| 1 | //! Narrow provider-native web-search client. |
| 2 | //! |
| 3 | //! This adapter reuses the active route's authenticated HTTP client without |
| 4 | //! exposing credentials to tool code. Route capability facts decide whether |
| 5 | //! the adapter is attached; this module only speaks the three documented |
| 6 | //! first-party wire contracts. |
| 7 | |
| 8 | use anyhow::{Context, Result, bail}; |
| 9 | use reqwest::header::{HeaderName, HeaderValue}; |
| 10 | use serde_json::{Value, json}; |
| 11 | |
| 12 | use crate::config::ApiProvider; |
| 13 | |
| 14 | use super::{CodewhaleClient, api_url, responses_api_url}; |
| 15 | |
| 16 | mod zai; |
| 17 | |
| 18 | mod kimi; |
| 19 | |
| 20 | const MAX_NATIVE_ANSWER_CHARS: usize = 4_000; |
| 21 | |
| 22 | #[derive(Clone)] |
| 23 | pub(crate) struct ProviderNativeSearchClient { |
| 24 | pub(super) inner: CodewhaleClient, |
| 25 | } |
| 26 | |
| 27 | #[derive(Clone)] |
| 28 | pub(crate) struct ProviderNativeSearchRequest { |
| 29 | pub(crate) query: String, |
| 30 | pub(crate) max_results: u8, |
| 31 | pub(crate) domains: Vec<String>, |
| 32 | } |
| 33 | |
| 34 | #[derive(Clone, PartialEq, Eq)] |
| 35 | pub(crate) struct ProviderNativeCitation { |
| 36 | pub(crate) url: String, |
| 37 | pub(crate) title: String, |
| 38 | pub(crate) snippet: Option<String>, |
| 39 | pub(crate) published: Option<String>, |
| 40 | } |
| 41 | |
| 42 | #[derive(Clone, PartialEq, Eq)] |
| 43 | pub(crate) struct ProviderNativeSearchResponse { |
| 44 | pub(crate) answer: Option<String>, |
| 45 | pub(crate) citations: Vec<ProviderNativeCitation>, |
| 46 | } |
| 47 | |
| 48 | impl ProviderNativeSearchClient { |
| 49 | #[must_use] |
| 50 | pub(crate) fn new(inner: CodewhaleClient) -> Option<Self> { |
| 51 | matches!( |
| 52 | inner.api_provider, |
| 53 | ApiProvider::Openai |
| 54 | | ApiProvider::Anthropic |
| 55 | | ApiProvider::Xai |
| 56 | | ApiProvider::XiaomiMimo |
| 57 | | ApiProvider::Zai |
| 58 | | ApiProvider::ModelstudioTokenPlan |
| 59 | | ApiProvider::Deepseek |
| 60 | | ApiProvider::DeepseekCN |
| 61 | | ApiProvider::Moonshot |
| 62 | ) |
| 63 | .then_some(Self { inner }) |
| 64 | } |
| 65 | |
| 66 | #[must_use] |
| 67 | pub(crate) fn provider(&self) -> ApiProvider { |
| 68 | self.inner.api_provider |
| 69 | } |
| 70 | |
| 71 | #[must_use] |
| 72 | pub(crate) fn model(&self) -> &str { |
| 73 | &self.inner.default_model |
| 74 | } |
| 75 | |
| 76 | #[must_use] |
| 77 | pub(crate) fn base_url(&self) -> &str { |
| 78 | &self.inner.base_url |
| 79 | } |
| 80 | |
| 81 | #[must_use] |
| 82 | pub(crate) fn host(&self) -> Option<String> { |
| 83 | reqwest::Url::parse(&self.inner.base_url) |
| 84 | .ok() |
| 85 | .and_then(|url| url.host_str().map(str::to_ascii_lowercase)) |
| 86 | } |
| 87 | |
| 88 | #[must_use] |
| 89 | pub(crate) fn cache_identity(&self) -> String { |
| 90 | format!( |
| 91 | "provider-native://{}/{}/{}", |
| 92 | self.inner.api_provider.as_str(), |
| 93 | self.host().as_deref().unwrap_or("unknown-host"), |
| 94 | self.inner.default_model |
| 95 | ) |
| 96 | } |
| 97 | |
| 98 | #[must_use] |
| 99 | pub(crate) const fn maximum_domain_count(&self) -> Option<usize> { |
| 100 | match self.inner.api_provider { |
| 101 | ApiProvider::Xai => Some(5), |
| 102 | ApiProvider::Openai => Some(100), |
| 103 | ApiProvider::Anthropic => None, |
| 104 | _ => Some(0), |
| 105 | } |
| 106 | } |
| 107 | |
| 108 | pub(crate) async fn search( |
| 109 | &self, |
| 110 | request: &ProviderNativeSearchRequest, |
| 111 | ) -> Result<ProviderNativeSearchResponse> { |
| 112 | // This adapter performs model-backed inference directly instead of |
| 113 | // calling `CodewhaleClient::create_message*`. It must therefore join |
| 114 | // the same attached-run ownership boundary explicitly. The guard is |
| 115 | // retained through response decode so a relay writer cannot start |
| 116 | // while this result is still able to feed the interactive turn. |
| 117 | let _inference = self.inner.acquire_remote_control_inference_permit().await; |
| 118 | if self.inner.api_provider == ApiProvider::Moonshot { |
| 119 | // Kimi/Moonshot runs a bounded multi-round agentic search with its |
| 120 | // own request/reply loop, so it cannot share the single-shot body |
| 121 | // dispatch below. It still runs under the inference permit above. |
| 122 | let mut parsed = kimi::search(self, request).await?; |
| 123 | parsed.citations.truncate(usize::from(request.max_results)); |
| 124 | return Ok(parsed); |
| 125 | } |
| 126 | let body = match self.inner.api_provider { |
| 127 | ApiProvider::Openai => build_responses_search_body( |
| 128 | &self.inner.default_model, |
| 129 | request, |
| 130 | ResponsesSearchDialect::Openai, |
| 131 | ), |
| 132 | ApiProvider::Xai => build_responses_search_body( |
| 133 | &self.inner.default_model, |
| 134 | request, |
| 135 | ResponsesSearchDialect::Xai, |
| 136 | ), |
| 137 | ApiProvider::ModelstudioTokenPlan => build_responses_search_body( |
| 138 | &self.inner.default_model, |
| 139 | request, |
| 140 | ResponsesSearchDialect::ModelStudio, |
| 141 | ), |
