| 1 | //! Provider/model inventory for routing policy. |
| 2 | //! |
| 3 | //! This is the high-level "what can this user actually run?" object. Auto |
| 4 | //! routing, fleet workers, and sub-agent policy should consume this shape |
| 5 | //! instead of guessing model strings from global defaults. |
| 6 | |
| 7 | use serde::Serialize; |
| 8 | |
| 9 | use crate::client::system_one::DecisionRouterRoute; |
| 10 | use crate::config::{ |
| 11 | AutoRouterKind, Config, ProviderIdentity, ProviderKind, has_api_key_for, |
| 12 | normalize_model_name_for_provider, provider_capability, |
| 13 | }; |
| 14 | use crate::provider_lake::models_for_provider; |
| 15 | |
| 16 | #[derive(Debug, Clone, PartialEq, Eq, Serialize)] |
| 17 | #[serde(rename_all = "snake_case")] |
| 18 | pub(crate) enum ModelAuthSource { |
| 19 | Config, |
| 20 | Env, |
| 21 | OAuthCli, |
| 22 | ImportedToken, |
| 23 | NoAuth, |
| 24 | KeylessLocal, |
| 25 | } |
| 26 | |
| 27 | #[derive(Debug, Clone, PartialEq, Eq, Serialize)] |
| 28 | pub(crate) struct ModelRouteCandidate { |
| 29 | #[serde(rename = "provider")] |
| 30 | pub(crate) provider_tag: String, |
| 31 | #[serde(skip)] |
| 32 | pub(crate) provider: ProviderKind, |
| 33 | #[serde(skip)] |
| 34 | pub(crate) identity: ProviderIdentity, |
| 35 | pub(crate) provider_name: String, |
| 36 | pub(crate) provider_display_name: String, |
| 37 | pub(crate) model: String, |
| 38 | /// Explicit declarations keep case-sensitive wire identity; bundled aliases |
| 39 | /// retain the existing case-insensitive convenience lookup. |
| 40 | #[serde(skip_serializing_if = "std::ops::Not::not")] |
| 41 | pub(crate) user_declared: bool, |
| 42 | pub(crate) context_window: u32, |
| 43 | /// The context window came from the legacy capability fallback (an `_Nk` |
| 44 | /// name-suffix parse or a vendor-family heuristic), not a route fact |
| 45 | /// (#5441). Serialized only when true so existing payloads stay stable. |
| 46 | #[serde(skip_serializing_if = "std::ops::Not::not")] |
| 47 | pub(crate) context_window_unverified: bool, |
| 48 | /// Known output ceiling, or `None` when this route publishes none. The |
| 49 | /// classifier is told "unknown" rather than a fabricated number. |
| 50 | #[serde(skip_serializing_if = "Option::is_none")] |
| 51 | pub(crate) max_output: Option<u32>, |
| 52 | pub(crate) thinking_supported: bool, |
| 53 | pub(crate) cache_telemetry_supported: bool, |
| 54 | pub(crate) auth_source: ModelAuthSource, |
| 55 | pub(crate) readiness: crate::provider_readiness::ResolvedProviderReadiness, |
| 56 | pub(crate) default_for_provider: bool, |
| 57 | pub(crate) tags: Vec<&'static str>, |
| 58 | } |
| 59 | |
| 60 | /// Why a declared `[auto.router]` cannot run. |
| 61 | #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)] |
| 62 | #[serde(rename_all = "snake_case")] |
| 63 | pub(crate) enum AutoRouterSetupIssue { |
| 64 | /// `kind` is neither `"chat"` nor `"decision"`. |
| 65 | UnknownKind, |
| 66 | /// `kind = "decision"` names a provider that serves no decision API. |
| 67 | UnsupportedDecisionProvider, |
| 68 | /// `provider` or `model` is missing (or the chat provider is unknown). |
| 69 | Incomplete, |
| 70 | /// The route is complete but its credential is missing. |
| 71 | MissingKey, |
| 72 | /// `thinking` is not a reasoning tier; the classifier call would carry |
| 73 | /// an effort the provider rejects or silently reinterprets. |
| 74 | InvalidThinking, |
| 75 | /// `model` is not one the chat router's provider serves (checked where |
| 76 | /// the provider's model namespace is known; pass-through providers such |
| 77 | /// as OpenRouter or a custom endpoint are validated upstream). |
