
| 976 | "usage": _usage(data), |
| 977 | "outputTypes": _output_types(data), |
| 978 | **selection, |
| 979 | } |
| 980 | if background_metadata: |
| 981 | artifact.update(background_metadata) |
| 982 | return items, artifact |
| 983 | |
| 984 | |
| 985 | def _agent_search( |
| 986 | query: str, |
| 987 | date_range: tuple[str, str], |
| 988 | config: dict[str, Any], |
| 989 | api_key: str, |
| 990 | deep: bool, |
| 991 | ) -> tuple[list[dict[str, Any]], dict[str, Any]]: |
| 992 | from_date, to_date = date_range |
| 993 | prompt = ( |
| 994 | f"What has been happening with {query} between {from_date} and {to_date}? " |
| 995 | "Include specific dates, names, numbers, and sources." |
| 996 | ) |
| 997 | payload, selection = _build_agent_payload(prompt, date_range, config, deep) |
| 998 | headers = { |
| 999 | "Authorization": f"Bearer {api_key}", |
| 1000 | "Content-Type": "application/json", |
| 1001 | } |
| 1002 | _log(f"Querying Perplexity Agent API for '{query}' ({from_date} to {to_date})") |
| 1003 | |
| 1004 | try: |
| 1005 | if deep: |
| 1006 | data, background_metadata = _poll_agent_background(payload, headers, config) |
| 1007 | else: |
| 1008 | data = http.post( |
| 1009 | PERPLEXITY_AGENT_URL, |
| 1010 | payload, |
| 1011 | headers=headers, |
| 1012 | timeout=_agent_timeout(config), |
| 1013 | retries=1, |
| 1014 | ) |
| 1015 | background_metadata = None |
| 1016 | except AgentBackgroundTimeout as exc: |
| 1017 | _log(f"Agent background request timed out: {exc}") |
| 1018 | return [], _agent_failure_artifact( |
| 1019 | query, |
| 1020 | deep, |
| 1021 | selection, |
| 1022 | error="timeout", |
| 1023 | metadata=exc.metadata, |
| 1024 | ) |
| 1025 | except AgentBackgroundFailed as exc: |
| 1026 | _log(f"Agent background request failed: {exc}") |
| 1027 | return [], _agent_failure_artifact( |
| 1028 | query, |
| 1029 | deep, |
| 1030 | selection, |
| 1031 | error="failed", |
| 1032 | metadata=exc.metadata, |
| 1033 | ) |
| 1034 | except AgentBackgroundPollError as exc: |
| 1035 | _log(f"Agent background poll failed: {exc}") |
| 1036 | return [], _agent_failure_artifact( |
| 1037 | query, |
| 1038 | deep, |
| 1039 | selection, |
| 1040 | error="poll_error", |
| 1041 | metadata=exc.metadata, |
| 1042 | ) |
| 1043 | |
| 1044 | return _agent_result( |
| 1045 | query, |
| 1046 | date_range, |
| 1047 | deep, |
| 1048 | data, |
| 1049 | selection, |
| 1050 | background_metadata, |
| 1051 | ) |
| 1052 | |
| 1053 | |
| 1054 | def _openrouter_sonar_search( |
| 1055 | query: str, |
| 1056 | date_range: tuple[str, str], |
| 1057 | api_key: str, |
| 1058 | deep: bool, |
| 1059 | ) -> tuple[list[dict[str, Any]], dict[str, Any]]: |
| 1060 | """Preserve the pre-Agent OpenRouter Sonar compatibility path.""" |
| 1061 | from_date, to_date = date_range |
| 1062 | model = ( |
| 1063 | OPENROUTER_MODEL_DEEP_RESEARCH |
| 1064 | if deep |
| 1065 | else OPENROUTER_MODEL_SONAR_PRO |
| 1066 | ) |
| 1067 | prompt = ( |
| 1068 | f"What has been happening with {query} between {from_date} and {to_date}? " |
| 1069 | "Include specific dates, names, numbers, and sources." |
| 1070 | ) |
| 1071 | payload = { |
| 1072 | "model": model, |
| 1073 | "messages": [{"role": "user", "content": prompt}], |
| 1074 | } |
| 1075 | headers = { |