| 142 | ApiProvider::Deepseek | ApiProvider::DeepseekCN => build_responses_search_body( |
| 143 | &self.inner.default_model, |
| 144 | request, |
| 145 | ResponsesSearchDialect::Deepseek, |
| 146 | ), |
| 147 | ApiProvider::Anthropic => { |
| 148 | let route_cap = self |
| 149 | .inner |
| 150 | .effective_max_output_tokens(&self.inner.default_model); |
| 151 | build_anthropic_search_body( |
| 152 | &self.inner.default_model, |
| 153 | request, |
| 154 | 2_048_u32.min(route_cap), |
| 155 | ) |
| 156 | } |
| 157 | ApiProvider::XiaomiMimo => build_mimo_search_body(&self.inner.default_model, request), |
| 158 | ApiProvider::Zai => zai::build_body(request, &self.inner.base_url)?, |
| 159 | _ => bail!("active provider has no native web-search adapter"), |
| 160 | }; |
| 161 | let url = match self.inner.api_provider { |
| 162 | ApiProvider::Openai | ApiProvider::Xai | ApiProvider::ModelstudioTokenPlan => { |
| 163 | api_url(&self.inner.base_url, "responses") |
| 164 | } |
| 165 | ApiProvider::Deepseek | ApiProvider::DeepseekCN => { |
| 166 | responses_api_url(&self.inner.base_url, self.inner.api_provider) |
| 167 | } |
| 168 | ApiProvider::Anthropic => anthropic_messages_url(&self.inner.base_url), |
| 169 | ApiProvider::XiaomiMimo => api_url(&self.inner.base_url, "chat/completions"), |
| 170 | ApiProvider::Zai => api_url(&self.inner.base_url, "web_search"), |
| 171 | _ => unreachable!("provider checked above"), |
| 172 | }; |
| 173 | let body_bytes = serde_json::to_vec(&body) |
| 174 | .context("failed to serialize provider-native web-search request")?; |
| 175 | let response = self |
| 176 | .inner |
| 177 | .send_with_retry(|| { |
| 178 | self.inner |
| 179 | .http_client |
| 180 | .post(&url) |
| 181 | .header("Accept", "application/json") |
| 182 | .body(body_bytes.clone()) |
| 183 | }) |
| 184 | .await |
| 185 | .context("provider-native web search request failed")?; |
| 186 | let payload = response |
| 187 | .json::<Value>() |
| 188 | .await |
| 189 | .context("provider-native web search returned invalid JSON")?; |
| 190 | let mut parsed = match self.inner.api_provider { |
| 191 | ApiProvider::Openai |
| 192 | | ApiProvider::Xai |
| 193 | | ApiProvider::ModelstudioTokenPlan |
| 194 | | ApiProvider::Deepseek |
| 195 | | ApiProvider::DeepseekCN => parse_responses_search(&payload), |
| 196 | ApiProvider::Anthropic => parse_anthropic_search(&payload), |
| 197 | ApiProvider::XiaomiMimo => parse_mimo_search(&payload), |
| 198 | ApiProvider::Zai => zai::parse(&payload), |
| 199 | _ => unreachable!("provider checked above"), |
| 200 | }; |
| 201 | parsed.citations.truncate(usize::from(request.max_results)); |
| 202 | Ok(parsed) |
| 203 | } |
| 204 | |
| 205 | pub(super) async fn post_json( |
| 206 | &self, |
| 207 | url: &str, |
| 208 | body: &Value, |
| 209 | headers: &[(HeaderName, HeaderValue)], |
| 210 | ) -> Result<Value> { |
| 211 | let body_bytes = serde_json::to_vec(&body) |
| 212 | .context("failed to serialize provider-native web-search request")?; |
| 213 | let headers = headers.to_vec(); |
| 214 | let response = self |
| 215 | .inner |
| 216 | .send_with_retry(|| { |
| 217 | let mut request = self |
| 218 | .inner |
| 219 | .http_client |
| 220 | .post(url) |
| 221 | .header("Accept", "application/json") |
| 222 | .body(body_bytes.clone()); |
| 223 | for (name, value) in &headers { |
| 224 | request = request.header(name, value); |
| 225 | } |
| 226 | request |
| 227 | }) |
| 228 | .await |
| 229 | .context("provider-native web search request failed")?; |
| 230 | response |
| 231 | .json::<Value>() |
| 232 | .await |
| 233 | .context("provider-native web search returned invalid JSON") |
| 234 | } |
| 235 | |
| 236 | pub(super) async fn get_json(&self, url: &str) -> Result<Value> { |
| 237 | let response = self |
| 238 | .inner |
| 239 | .send_with_retry(|| { |
| 240 | self.inner |
| 241 | .http_client |
| 242 | .get(url) |
| 243 | .header("Accept", "application/json") |
| 244 | }) |
| 245 | .await |
| 246 | .context("provider-native web search request failed")?; |
| 247 | response |
| 248 | .json::<Value>() |
| 249 | .await |
| 250 | .context("provider-native web search returned invalid JSON") |
| 251 | } |
| 252 | } |
| 253 | |
| 254 | #[derive(Clone, Copy)] |
| 255 | enum ResponsesSearchDialect { |
| 256 | Openai, |
| 257 | Xai, |
| 258 | ModelStudio, |
| 259 | Deepseek, |
| 260 | } |
| 261 | |
| 262 | fn search_prompt(request: &ProviderNativeSearchRequest) -> String { |
| 263 | format!( |
| 264 | "Search the web for the following query and answer only from web sources. \ |
| 265 | Use concise prose with citations and prefer at most {} distinct sources.\n\n{}", |
| 266 | request.max_results, request.query |
| 267 | ) |
| 268 | } |
| 269 | |
| 270 | fn build_responses_search_body( |
| 271 | model: &str, |
| 272 | request: &ProviderNativeSearchRequest, |
| 273 | dialect: ResponsesSearchDialect, |
| 274 | ) -> Value { |
| 275 | let mut tool = json!({ "type": "web_search" }); |
| 276 | if !request.domains.is_empty() |
| 277 | && matches!( |
| 278 | dialect, |