| 78 | InvalidModel, |
| 79 | } |
| 80 | |
| 81 | impl AutoRouterSetupIssue { |
| 82 | #[must_use] |
| 83 | pub(crate) fn label(self) -> &'static str { |
| 84 | match self { |
| 85 | Self::UnknownKind => "unknown [auto.router] kind (expected \"chat\" or \"decision\")", |
| 86 | Self::UnsupportedDecisionProvider => { |
| 87 | "decision routers are served by OpenRouter or TypeSafe" |
| 88 | } |
| 89 | Self::Incomplete => "[auto.router] needs a known provider and a model", |
| 90 | Self::MissingKey => "no API key for the router route", |
| 91 | Self::InvalidThinking => { |
| 92 | "[auto.router] thinking must be auto, off, minimal, low, medium, high, xhigh, ultra, or max" |
| 93 | } |
| 94 | Self::InvalidModel => "[auto.router] model is not served by the router provider", |
| 95 | } |
| 96 | } |
| 97 | } |
| 98 | |
| 99 | /// Probability or confidence (0..=1) as basis points, so receipts and the |
| 100 | /// inventory stay `Eq` and never re-render a float. |
| 101 | #[must_use] |
| 102 | pub(crate) fn probability_bp(value: f64) -> u16 { |
| 103 | if !value.is_finite() { |
| 104 | return 0; |
| 105 | } |
| 106 | // Bounded to 0..=10000 before the cast. |
| 107 | (value.clamp(0.0, 1.0) * 10_000.0).round() as u16 |
| 108 | } |
| 109 | |
| 110 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 111 | pub(crate) struct ModelInventory { |
| 112 | pub(crate) active_provider: ProviderKind, |
| 113 | pub(crate) active_identity: ProviderIdentity, |
| 114 | pub(crate) router_provider: ProviderKind, |
| 115 | pub(crate) router_identity: Option<ProviderIdentity>, |
| 116 | pub(crate) router_model: String, |
| 117 | /// Thinking tier for the classifier call (None = off) (#auto.router). |
| 118 | pub(crate) router_thinking: Option<String>, |
| 119 | /// Classifier call timeout in seconds (default 4; clamped at config load). |
| 120 | pub(crate) router_timeout_secs: u64, |
| 121 | /// Whether an explicit legacy `[auto.router]` classifier route is |
| 122 | /// configured. Absent configuration means legacy Auto stays local/free — |
| 123 | /// holding a provider key never elects a network classifier by itself. |
| 124 | pub(crate) router_configured: bool, |
| 125 | pub(crate) router_available: bool, |
| 126 | /// `[auto.router] kind` (#6525); an unknown kind leaves the router |
| 127 | /// unconfigured and sets [`Self::router_setup_issue`]. |
| 128 | pub(crate) router_kind: AutoRouterKind, |
| 129 | /// Endpoint for a decision router (`None` for chat routers). |
| 130 | pub(crate) router_decision_route: Option<DecisionRouterRoute>, |
| 131 | /// Decision-router endpoint override (TypeSafe only). |
| 132 | pub(crate) router_base_url: Option<String>, |
| 133 | /// Decision-router confidence floor in basis points (0..=10000). |
| 134 | pub(crate) router_min_confidence_bp: u16, |
| 135 | /// Why a declared `[auto.router]` cannot run. `None` when no router is |
| 136 | /// declared or the router is available. A declared-but-unusable router is |
| 137 | /// shown as failing, never silently ignored. |
| 138 | pub(crate) router_setup_issue: Option<AutoRouterSetupIssue>, |
| 139 | /// `[auto] cross_provider = true` opt-in (#4411). When false (the |
| 140 | /// default), Auto routing — classifier payload included — is confined to |
| 141 | /// `active_provider`. The full candidate list still carries every |
| 142 | /// authenticated provider because pickers and explicit `/model` lookups |
| 143 | /// legitimately need it; only the Auto paths are scoped. |
| 144 | pub(crate) cross_provider_auto: bool, |
| 145 | pub(crate) candidates: Vec<ModelRouteCandidate>, |
| 146 | } |
| 147 | |