| 1076 | "Authorization": f"Bearer {api_key}", |
| 1077 | "Content-Type": "application/json", |
| 1078 | } |
| 1079 | _log(f"Querying OpenRouter {model} for '{query}' ({from_date} to {to_date})") |
| 1080 | data = http.post( |
| 1081 | OPENROUTER_URL, |
| 1082 | payload, |
| 1083 | headers=headers, |
| 1084 | timeout=120 if deep else 30, |
| 1085 | retries=1, |
| 1086 | ) |
| 1087 | |
| 1088 | choices = data.get("choices") |
| 1089 | if not isinstance(choices, list) or not choices: |
| 1090 | return [], { |
| 1091 | "label": "perplexity", |
| 1092 | "provider": "openrouter", |
| 1093 | "mode": PERPLEXITY_MODE_SONAR, |
| 1094 | "endpoint": "openrouter-chat-completions", |
| 1095 | "model": model, |
| 1096 | "deep": deep, |
| 1097 | "query": query, |
| 1098 | "error": "empty_choices", |
| 1099 | "responseId": data.get("id"), |
| 1100 | "servedModel": data.get("model") or model, |
| 1101 | "usage": _usage(data), |
| 1102 | } |
| 1103 | |
| 1104 | choice = choices[0] if isinstance(choices[0], dict) else {} |
| 1105 | message = choice.get("message") |
| 1106 | message = message if isinstance(message, dict) else {} |
| 1107 | synthesis = message.get("content") |
| 1108 | synthesis = synthesis if isinstance(synthesis, str) else "" |
| 1109 | if not synthesis: |
| 1110 | return [], { |
| 1111 | "label": "perplexity", |
| 1112 | "provider": "openrouter", |
| 1113 | "mode": PERPLEXITY_MODE_SONAR, |
| 1114 | "endpoint": "openrouter-chat-completions", |
| 1115 | "model": model, |
| 1116 | "deep": deep, |
| 1117 | "query": query, |
| 1118 | "error": "empty_synthesis", |
| 1119 | "responseId": data.get("id"), |
| 1120 | "servedModel": data.get("model") or model, |
| 1121 | "usage": _usage(data), |
| 1122 | } |
| 1123 | |
| 1124 | citations = _extract_openrouter_citations(data, choice) |
| 1125 | title_mode = "Deep Research" if deep else "Sonar" |
| 1126 | items: list[dict[str, Any]] = [ |
| 1127 | { |
| 1128 | "id": "PX1", |
| 1129 | "title": f"Perplexity {title_mode}: {query}", |
| 1130 | "url": "", |
| 1131 | "source_domain": "perplexity.ai", |
| 1132 | "snippet": synthesis[:2000], |
| 1133 | "date": to_date, |
| 1134 | "relevance": 0.9, |
| 1135 | "why_relevant": f"AI synthesis of recent activity for '{query}'", |
| 1136 | "engagement": {"citations": len(citations)}, |
| 1137 | "metadata": { |
| 1138 | "citations": citations, |
| 1139 | "usage": _usage(data), |
| 1140 | "openrouter_response_id": data.get("id"), |
| 1141 | }, |
| 1142 | } |
| 1143 | ] |
| 1144 | for index, citation in enumerate(citations): |
| 1145 | items.append( |
| 1146 | { |
| 1147 | "id": f"PX{index + 2}", |
| 1148 | "title": citation["title"] or _domain(citation["url"]), |
| 1149 | "url": citation["url"], |
| 1150 | "source_domain": _domain(citation["url"]), |
| 1151 | "snippet": citation.get("snippet") or "", |
| 1152 | "date": citation.get("date"), |
| 1153 | "relevance": 0.7, |
| 1154 | "why_relevant": f"Cited in Perplexity synthesis for '{query}'", |
| 1155 | "engagement": {"citations": 1}, |
| 1156 | "metadata": {"citations": [citation]}, |
| 1157 | } |
| 1158 | ) |
| 1159 | |
| 1160 | return items, { |
| 1161 | "label": "perplexity", |
| 1162 | "provider": "openrouter", |
| 1163 | "mode": PERPLEXITY_MODE_SONAR, |
| 1164 | "endpoint": "openrouter-chat-completions", |
| 1165 | "model": model, |