| 279 | ResponsesSearchDialect::Openai | ResponsesSearchDialect::Xai |
| 280 | ) |
| 281 | { |
| 282 | tool["filters"] = json!({ "allowed_domains": request.domains }); |
| 283 | } |
| 284 | let mut body = json!({ |
| 285 | "model": model, |
| 286 | "input": search_prompt(request), |
| 287 | "tools": [tool], |
| 288 | }); |
| 289 | match dialect { |
| 290 | ResponsesSearchDialect::Openai => { |
| 291 | body["tool_choice"] = json!("required"); |
| 292 | body["store"] = json!(false); |
| 293 | body["include"] = json!(["web_search_call.action.sources"]); |
| 294 | } |
| 295 | ResponsesSearchDialect::Xai => { |
| 296 | body["tool_choice"] = json!("required"); |
| 297 | } |
| 298 | ResponsesSearchDialect::ModelStudio => { |
| 299 | body["tool_choice"] = json!("required"); |
| 300 | } |
| 301 | ResponsesSearchDialect::Deepseek => { |
| 302 | body["tool_choice"] = json!({ "type": "web_search" }); |
| 303 | } |
| 304 | } |
| 305 | body |
| 306 | } |
| 307 | |
| 308 | fn build_anthropic_search_body( |
| 309 | model: &str, |
| 310 | request: &ProviderNativeSearchRequest, |
| 311 | max_tokens: u32, |
| 312 | ) -> Value { |
| 313 | let mut tool = json!({ |
| 314 | "type": "web_search_20250305", |
| 315 | "name": "web_search", |
| 316 | "max_uses": 1, |
| 317 | }); |
| 318 | if !request.domains.is_empty() { |
| 319 | tool["allowed_domains"] = json!(request.domains); |
| 320 | } |
| 321 | json!({ |
| 322 | "model": model, |
| 323 | "max_tokens": max_tokens, |
| 324 | "messages": [{ "role": "user", "content": search_prompt(request) }], |
| 325 | "tools": [tool], |
| 326 | }) |
| 327 | } |
| 328 | |
| 329 | fn build_mimo_search_body(model: &str, request: &ProviderNativeSearchRequest) -> Value { |
| 330 | json!({ |
| 331 | "model": model, |
| 332 | "messages": [{ "role": "user", "content": search_prompt(request) }], |
| 333 | "tools": [{ |
| 334 | "type": "web_search", |
| 335 | "max_keyword": 1, |
| 336 | "force_search": true, |
| 337 | "limit": request.max_results, |
| 338 | }], |
| 339 | "tool_choice": "auto", |
| 340 | "max_completion_tokens": 2_048, |
| 341 | "stream": false, |
| 342 | "thinking": { "type": "disabled" }, |
| 343 | }) |
| 344 | } |
| 345 | |
| 346 | fn anthropic_messages_url(base_url: &str) -> String { |
| 347 | let base = base_url.trim_end_matches('/'); |
| 348 | if base.ends_with("/v1") { |
| 349 | format!("{base}/messages") |
| 350 | } else { |
| 351 | format!("{base}/v1/messages") |
| 352 | } |
| 353 | } |
| 354 | |
| 355 | fn parse_responses_search(payload: &Value) -> ProviderNativeSearchResponse { |
| 356 | let mut answer_parts = Vec::new(); |
| 357 | let mut citations = Vec::new(); |
| 358 | if let Some(output) = payload.get("output").and_then(Value::as_array) { |
| 359 | for item in output { |
| 360 | let item_type = item.get("type").and_then(Value::as_str); |
| 361 | if item_type == Some("web_search_call") |
| 362 | && let Some(action) = item.get("action") |
| 363 | { |
| 364 | if let Some(sources) = action.get("sources").and_then(Value::as_array) { |
| 365 | for source in sources { |
| 366 | push_citation(&mut citations, citation_from_value(source, None, None)); |
| 367 | } |
| 368 | } |
| 369 | push_citation(&mut citations, citation_from_value(action, None, None)); |
| 370 | } |
| 371 | |
| 372 | if item_type == Some("message") |
| 373 | && let Some(content) = item.get("content").and_then(Value::as_array) |
| 374 | { |
| 375 | for block in content { |
| 376 | if matches!( |
| 377 | block.get("type").and_then(Value::as_str), |
| 378 | Some("output_text" | "text") |
| 379 | ) && let Some(text) = block.get("text").and_then(Value::as_str) |
| 380 | && !text.trim().is_empty() |
| 381 | { |
| 382 | answer_parts.push(text.trim().to_string()); |
| 383 | } |
| 384 | if let Some(annotations) = block.get("annotations").and_then(Value::as_array) { |
| 385 | for annotation in annotations { |
| 386 | push_citation( |
| 387 | &mut citations, |
| 388 | citation_from_value(annotation, None, None), |
| 389 | ); |
| 390 | } |
| 391 | } |
| 392 | } |
| 393 | } |
| 394 | } |
| 395 | } |
| 396 | if answer_parts.is_empty() |
| 397 | && let Some(output_text) = payload.get("output_text").and_then(Value::as_str) |
| 398 | && !output_text.trim().is_empty() |
| 399 | { |
| 400 | answer_parts.push(output_text.trim().to_string()); |
| 401 | } |
| 402 | for answer in &answer_parts { |
| 403 | for citation in citations_from_text(answer) { |
| 404 | push_citation(&mut citations, Some(citation)); |
| 405 | } |
| 406 | } |
| 407 | if let Some(top_level) = payload.get("citations").and_then(Value::as_array) { |
| 408 | for citation in top_level { |
| 409 | let parsed = citation |
| 410 | .as_str() |
| 411 | .and_then(|url| citation_from_url(url, None, None, None)) |
| 412 | .or_else(|| citation_from_value(citation, None, None)); |
| 413 | push_citation(&mut citations, parsed); |
| 414 | } |
| 415 | } |
| 416 | ProviderNativeSearchResponse { |
| 417 | answer: bounded_answer(answer_parts), |
| 418 | citations, |
| 419 | } |
| 420 | } |
| 421 | |
| 422 | fn parse_anthropic_search(payload: &Value) -> ProviderNativeSearchResponse { |