| 148 | impl Serialize for ModelInventory { |
| 149 | fn serialize<S: serde::Serializer>(&self, serializer: S) -> Result<S::Ok, S::Error> { |
| 150 | use serde::ser::SerializeStruct; |
| 151 | let active = codewhale_config::descriptors::tui_wire_tag_for_route( |
| 152 | self.active_provider, |
| 153 | self.active_identity.key.as_str(), |
| 154 | ) |
| 155 | .ok_or_else(|| serde::ser::Error::custom("contradictory active inventory identity"))?; |
| 156 | let router = self |
| 157 | .router_identity |
| 158 | .as_ref() |
| 159 | .map_or_else( |
| 160 | || { |
| 161 | Some( |
| 162 | codewhale_config::descriptors::compatibility_for_kind(self.router_provider) |
| 163 | .tui_wire_tag, |
| 164 | ) |
| 165 | }, |
| 166 | |identity| { |
| 167 | codewhale_config::descriptors::tui_wire_tag_for_route( |
| 168 | self.router_provider, |
| 169 | identity.key.as_str(), |
| 170 | ) |
| 171 | }, |
| 172 | ) |
| 173 | .ok_or_else(|| serde::ser::Error::custom("contradictory router inventory identity"))?; |
| 174 | let mut state = serializer.serialize_struct("ModelInventory", 14)?; |
| 175 | state.serialize_field("active_provider", active)?; |
| 176 | state.serialize_field("router_provider", router)?; |
| 177 | state.serialize_field("router_model", &self.router_model)?; |
| 178 | state.serialize_field("router_thinking", &self.router_thinking)?; |
| 179 | state.serialize_field("router_timeout_secs", &self.router_timeout_secs)?; |
| 180 | state.serialize_field("router_configured", &self.router_configured)?; |
| 181 | state.serialize_field("router_available", &self.router_available)?; |
| 182 | state.serialize_field("router_kind", &self.router_kind)?; |
| 183 | state.serialize_field("router_decision_route", &self.router_decision_route)?; |
| 184 | state.serialize_field("router_base_url", &self.router_base_url)?; |
| 185 | state.serialize_field("router_min_confidence_bp", &self.router_min_confidence_bp)?; |
| 186 | state.serialize_field("router_setup_issue", &self.router_setup_issue)?; |
| 187 | state.serialize_field("cross_provider_auto", &self.cross_provider_auto)?; |
| 188 | state.serialize_field("candidates", &self.candidates)?; |
| 189 | state.end() |
| 190 | } |
| 191 | } |
| 192 | |
| 193 | impl ModelInventory { |
| 194 | pub(crate) fn from_config(config: &Config) -> Result<Self, String> { |
| 195 | Self::from_config_with_health( |
| 196 | config, |
| 197 | &crate::provider_readiness::ProviderReadinessSnapshot::default(), |
| 198 | ) |
| 199 | } |
| 200 | |
| 201 | pub(crate) fn from_config_with_health( |
| 202 | config: &Config, |
| 203 | health: &crate::provider_readiness::ProviderReadinessSnapshot, |
| 204 | ) -> Result<Self, String> { |
| 205 | let active_identity = config.active_provider_identity()?; |
| 206 | let active_provider = active_identity.provider; |
| 207 | let mut candidates = Vec::new(); |
| 208 | |
| 209 | for identity in config.provider_identities() { |
| 210 | let provider = identity.provider; |
| 211 | let Some(auth_source) = auth_source_for_provider(config, &identity) else { |
| 212 | continue; |
| 213 | }; |
| 214 | let default_model = provider_default_model(config, &identity); |
| 215 | let mut models = Vec::<String>::new(); |
| 216 | if let Some(model) = configured_model_for_provider(config, &identity) { |
| 217 | push_model(&mut models, provider, &model); |
| 218 | } |
| 219 | if active_identity == identity { |
| 220 | let active_model = config.default_model(); |
| 221 | if !active_model.trim().eq_ignore_ascii_case("auto") { |
| 222 | push_model(&mut models, provider, &active_model); |
| 223 | } |
| 224 | } |
| 225 | for model in models_for_provider(config, &identity) { |