| 1166 | "deep": deep, |
| 1167 | "query": query, |
| 1168 | "synthesisLength": len(synthesis), |
| 1169 | "citationCount": len(citations), |
| 1170 | "responseId": data.get("id"), |
| 1171 | "servedModel": data.get("model") or model, |
| 1172 | "usage": _usage(data), |
| 1173 | } |
| 1174 | |
| 1175 | |
| 1176 | def _merge_agent_and_search( |
| 1177 | agent_items: list[dict[str, Any]], |
| 1178 | search_items: list[dict[str, Any]], |
| 1179 | ) -> list[dict[str, Any]]: |
| 1180 | if not agent_items: |
| 1181 | return search_items |
| 1182 | merged = agent_items[:1] |
| 1183 | seen_urls = {item.get("url") for item in merged if item.get("url")} |
| 1184 | for item in [*search_items, *agent_items[1:]]: |
| 1185 | url = item.get("url") |
| 1186 | if url and url in seen_urls: |
| 1187 | continue |
| 1188 | if url: |
| 1189 | seen_urls.add(url) |
| 1190 | merged.append(item) |
| 1191 | return merged |
| 1192 | |
| 1193 | |
| 1194 | def _top_level_failure( |
| 1195 | query: str, |
| 1196 | mode: str, |
| 1197 | deep: bool, |
| 1198 | exc: Exception, |
| 1199 | provider: str = "perplexity", |
| 1200 | requested_search_type: str | None = None, |
| 1201 | ) -> dict[str, Any]: |
| 1202 | artifact = _error_artifact(exc) |
| 1203 | if provider == "perplexity" and requested_search_type is not None: |
| 1204 | artifact["requested_search_type"] = requested_search_type |
| 1205 | if provider == "openrouter": |
| 1206 | endpoint = "openrouter-chat-completions" |
| 1207 | elif deep: |
| 1208 | endpoint = "agent-background" |
| 1209 | elif mode == PERPLEXITY_MODE_SEARCH: |
| 1210 | endpoint = "search" |
| 1211 | else: |
| 1212 | endpoint = "agent" |
| 1213 | artifact.update( |
| 1214 | { |
| 1215 | "label": "perplexity", |
| 1216 | "provider": provider, |
| 1217 | "mode": mode, |
| 1218 | "endpoint": endpoint, |
| 1219 | "deep": deep, |
| 1220 | "query": query, |
| 1221 | } |
| 1222 | ) |
| 1223 | return artifact |
| 1224 | |
| 1225 | |
| 1226 | def search( |
| 1227 | query: str, |
| 1228 | date_range: tuple[str, str], |
| 1229 | config: dict[str, Any], |
| 1230 | deep: bool = False, |
| 1231 | ) -> tuple[list[dict[str, Any]], dict[str, Any]]: |
| 1232 | """Search through Perplexity's Agent API or raw Search API. |
| 1233 | |
| 1234 | Normal synthesis uses the controlled Agent profile. Search API mode remains |
| 1235 | available for raw ranked rows. Deep Research is a dynamic high-preset |
| 1236 | background run and requires an explicit --deep-research invocation. |
| 1237 | """ |
| 1238 | resolved = _provider(config) |
| 1239 | if not resolved: |
| 1240 | _log( |
| 1241 | "No PERPLEXITY_API_KEY or OPENROUTER_API_KEY configured, skipping" |
| 1242 | ) |
| 1243 | return [], {} |
| 1244 | provider, api_key = resolved |
| 1245 | |
| 1246 | mode = _mode(config, deep, provider) |
| 1247 | search_type = None |
| 1248 | try: |
| 1249 | if provider == "openrouter": |
| 1250 | # Sonar ignores the direct-only setting, so never validate it here: |
| 1251 | # an unusable value must not block an otherwise working search. |
| 1252 | if _config_text(config, "LAST30DAYS_PERPLEXITY_SEARCH_TYPE"): |
| 1253 | _log("LAST30DAYS_PERPLEXITY_SEARCH_TYPE requires PERPLEXITY_API_KEY; " |
| 1254 | "it does not change OpenRouter Sonar search") |