| 423 | let mut answer_parts = Vec::new(); |
| 424 | let mut citations = Vec::new(); |
| 425 | if let Some(content) = payload.get("content").and_then(Value::as_array) { |
| 426 | for block in content { |
| 427 | match block.get("type").and_then(Value::as_str) { |
| 428 | Some("web_search_tool_result") => { |
| 429 | if let Some(results) = block.get("content").and_then(Value::as_array) { |
| 430 | for result in results { |
| 431 | let published = result |
| 432 | .get("page_age") |
| 433 | .and_then(Value::as_str) |
| 434 | .map(str::to_string); |
| 435 | push_citation( |
| 436 | &mut citations, |
| 437 | citation_from_value(result, None, published), |
| 438 | ); |
| 439 | } |
| 440 | } |
| 441 | } |
| 442 | Some("text") => { |
| 443 | if let Some(text) = block.get("text").and_then(Value::as_str) |
| 444 | && !text.trim().is_empty() |
| 445 | { |
| 446 | answer_parts.push(text.trim().to_string()); |
| 447 | } |
| 448 | if let Some(block_citations) = block.get("citations").and_then(Value::as_array) |
| 449 | { |
| 450 | for citation in block_citations { |
| 451 | let snippet = citation |
| 452 | .get("cited_text") |
| 453 | .and_then(Value::as_str) |
| 454 | .map(str::to_string); |
| 455 | push_citation( |
| 456 | &mut citations, |
| 457 | citation_from_value(citation, snippet, None), |
| 458 | ); |
| 459 | } |
| 460 | } |
| 461 | } |
| 462 | _ => {} |
| 463 | } |
| 464 | } |
| 465 | } |
| 466 | ProviderNativeSearchResponse { |
| 467 | answer: bounded_answer(answer_parts), |
| 468 | citations, |
| 469 | } |
| 470 | } |
| 471 | |
| 472 | fn parse_mimo_search(payload: &Value) -> ProviderNativeSearchResponse { |
| 473 | let message = payload.pointer("/choices/0/message"); |
| 474 | let answer = message |
| 475 | .and_then(|value| value.get("content")) |
| 476 | .and_then(Value::as_str) |
| 477 | .map(str::trim) |
| 478 | .filter(|text| !text.is_empty()) |
| 479 | .map(str::to_string); |
| 480 | let mut citations = Vec::new(); |
| 481 | if let Some(annotations) = message |
| 482 | .and_then(|value| value.get("annotations")) |
| 483 | .and_then(Value::as_array) |
| 484 | .or_else(|| payload.get("annotations").and_then(Value::as_array)) |
| 485 | { |
| 486 | for annotation in annotations { |
| 487 | let Some(url) = annotation.get("url").and_then(Value::as_str) else { |
| 488 | continue; |
| 489 | }; |
| 490 | let title = annotation |
| 491 | .get("title") |
| 492 | .and_then(Value::as_str) |
| 493 | .map(str::to_string); |
| 494 | let snippet = annotation |
| 495 | .get("summary") |
| 496 | .and_then(Value::as_str) |
| 497 | .map(str::to_string); |
| 498 | let published = annotation |
| 499 | .get("publish_time") |
| 500 | .and_then(Value::as_str) |
| 501 | .map(str::to_string); |
| 502 | push_citation( |
| 503 | &mut citations, |
| 504 | citation_from_url(url, title, snippet, published), |
| 505 | ); |
| 506 | } |
| 507 | } |
| 508 | ProviderNativeSearchResponse { |
| 509 | answer: bounded_answer(answer.into_iter().collect()), |
| 510 | citations, |
| 511 | } |
| 512 | } |
| 513 | |
| 514 | fn citation_from_value( |
| 515 | value: &Value, |
| 516 | snippet: Option<String>, |
| 517 | published: Option<String>, |
| 518 | ) -> Option<ProviderNativeCitation> { |
| 519 | let url = value.get("url").and_then(Value::as_str)?.trim(); |
| 520 | let title = value |
| 521 | .get("title") |
| 522 | .and_then(Value::as_str) |
| 523 | .map(str::trim) |
| 524 | .filter(|title| !title.is_empty()) |
| 525 | .map(str::to_string); |
| 526 | citation_from_url(url, title, snippet, published) |
| 527 | } |
| 528 | |
| 529 | fn citation_from_url( |
| 530 | url: &str, |
| 531 | title: Option<String>, |
| 532 | snippet: Option<String>, |
| 533 | published: Option<String>, |
| 534 | ) -> Option<ProviderNativeCitation> { |
| 535 | let parsed = reqwest::Url::parse(url).ok()?; |
| 536 | if !matches!(parsed.scheme(), "http" | "https") || parsed.host_str().is_none() { |
| 537 | return None; |
| 538 | } |
| 539 | Some(ProviderNativeCitation { |
| 540 | url: url.to_string(), |
| 541 | title: title.unwrap_or_else(|| fallback_title(url)), |
| 542 | snippet, |
| 543 | published, |
| 544 | }) |
| 545 | } |
| 546 | |
| 547 | fn push_citation( |
| 548 | citations: &mut Vec<ProviderNativeCitation>, |
| 549 | candidate: Option<ProviderNativeCitation>, |
| 550 | ) { |
| 551 | let Some(candidate) = candidate else { |
| 552 | return; |
| 553 | }; |
| 554 | if let Some(existing) = citations |
| 555 | .iter_mut() |
| 556 | .find(|existing| existing.url == candidate.url) |
| 557 | { |
| 558 | if existing.title == fallback_title(&existing.url) |
| 559 | && candidate.title != fallback_title(&candidate.url) |
| 560 | { |
| 561 | existing.title = candidate.title; |
| 562 | } |
| 563 | if existing.snippet.is_none() { |
| 564 | existing.snippet = candidate.snippet; |
| 565 | } |
| 566 | if existing.published.is_none() { |
| 567 | existing.published = candidate.published; |
| 568 | } |
| 569 | return; |
| 570 | } |
| 571 | citations.push(candidate); |
| 572 | } |
| 573 | |
| 574 | fn citations_from_text(text: &str) -> Vec<ProviderNativeCitation> { |