| 226 | push_model(&mut models, provider, &model); |
| 227 | } |
| 228 | for declaration in config.custom_models.as_deref().unwrap_or_default() { |
| 229 | if crate::provider_lake::configured_model_for_route( |
| 230 | config, |
| 231 | provider, |
| 232 | identity.key.as_str(), |
| 233 | &config.base_url_for_route(&identity), |
| 234 | &declaration.id, |
| 235 | ) |
| 236 | .is_some() |
| 237 | && !models.contains(&declaration.id) |
| 238 | { |
| 239 | models.push(declaration.id.clone()); |
| 240 | } |
| 241 | } |
| 242 | if models.is_empty() { |
| 243 | push_model(&mut models, provider, &default_model); |
| 244 | } |
| 245 | |
| 246 | for model in models { |
| 247 | let readiness = |
| 248 | crate::provider_readiness::resolve_for_model(config, &identity, &model, health); |
| 249 | let mut capability = provider_capability(provider, &model); |
| 250 | let mut user_declared = false; |
| 251 | // #5239/#5441: a candidate whose window came from the legacy |
| 252 | // capability fallback (a `_Nk` name-suffix parse or a |
| 253 | // vendor-family heuristic) carries the number *and* the fact |
| 254 | // that nobody verified it — the auto-router must not read a |
| 255 | // guessed window as a route capability. |
| 256 | let mut context_window_unverified = |
| 257 | codewhale_config::catalog::reviewed::intrinsic_model(&model) |
| 258 | .and_then(|row| row.context_window) |
| 259 | .is_none(); |
| 260 | if let Ok(route) = crate::route_runtime::resolve_runtime_route_for_identity( |
| 261 | config, |
| 262 | &identity, |
| 263 | Some(&model), |
| 264 | ) { |
| 265 | if let Some(context_window) = route.candidate.limits().context_tokens { |
| 266 | capability.context_window = context_window.min(u64::from(u32::MAX)) as u32; |
| 267 | context_window_unverified = !route.context_window.source.is_verified(); |
| 268 | } |
| 269 | // A concrete offering maximum is a stronger fact than the |
| 270 | // static compatibility matrix — and is the only way a |
| 271 | // membership route (no static cap) gets a known ceiling. |
| 272 | if let Some(max_output) = route |
| 273 | .candidate |
| 274 | .limits() |
| 275 | .output_tokens |
| 276 | .and_then(|tokens| u32::try_from(tokens).ok()) |
| 277 | .filter(|tokens| *tokens > 0) |
| 278 | { |
| 279 | capability.max_output = Some(max_output); |
| 280 | } |
| 281 | user_declared = route |
| 282 | .candidate |
| 283 | .applied_limit_overrides() |
| 284 | .iter() |
| 285 | .any(|entry| { |
| 286 | entry.source |
| 287 | == codewhale_config::route::OverrideSource::UserModelMetadata |
| 288 | }); |
| 289 | if user_declared { |
| 290 | context_window_unverified = !route.context_window.source.is_verified(); |
| 291 | capability.context_window = route.context_window.tokens; |
| 292 | capability.max_output = route |
| 293 | .candidate |
| 294 | .limits() |
| 295 | .output_tokens |
| 296 | .and_then(|value| u32::try_from(value).ok()); |
| 297 | capability.thinking_supported = route.candidate.capabilities().reasoning |
| 298 | == codewhale_config::route::CapabilityState::Supported; |
| 299 | } |
| 300 | // Do not promote bare `k3` into the global capability |
| 301 | // catalog. Its thinking trace contract belongs only to |
| 302 | // Kimi Code's exact membership-plan route. |
| 303 | if !user_declared |
| 304 | && crate::config::is_exact_kimi_code_k3_route( |
| 305 | provider, |
| 306 | &route.candidate.endpoint().base_url, |
| 307 | route.candidate.wire_model_id().as_str(), |
| 308 | ) |
| 309 | { |
| 310 | capability.thinking_supported = true; |
| 311 | } |
| 312 | } |
| 313 | let mut tags = Vec::new(); |
| 314 | if capability.context_window >= 1_000_000 { |