| 1255 | result = _openrouter_sonar_search( |
| 1256 | query, |
| 1257 | date_range, |
| 1258 | api_key, |
| 1259 | deep, |
| 1260 | ) |
| 1261 | if deep: |
| 1262 | _log_deep_receipt(result[1]) |
| 1263 | return result |
| 1264 | search_type = _search_type(config) |
| 1265 | if mode == PERPLEXITY_MODE_SEARCH: |
| 1266 | return _search_api(query, date_range, config, api_key) |
| 1267 | if mode == PERPLEXITY_MODE_BOTH: |
| 1268 | search_items: list[dict[str, Any]] = [] |
| 1269 | agent_items: list[dict[str, Any]] = [] |
| 1270 | search_artifact: dict[str, Any] = {} |
| 1271 | agent_artifact: dict[str, Any] = {} |
| 1272 | try: |
| 1273 | search_items, search_artifact = _search_api( |
| 1274 | query, |
| 1275 | date_range, |
| 1276 | config, |
| 1277 | api_key, |
| 1278 | ) |
| 1279 | except Exception as exc: |
| 1280 | _log(f"Search API leg failed in both mode: {exc}") |
| 1281 | search_artifact = _error_artifact(exc) |
| 1282 | if search_type is not None: |
| 1283 | search_artifact["requested_search_type"] = search_type |
| 1284 | try: |
| 1285 | agent_items, agent_artifact = _agent_search( |
| 1286 | query, |
| 1287 | date_range, |
| 1288 | config, |
| 1289 | api_key, |
| 1290 | deep=False, |
| 1291 | ) |
| 1292 | except Exception as exc: |
| 1293 | _log(f"Agent API leg failed in both mode: {exc}") |
| 1294 | agent_artifact = _error_artifact(exc) |
| 1295 | if search_type is not None: |
| 1296 | agent_artifact["requested_search_type"] = search_type |
| 1297 | items = _merge_agent_and_search(agent_items, search_items) |
| 1298 | return items, { |
| 1299 | "label": "perplexity", |
| 1300 | "provider": "perplexity", |
| 1301 | "mode": PERPLEXITY_MODE_BOTH, |
| 1302 | "query": query, |
| 1303 | "search": search_artifact, |
| 1304 | "agent": agent_artifact, |
| 1305 | "itemCount": len(items), |
| 1306 | } |
| 1307 | result = _agent_search(query, date_range, config, api_key, deep) |
| 1308 | if deep: |
| 1309 | _log_deep_receipt(result[1]) |
| 1310 | return result |
| 1311 | except http.HTTPError as exc: |
| 1312 | if exc.status_code == 401: |
| 1313 | _log(f"Invalid {provider} API key (401)") |
| 1314 | elif exc.status_code == 429: |
| 1315 | _log(f"Rate limited by {provider} (429)") |
| 1316 | else: |
| 1317 | _log(f"HTTP error: {exc}") |
| 1318 | artifact = _top_level_failure( |
| 1319 | query, |
| 1320 | mode, |
| 1321 | deep, |
| 1322 | exc, |
| 1323 | provider=provider, |
| 1324 | requested_search_type=search_type, |
| 1325 | ) |
| 1326 | if deep: |
| 1327 | _log_deep_receipt(artifact) |
| 1328 | return [], artifact |
| 1329 | except TimeoutError as exc: |
| 1330 | _log(f"Request timed out: {exc}") |
| 1331 | artifact = _top_level_failure( |
| 1332 | query, |
| 1333 | mode, |
| 1334 | deep, |
| 1335 | exc, |
| 1336 | provider=provider, |
| 1337 | requested_search_type=search_type, |
| 1338 | ) |
| 1339 | if deep: |
| 1340 | _log_deep_receipt(artifact) |
| 1341 | return [], artifact |
| 1342 | except Exception as exc: |
| 1343 | _log(f"Request failed: {exc}") |
| 1344 | artifact = _top_level_failure( |
| 1345 | query, |
| 1346 | mode, |
| 1347 | deep, |
| 1348 | exc, |
| 1349 | provider=provider, |
| 1350 | requested_search_type=search_type, |
| 1351 | ) |
| 1352 | if deep: |
| 1353 | _log_deep_receipt(artifact) |
| 1354 | return [], artifact |
| 1355 |