| 575 | let mut citations = Vec::new(); |
| 576 | let mut offset = 0; |
| 577 | while offset < text.len() { |
| 578 | let remaining = &text[offset..]; |
| 579 | let relative_start = match (remaining.find("https://"), remaining.find("http://")) { |
| 580 | (Some(https), Some(http)) => Some(https.min(http)), |
| 581 | (Some(https), None) => Some(https), |
| 582 | (None, Some(http)) => Some(http), |
| 583 | (None, None) => None, |
| 584 | }; |
| 585 | let Some(relative_start) = relative_start else { |
| 586 | break; |
| 587 | }; |
| 588 | let start = offset + relative_start; |
| 589 | let tail = &text[start..]; |
| 590 | // Balanced parentheses belong to the URL (Wikipedia titles such as |
| 591 | // `Foo_(bar)` keep their closing paren); an unmatched closer ends it. |
| 592 | let mut open_parens = 0_usize; |
| 593 | let end = tail |
| 594 | .char_indices() |
| 595 | .find_map(|(index, ch)| match ch { |
| 596 | '(' => { |
| 597 | open_parens += 1; |
| 598 | None |
| 599 | } |
| 600 | ')' if open_parens > 0 => { |
| 601 | open_parens -= 1; |
| 602 | None |
| 603 | } |
| 604 | _ => (index > 0 |
| 605 | && (ch.is_whitespace() |
| 606 | || matches!(ch, ')' | ']' | '}' | '>' | '"' | '\'' | '`'))) |
| 607 | .then_some(index), |
| 608 | }) |
| 609 | .unwrap_or(tail.len()); |
| 610 | let url = tail[..end].trim_end_matches(['.', ',', ';', ':', '!', '?']); |
| 611 | push_citation(&mut citations, citation_from_url(url, None, None, None)); |
| 612 | offset = start + end.max(1); |
| 613 | } |
| 614 | citations |
| 615 | } |
| 616 | |
| 617 | fn fallback_title(url: &str) -> String { |
| 618 | reqwest::Url::parse(url) |
| 619 | .ok() |
| 620 | .and_then(|parsed| parsed.host_str().map(str::to_string)) |
| 621 | .unwrap_or_else(|| "Web source".to_string()) |
| 622 | } |
| 623 | |
| 624 | fn bounded_answer(parts: Vec<String>) -> Option<String> { |
| 625 | let joined = parts.join("\n\n"); |
| 626 | let trimmed = joined.trim(); |
| 627 | if trimmed.is_empty() { |
| 628 | return None; |
| 629 | } |
| 630 | if trimmed.chars().count() <= MAX_NATIVE_ANSWER_CHARS { |
| 631 | return Some(trimmed.to_string()); |
| 632 | } |
| 633 | let mut bounded = trimmed |
| 634 | .chars() |
| 635 | .take(MAX_NATIVE_ANSWER_CHARS.saturating_sub(1)) |
| 636 | .collect::<String>(); |
| 637 | bounded.push('…'); |
| 638 | Some(bounded) |
| 639 | } |
| 640 | |
| 641 | #[cfg(test)] |
| 642 | mod tests { |
| 643 | use super::*; |
| 644 | use crate::config::{Config, ProviderConfig, ProvidersConfig}; |
| 645 | use wiremock::matchers::{body_partial_json, header, method, path}; |
| 646 | use wiremock::{Mock, MockServer, ResponseTemplate}; |
| 647 | |
| 648 | fn request() -> ProviderNativeSearchRequest { |
| 649 | ProviderNativeSearchRequest { |
| 650 | query: "current release".to_string(), |
| 651 | max_results: 3, |
| 652 | domains: vec!["example.com".to_string()], |
| 653 | } |
| 654 | } |
| 655 | |
| 656 | fn xai_client_with_boundary( |
| 657 | server: &MockServer, |
| 658 | isolated: bool, |
| 659 | unrelated: bool, |
| 660 | ) -> ProviderNativeSearchClient { |
| 661 | let config = Config { |
| 662 | provider: Some("xai".to_string()), |
| 663 | providers: Some(ProvidersConfig { |
| 664 | xai: ProviderConfig { |
| 665 | api_key: Some("xai-test-key".to_string()), |
| 666 | base_url: Some(format!("{}/v1", server.uri())), |
| 667 | model: Some("grok-4.5".to_string()), |
| 668 | ..ProviderConfig::default() |
| 669 | }, |
| 670 | ..ProvidersConfig::default() |
| 671 | }), |
| 672 | runtime_chat_isolated: isolated, |
| 673 | runtime_thread_inference_unrelated: unrelated, |
| 674 | ..Config::default() |
| 675 | }; |
| 676 | ProviderNativeSearchClient::new(CodewhaleClient::new(&config).expect("test xAI client")) |
| 677 | .expect("xAI native adapter") |
| 678 | } |
| 679 | |
| 680 | #[test] |
| 681 | fn responses_payload_requires_search_and_keeps_domains_provider_side() { |
| 682 | let body = |
| 683 | build_responses_search_body("gpt-5.6", &request(), ResponsesSearchDialect::Openai); |
| 684 | assert_eq!(body["tools"][0]["type"], "web_search"); |
| 685 | assert_eq!( |
| 686 | body["tools"][0]["filters"]["allowed_domains"][0], |
| 687 | "example.com" |
| 688 | ); |
| 689 | assert_eq!(body["tool_choice"], "required"); |
| 690 | assert_eq!(body["include"][0], "web_search_call.action.sources"); |
| 691 | } |
| 692 | |
| 693 | #[test] |
| 694 | fn modelstudio_payload_uses_required_harness_search_without_filters() { |
| 695 | let body = build_responses_search_body( |
| 696 | "qwen3.8-max", |
| 697 | &request(), |
| 698 | ResponsesSearchDialect::ModelStudio, |
| 699 | ); |
| 700 | assert_eq!(body["tools"][0]["type"], "web_search"); |
| 701 | assert!(body["tools"][0].get("filters").is_none()); |
| 702 | assert_eq!(body["tool_choice"], "required"); |
| 703 | assert!(body.get("include").is_none()); |
| 704 | assert!(body.get("store").is_none()); |
| 705 | } |
| 706 | |
| 707 | #[test] |
| 708 | fn deepseek_payload_uses_its_responses_search_contract() { |
| 709 | let body = build_responses_search_body( |
| 710 | "deepseek-v4-flash", |
| 711 | &request(), |
| 712 | ResponsesSearchDialect::Deepseek, |