| 315 | tags.push("long_context"); |
| 316 | } |
| 317 | if capability.thinking_supported { |
| 318 | tags.push("thinking"); |
| 319 | } |
| 320 | if matches!( |
| 321 | provider, |
| 322 | ProviderKind::Ollama | ProviderKind::Sglang | ProviderKind::Vllm |
| 323 | ) { |
| 324 | tags.push("local"); |
| 325 | } |
| 326 | // Unready routes stay visible (annotated) so an operator can |
| 327 | // override explicitly, but they are never a silent default. |
| 328 | let default_for_provider = readiness.can_attempt() |
| 329 | && (model == default_model |
| 330 | || (!user_declared && model.eq_ignore_ascii_case(&default_model))); |
| 331 | if default_for_provider { |
| 332 | tags.push("default"); |
| 333 | } |
| 334 | if !readiness.can_attempt() { |
| 335 | tags.push("unready"); |
| 336 | } |
| 337 | |
| 338 | candidates.push(ModelRouteCandidate { |
| 339 | provider, |
| 340 | provider_tag: codewhale_config::descriptors::tui_wire_tag_for_route( |
| 341 | provider, |
| 342 | identity.key.as_str(), |
| 343 | ) |
| 344 | .expect("admitted identity has a wire projection") |
| 345 | .to_string(), |
| 346 | identity: identity.clone(), |
| 347 | provider_name: identity.key.to_string(), |
| 348 | provider_display_name: identity.compatibility().map_or_else( |
| 349 | || format!("{} (custom)", identity.key), |
| 350 | |row| row.label.to_string(), |
| 351 | ), |
| 352 | default_for_provider, |
| 353 | model, |
| 354 | user_declared, |
| 355 | context_window: capability.context_window, |
| 356 | context_window_unverified, |
| 357 | max_output: capability.max_output, |
| 358 | thinking_supported: capability.thinking_supported, |
| 359 | cache_telemetry_supported: capability.cache_telemetry_supported, |
| 360 | auth_source: auth_source.clone(), |
| 361 | readiness: readiness.clone(), |
| 362 | tags, |
| 363 | }); |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | // `[auto.router]` is legacy `model = auto` configuration and stays that |
| 368 | // way — it is NOT a Fleet Router. Explicit configuration still works. |
| 369 | // |
| 370 | // What is gone is the implicit half: merely holding a DeepSeek key used |
| 371 | // to silently elect `deepseek-v4-flash` as a network classifier for |
| 372 | // every Auto turn, spending a user's tokens on a route they never asked |
| 373 | // for and privileging one provider. With no explicit `[auto.router]`, |
| 374 | // legacy Auto is now local/free (heuristic-only). |
| 375 | let router_table = config.auto.as_ref().and_then(|auto| auto.router.as_ref()); |
| 376 | // Any populated key declares a router: a table holding only |
| 377 | // `timeout_secs` or `base_url` is a malformed router to diagnose, not |
| 378 | // an absent one to skip silently. |
| 379 | let router_declared = router_table.is_some_and(|router| { |
| 380 | router.provider.is_some() |
| 381 | || router.model.is_some() |
| 382 | || router.kind.is_some() |
| 383 | || router.thinking.is_some() |
| 384 | || router.timeout_secs.is_some() |
| 385 | || router.min_confidence.is_some() |
| 386 | || router.base_url.is_some() |
| 387 | }); |
| 388 | let router_kind = AutoRouterKind::parse(router_table.and_then(|r| r.kind.as_deref())); |
| 389 | let router_model_setting = router_table |
| 390 | .and_then(|router| router.model.as_deref()) |
| 391 | .map(str::trim) |
| 392 | .filter(|model| !model.is_empty()); |
| 393 | let router_provider_setting = router_table |
| 394 | .and_then(|router| router.provider.as_deref()) |
| 395 | .map(str::trim) |
| 396 | .filter(|provider| !provider.is_empty()); |
| 397 | let mut router_setup_issue = None; |
| 398 | let mut router_decision_route = None; |
| 399 | let explicit_router = match router_kind { |
| 400 | None => { |
| 401 | router_setup_issue = Some(AutoRouterSetupIssue::UnknownKind); |
| 402 | None |
| 403 | } |
| 404 | Some(AutoRouterKind::Chat) => { |
| 405 | let thinking = router_table |
| 406 | .and_then(|router| router.thinking.as_deref()) |
| 407 | .map(str::trim) |
| 408 | .filter(|t| !t.is_empty()); |
| 409 | // A typo here would otherwise ride every classifier request |
| 410 | // as an effort string the provider rejects, failing Auto back |
| 411 | // to the local fallback on every turn. |
| 412 | if thinking.is_some_and(|thinking| { |
| 413 | crate::reasoning_preference::ReasoningEffort::parse_strict(thinking).is_err() |
| 414 | }) { |
| 415 | router_setup_issue = Some(AutoRouterSetupIssue::InvalidThinking); |
| 416 | None |
| 417 | } else { |
| 418 | let route = router_provider_setting |
| 419 | .and_then(|name| config.resolve_provider_pin_identity(name).ok()) |
| 420 | .zip(router_model_setting); |
| 421 | // Same check a session route gets: a model the provider |
| 422 | // cannot serve would 404 every classifier call and fall |
| 423 | // back to local routing on every turn. |
| 424 | if route.as_ref().is_some_and(|(identity, model)| { |
| 425 | crate::config::validate_route(identity.provider, model).is_err() |
| 426 | }) { |
| 427 | router_setup_issue = Some(AutoRouterSetupIssue::InvalidModel); |
| 428 | None |
| 429 | } else { |
| 430 | route.map(|(identity, model)| { |
| 431 | ( |
| 432 | identity.provider, |
| 433 | Some(identity), |
| 434 | model.to_string(), |
| 435 | thinking.map(str::to_string), |
| 436 | ) |
| 437 | }) |
| 438 | } |
| 439 | } |
| 440 | } |
| 441 | Some(AutoRouterKind::Decision) => { |
| 442 | match router_provider_setting.map(DecisionRouterRoute::parse) { |
| 443 | Some(None) => { |
| 444 | router_setup_issue = |
| 445 | Some(AutoRouterSetupIssue::UnsupportedDecisionProvider); |
| 446 | None |
| 447 | } |
| 448 | Some(Some(route)) => router_model_setting.map(|model| { |
| 449 | router_decision_route = Some(route); |
| 450 | // A decision model has no reasoning knob, so the |
| 451 | // configured `thinking` is ignored. TypeSafe is not a |
| 452 | // chat provider; `router_provider` is only a label |
| 453 | // there, and `router_decision_route` is authoritative. |
| 454 | ( |
| 455 | ProviderKind::Openrouter, |
| 456 | config |
| 457 | .builtin_provider_identity(ProviderKind::Openrouter) |
| 458 | .ok(), |
| 459 | model.to_string(), |
| 460 | None, |
| 461 | ) |
| 462 | }), |
| 463 | None => None, |
| 464 | } |
| 465 | } |
| 466 | }; |
| 467 | let router_configured = explicit_router.is_some(); |
| 468 | let (router_provider, router_identity, router_model, router_thinking) = explicit_router |
| 469 | // Kept only as an inert display/default label for the router fields; |
| 470 | // `router_available` below is what gates any classifier call. |
| 471 | .unwrap_or_else(|| { |
| 472 | ( |
| 473 | ProviderKind::Deepseek, |
| 474 | None, |
| 475 | "deepseek-v4-flash".to_string(), |
| 476 | None, |
| 477 | ) |
| 478 | }); |
| 479 | let router_available = router_configured |
| 480 | && match router_decision_route { |
| 481 | Some(route) => route.has_key(config), |
| 482 | None => router_identity |
| 483 | .as_ref() |
| 484 | .is_some_and(|identity| has_api_key_for(config, identity)), |
| 485 | }; |
| 486 | if router_declared && router_setup_issue.is_none() && !router_available { |
| 487 | router_setup_issue = Some(if router_configured { |
| 488 | AutoRouterSetupIssue::MissingKey |
| 489 | } else { |
| 490 | AutoRouterSetupIssue::Incomplete |
| 491 | }); |
| 492 | } |
| 493 | |
| 494 | let cross_provider_auto = config.auto_cross_provider(); |