| 713 | ); |
| 714 | assert_eq!(body["tools"][0]["type"], "web_search"); |
| 715 | assert!(body["tools"][0].get("filters").is_none()); |
| 716 | assert_eq!(body["tool_choice"]["type"], "web_search"); |
| 717 | assert!(body.get("include").is_none()); |
| 718 | assert!(body.get("store").is_none()); |
| 719 | } |
| 720 | |
| 721 | #[test] |
| 722 | fn anthropic_payload_uses_basic_direct_search_contract() { |
| 723 | let body = build_anthropic_search_body("claude-opus-4-8", &request(), 2_048); |
| 724 | assert_eq!(body["tools"][0]["type"], "web_search_20250305"); |
| 725 | assert_eq!(body["tools"][0]["max_uses"], 1); |
| 726 | assert_eq!(body["tools"][0]["allowed_domains"][0], "example.com"); |
| 727 | assert_eq!(body["max_tokens"], 2_048); |
| 728 | |
| 729 | let tiny_route = build_anthropic_search_body("claude-opus-4-8", &request(), 128); |
| 730 | assert_eq!(tiny_route["max_tokens"], 128); |
| 731 | } |
| 732 | |
| 733 | #[test] |
| 734 | fn mimo_payload_forces_bounded_web_search_plugin() { |
| 735 | let body = build_mimo_search_body("mimo-v2.5-pro", &request()); |
| 736 | assert_eq!(body["tools"][0]["type"], "web_search"); |
| 737 | assert_eq!(body["tools"][0]["force_search"], true); |
| 738 | assert_eq!(body["tools"][0]["limit"], 3); |
| 739 | assert_eq!(body["max_completion_tokens"], 2_048); |
| 740 | assert_eq!(body["thinking"]["type"], "disabled"); |
| 741 | } |
| 742 | |
| 743 | #[test] |
| 744 | fn responses_parser_separates_answer_and_deduplicated_citations() { |
| 745 | let payload = json!({ |
| 746 | "output": [ |
| 747 | { |
| 748 | "type": "web_search_call", |
| 749 | "action": { "sources": [ |
| 750 | { "url": "https://example.com/a", "title": "Source A" } |
| 751 | ] } |
| 752 | }, |
| 753 | { |
| 754 | "type": "message", |
| 755 | "content": [{ |
| 756 | "type": "output_text", |
| 757 | "text": "Grounded answer.", |
| 758 | "annotations": [ |
| 759 | { "type": "url_citation", "url": "https://example.com/a", "title": "Source A" }, |
| 760 | { "type": "url_citation", "url": "https://example.org/b", "title": "Source B" } |
| 761 | ] |
| 762 | }] |
| 763 | } |
| 764 | ] |
| 765 | }); |
| 766 | let parsed = parse_responses_search(&payload); |
| 767 | assert_eq!(parsed.answer.as_deref(), Some("Grounded answer.")); |
| 768 | assert_eq!(parsed.citations.len(), 2); |
| 769 | assert_eq!(parsed.citations[0].title, "Source A"); |
| 770 | assert_eq!(parsed.citations[1].url, "https://example.org/b"); |
| 771 | } |
| 772 | |
| 773 | #[test] |
| 774 | fn responses_parser_keeps_final_message_and_opened_pages_only() { |
| 775 | let payload = json!({ |
| 776 | "output": [ |
| 777 | { |
| 778 | "type": "reasoning", |
| 779 | "content": [{ |
| 780 | "type": "reasoning_text", |
| 781 | "text": "private analysis https://reasoning.example/ must stay hidden" |
| 782 | }] |
| 783 | }, |
| 784 | { |
| 785 | "type": "web_search_call", |
| 786 | "action": { |
| 787 | "type": "open_page", |
| 788 | "url": "https://github.com/Hmbown/CodeWhale" |
| 789 | } |
| 790 | }, |
| 791 | { |
| 792 | "type": "message", |
| 793 | "content": [{ |
| 794 | "type": "output_text", |
| 795 | "text": "Official repository: https://github.com/Hmbown/CodeWhale", |
| 796 | "annotations": [] |
| 797 | }] |
| 798 | } |
| 799 | ] |
| 800 | }); |
| 801 | |
| 802 | let parsed = parse_responses_search(&payload); |
| 803 | |
| 804 | assert_eq!( |
| 805 | parsed.answer.as_deref(), |
| 806 | Some("Official repository: https://github.com/Hmbown/CodeWhale") |
| 807 | ); |
| 808 | assert_eq!(parsed.citations.len(), 1); |
| 809 | assert_eq!( |
| 810 | parsed.citations[0].url, |
| 811 | "https://github.com/Hmbown/CodeWhale" |
| 812 | ); |
| 813 | } |
| 814 | |
| 815 | #[test] |
| 816 | fn anthropic_parser_keeps_result_metadata_and_cited_text_separate() { |
| 817 | let payload = json!({ |
| 818 | "content": [ |
| 819 | { |
| 820 | "type": "web_search_tool_result", |
| 821 | "content": [{ |
| 822 | "type": "web_search_result", |
| 823 | "url": "https://example.com/a", |
| 824 | "title": "Source A", |
| 825 | "page_age": "July 18, 2026" |
| 826 | }] |
| 827 | }, |
| 828 | { |
| 829 | "type": "text", |
| 830 | "text": "Grounded answer.", |
| 831 | "citations": [{ |
| 832 | "type": "web_search_result_location", |
| 833 | "url": "https://example.com/a", |
| 834 | "title": "Source A", |
| 835 | "cited_text": "Supporting passage" |
| 836 | }] |
| 837 | } |
| 838 | ] |
| 839 | }); |
| 840 | let parsed = parse_anthropic_search(&payload); |
| 841 | assert_eq!(parsed.answer.as_deref(), Some("Grounded answer.")); |
| 842 | assert_eq!(parsed.citations.len(), 1); |
| 843 | assert_eq!( |
| 844 | parsed.citations[0].published.as_deref(), |
| 845 | Some("July 18, 2026") |
| 846 | ); |
| 847 | assert_eq!( |
| 848 | parsed.citations[0].snippet.as_deref(), |
| 849 | Some("Supporting passage") |
| 850 | ); |
| 851 | } |
| 852 | |
| 853 | #[test] |
| 854 | fn mimo_parser_keeps_non_streaming_annotations() { |
| 855 | let parsed = parse_mimo_search(&json!({ |
| 856 | "choices": [{ |
| 857 | "message": { |
| 858 | "content": "Grounded answer.", |
| 859 | "annotations": [{ |
| 860 | "type": "url_citation", |
| 861 | "url": "https://example.com/weather", |
| 862 | "title": "Weather", |
| 863 | "summary": "Forecast", |
| 864 | "publish_time": "2026-08-28" |
| 865 | }] |
| 866 | } |
| 867 | }] |
| 868 | })); |
| 869 | assert_eq!(parsed.answer.as_deref(), Some("Grounded answer.")); |
| 870 | assert_eq!(parsed.citations.len(), 1); |
| 871 | assert_eq!(parsed.citations[0].snippet.as_deref(), Some("Forecast")); |
| 872 | assert_eq!(parsed.citations[0].published.as_deref(), Some("2026-08-28")); |
| 873 | } |
| 874 | |
| 875 | #[test] |
| 876 | fn non_http_citations_are_rejected() { |
| 877 | let payload = json!({ "citations": ["javascript:alert(1)"] }); |
| 878 | assert!(parse_responses_search(&payload).citations.is_empty()); |
| 879 | } |
| 880 | |
| 881 | #[test] |
| 882 | fn answer_links_preserve_mixed_scheme_source_order() { |
| 883 | let citations = |
| 884 | citations_from_text("First http://legacy.example/a, then https://secure.example/b."); |
| 885 | assert_eq!(citations.len(), 2); |
| 886 | assert_eq!(citations[0].url, "http://legacy.example/a"); |
| 887 | assert_eq!(citations[1].url, "https://secure.example/b"); |
| 888 | } |
| 889 | |
| 890 | #[test] |
| 891 | fn answer_links_keep_balanced_parenthesis_segments() { |
| 892 | let citations = citations_from_text( |
| 893 | "See https://en.wikipedia.org/wiki/Foo_(bar) and https://en.wikipedia.org/wiki/Baz_(qux_(nested)) for details.", |
| 894 | ); |
| 895 | assert_eq!(citations.len(), 2); |
| 896 | assert_eq!( |
| 897 | citations[0].url, "https://en.wikipedia.org/wiki/Foo_(bar)", |
| 898 | "a balanced closing paren is part of the URL" |
| 899 | ); |
| 900 | assert_eq!( |
| 901 | citations[1].url, "https://en.wikipedia.org/wiki/Baz_(qux_(nested))", |
| 902 | "nested balanced parens stay intact" |
| 903 | ); |
| 904 | |
| 905 | let unbalanced = citations_from_text("Broken https://en.wikipedia.org/wiki/Foo_(bar here."); |
| 906 | assert_eq!( |
| 907 | unbalanced[0].url, "https://en.wikipedia.org/wiki/Foo_(bar", |
| 908 | "an unclosed paren cannot extend past the next whitespace" |
| 909 | ); |
| 910 | } |
| 911 | |
| 912 | #[tokio::test] |
| 913 | async fn xai_adapter_reuses_active_authenticated_transport() { |
| 914 | let server = MockServer::start().await; |
| 915 | Mock::given(method("POST")) |
| 916 | .and(path("/v1/responses")) |
| 917 | .and(header("authorization", "Bearer xai-test-key")) |
| 918 | .and(body_partial_json(json!({ |
| 919 | "model": "grok-4.5", |
| 920 | "tools": [{ |
| 921 | "type": "web_search", |
| 922 | "filters": { "allowed_domains": ["example.com"] } |
| 923 | }], |
| 924 | "tool_choice": "required" |
| 925 | }))) |
| 926 | .respond_with(ResponseTemplate::new(200).set_body_json(json!({ |
| 927 | "output_text": "Grounded answer.", |
| 928 | "citations": ["https://example.com/source"] |
| 929 | }))) |
| 930 | .expect(1) |
| 931 | .mount(&server) |
| 932 | .await; |
| 933 | let config = Config { |
| 934 | provider: Some("xai".to_string()), |
| 935 | providers: Some(ProvidersConfig { |
| 936 | xai: ProviderConfig { |
| 937 | api_key: Some("xai-test-key".to_string()), |
| 938 | base_url: Some(format!("{}/v1", server.uri())), |
| 939 | model: Some("grok-4.5".to_string()), |
| 940 | ..ProviderConfig::default() |
| 941 | }, |
| 942 | ..ProvidersConfig::default() |
| 943 | }), |
| 944 | ..Config::default() |
| 945 | }; |
| 946 | let inner = CodewhaleClient::new(&config).expect("test xAI client"); |
| 947 | let client = ProviderNativeSearchClient::new(inner).expect("xAI native adapter"); |
| 948 | let cache_identity = client.cache_identity(); |
| 949 | assert!(cache_identity.contains("provider-native://xai/")); |
| 950 | assert!(cache_identity.ends_with("/grok-4.5")); |
| 951 | assert!(!cache_identity.contains("xai-test-key")); |
| 952 | |
| 953 | let response = client.search(&request()).await.expect("native search"); |
| 954 | |
| 955 | assert_eq!(response.answer.as_deref(), Some("Grounded answer.")); |
| 956 | assert_eq!(response.citations.len(), 1); |
| 957 | assert_eq!(response.citations[0].url, "https://example.com/source"); |
| 958 | } |
| 959 | |
| 960 | #[tokio::test] |
| 961 | async fn modelstudio_adapter_uses_token_plan_responses_contract() { |
| 962 | let server = MockServer::start().await; |
| 963 | Mock::given(method("POST")) |
| 964 | .and(path("/v1/responses")) |
| 965 | .and(header("authorization", "Bearer modelstudio-test-key")) |
| 966 | .and(body_partial_json(json!({ |
| 967 | "model": "qwen3.8-max", |
| 968 | "tools": [{ "type": "web_search" }], |
| 969 | "tool_choice": "required" |
| 970 | }))) |
| 971 | .respond_with(ResponseTemplate::new(200).set_body_json(json!({ |
| 972 | "output": [{ |
| 973 | "type": "web_search_call", |
| 974 | "action": { |
| 975 | "sources": [{ |
| 976 | "url": "https://example.com/qwen", |
| 977 | "title": "Qwen source" |
| 978 | }] |
| 979 | } |
| 980 | }] |
| 981 | }))) |
| 982 | .expect(1) |
| 983 | .mount(&server) |
| 984 | .await; |
| 985 | let config = Config { |
| 986 | provider: Some("modelstudio-token-plan".to_string()), |
| 987 | providers: Some(ProvidersConfig { |
| 988 | modelstudio_token_plan: ProviderConfig { |
| 989 | api_key: Some("modelstudio-test-key".to_string()), |
| 990 | base_url: Some(format!("{}/v1", server.uri())), |
| 991 | model: Some("qwen3.8-max".to_string()), |
| 992 | ..ProviderConfig::default() |
| 993 | }, |
| 994 | ..ProvidersConfig::default() |
| 995 | }), |
| 996 | ..Config::default() |
| 997 | }; |
| 998 | let inner = CodewhaleClient::new(&config).expect("test ModelStudio client"); |
| 999 | let client = ProviderNativeSearchClient::new(inner).expect("Qwen native adapter"); |
| 1000 | |
| 1001 | let response = client.search(&request()).await.expect("native search"); |
| 1002 | |
| 1003 | assert_eq!(response.citations.len(), 1); |
| 1004 | assert_eq!(response.citations[0].url, "https://example.com/qwen"); |
| 1005 | } |
| 1006 | #[tokio::test] |
| 1007 | async fn deepseek_adapter_uses_authenticated_responses_endpoint() { |
| 1008 | let server = MockServer::start().await; |
| 1009 | Mock::given(method("POST")) |
| 1010 | .and(path("/v1/responses")) |
| 1011 | .and(header("authorization", "Bearer deepseek-test-key")) |
| 1012 | .and(body_partial_json(json!({ |
| 1013 | "model": "deepseek-v4-flash", |
| 1014 | "tools": [{ "type": "web_search" }], |
| 1015 | "tool_choice": { "type": "web_search" } |
| 1016 | }))) |
| 1017 | .respond_with(ResponseTemplate::new(200).set_body_json(json!({ |
| 1018 | "output": [ |
| 1019 | { |
| 1020 | "type": "web_search_call", |
| 1021 | "action": { |
| 1022 | "type": "open_page", |
| 1023 | "url": "https://example.com/deepseek" |
| 1024 | } |
| 1025 | }, |
| 1026 | { |
| 1027 | "type": "message", |
| 1028 | "content": [{ |
| 1029 | "type": "output_text", |
| 1030 | "text": "Grounded answer.", |
| 1031 | "annotations": [] |
| 1032 | }] |
| 1033 | } |
| 1034 | ] |
| 1035 | }))) |
| 1036 | .expect(1) |
| 1037 | .mount(&server) |
| 1038 | .await; |
| 1039 | let config = Config { |
| 1040 | provider: Some("deepseek".to_string()), |
| 1041 | providers: Some(ProvidersConfig { |
| 1042 | deepseek: ProviderConfig { |
| 1043 | api_key: Some("deepseek-test-key".to_string()), |
| 1044 | base_url: Some(format!("{}/v1", server.uri())), |
| 1045 | model: Some("deepseek-v4-flash".to_string()), |
| 1046 | ..ProviderConfig::default() |
| 1047 | }, |
| 1048 | ..ProvidersConfig::default() |
| 1049 | }), |
| 1050 | ..Config::default() |
| 1051 | }; |
| 1052 | let inner = CodewhaleClient::new(&config).expect("test DeepSeek client"); |
| 1053 | let client = ProviderNativeSearchClient::new(inner).expect("DeepSeek native adapter"); |
| 1054 | |
| 1055 | let response = client.search(&request()).await.expect("native search"); |
| 1056 | |
| 1057 | assert_eq!(response.answer.as_deref(), Some("Grounded answer.")); |
| 1058 | assert_eq!(response.citations.len(), 1); |
| 1059 | assert_eq!(response.citations[0].url, "https://example.com/deepseek"); |
| 1060 | } |
| 1061 | |
| 1062 | #[tokio::test] |
| 1063 | async fn native_search_obeys_attached_run_ownership_without_blocking_unrelated_runtime() { |
| 1064 | let server = MockServer::start().await; |
| 1065 | Mock::given(method("POST")) |
| 1066 | .and(path("/v1/responses")) |
| 1067 | .respond_with(ResponseTemplate::new(200).set_body_json(json!({ |
| 1068 | "output_text": "Grounded answer.", |
| 1069 | "citations": ["https://example.com/source"] |
| 1070 | }))) |
| 1071 | .expect(3) |
| 1072 | .mount(&server) |
| 1073 | .await; |
| 1074 | let participant = xai_client_with_boundary(&server, false, false); |
| 1075 | let isolated = xai_client_with_boundary(&server, true, false); |
| 1076 | let unrelated = xai_client_with_boundary(&server, false, true); |
| 1077 | |
| 1078 | let ownership = crate::client::acquire_runtime_chat_inference_ownership().await; |
| 1079 | let participant_request = request(); |
| 1080 | let mut waiting = |
| 1081 | tokio::spawn(async move { participant.search(&participant_request).await }); |
| 1082 | assert!( |
| 1083 | tokio::time::timeout(std::time::Duration::from_millis(40), &mut waiting) |
| 1084 | .await |
| 1085 | .is_err(), |
| 1086 | "provider-native inference from the attached run must wait behind Runtime Chat" |
| 1087 | ); |
| 1088 | |
| 1089 | tokio::time::timeout( |
| 1090 | std::time::Duration::from_secs(1), |
| 1091 | isolated.search(&request()), |
| 1092 | ) |
| 1093 | .await |
| 1094 | .expect("isolated relay request must not self-deadlock") |
| 1095 | .expect("isolated native search fixture"); |
| 1096 | tokio::time::timeout( |
| 1097 | std::time::Duration::from_secs(1), |
| 1098 | unrelated.search(&request()), |
| 1099 | ) |
| 1100 | .await |
| 1101 | .expect("unrelated Runtime manager stays concurrent") |
| 1102 | .expect("unrelated native search fixture"); |
| 1103 | |
| 1104 | drop(ownership); |
| 1105 | tokio::time::timeout(std::time::Duration::from_secs(1), waiting) |
| 1106 | .await |
| 1107 | .expect("attached participant resumes after relay settlement") |
| 1108 | .expect("participant task") |
| 1109 | .expect("participant native search fixture"); |
| 1110 | } |
| 1111 | } |
| 1112 |