| 495 | let router_timeout_secs = config.auto_router_timeout_secs(); |
| 496 | let router_min_confidence_bp = probability_bp(config.auto_router_min_confidence()); |
| 497 | |
| 498 | Ok(Self { |
| 499 | active_provider, |
| 500 | active_identity, |
| 501 | router_provider, |
| 502 | router_identity, |
| 503 | router_configured, |
| 504 | router_available, |
| 505 | router_model, |
| 506 | router_thinking, |
| 507 | router_timeout_secs, |
| 508 | router_kind: router_kind.unwrap_or(AutoRouterKind::Chat), |
| 509 | router_decision_route, |
| 510 | router_base_url: router_table |
| 511 | .and_then(|router| router.base_url.as_deref()) |
| 512 | .map(str::trim) |
| 513 | .filter(|url| !url.is_empty()) |
| 514 | .map(str::to_string), |
| 515 | router_min_confidence_bp, |
| 516 | router_setup_issue, |
| 517 | cross_provider_auto, |
| 518 | candidates, |
| 519 | }) |
| 520 | } |
| 521 | |
| 522 | /// Whether Auto routing may select `provider` (#4411). |
| 523 | pub(crate) fn auto_scope_allows(&self, provider_id: &str) -> bool { |
| 524 | self.cross_provider_auto || self.active_identity.key.as_str() == provider_id |
| 525 | } |
| 526 | |
| 527 | pub(crate) fn candidate(&self, provider_id: &str, model: &str) -> Option<&ModelRouteCandidate> { |
| 528 | let model = model.trim(); |
| 529 | self.candidates |
| 530 | .iter() |
| 531 | .find(|candidate| { |
| 532 | candidate.identity.key.as_str() == provider_id && candidate.model == model |
| 533 | }) |
| 534 | .or_else(|| { |
| 535 | self.candidates.iter().find(|candidate| { |
| 536 | candidate.identity.key.as_str() == provider_id |
| 537 | && !candidate.user_declared |
| 538 | && candidate.model.eq_ignore_ascii_case(model) |
| 539 | }) |
| 540 | }) |
| 541 | } |
| 542 | |
| 543 | pub(crate) fn active_default(&self) -> Option<&ModelRouteCandidate> { |
| 544 | self.candidates |
| 545 | .iter() |
| 546 | .find(|candidate| { |
| 547 | self.active_identity == candidate.identity && candidate.default_for_provider |
| 548 | }) |
| 549 | .or_else(|| { |
| 550 | self.candidates.iter().find(|candidate| { |
| 551 | self.active_identity == candidate.identity && candidate.readiness.can_attempt() |
| 552 | }) |
| 553 | }) |
| 554 | .or_else(|| { |
| 555 | // Falling through to another provider is a cross-provider Auto |
| 556 | // route (#4411): allowed only under the persisted opt-in. With |
| 557 | // it off, an unusable active provider surfaces as "no runnable |
| 558 | // candidate" instead of silently borrowing another provider's |
| 559 | // credentials. |
| 560 | self.cross_provider_auto |
| 561 | .then(|| { |
| 562 | self.candidates |
| 563 | .iter() |
| 564 | .find(|candidate| candidate.readiness.can_attempt()) |
| 565 | }) |
| 566 | .flatten() |
| 567 | }) |
| 568 | } |
| 569 | |
| 570 | pub(crate) fn router_context_json(&self) -> String { |
| 571 | #[derive(Serialize)] |
| 572 | struct RouterInventoryContext<'a> { |
| 573 | active_provider: &'a str, |
| 574 | candidates: Vec<RouterCandidateContext<'a>>, |
| 575 | } |
| 576 | |
| 577 | #[derive(Serialize)] |
| 578 | struct RouterCandidateContext<'a> { |
| 579 | provider: &'a str, |
| 580 | provider_name: &'a str, |
| 581 | provider_display_name: &'a str, |
| 582 | model: &'a str, |
| 583 | context_window: u32, |
| 584 | #[serde(skip_serializing_if = "std::ops::Not::not")] |
| 585 | context_window_unverified: bool, |
| 586 | #[serde(skip_serializing_if = "Option::is_none")] |
| 587 | max_output: Option<u32>, |
| 588 | thinking_supported: bool, |
| 589 | cache_telemetry_supported: bool, |
| 590 | default_for_provider: bool, |
| 591 | tags: &'a [&'static str], |
| 592 | } |
| 593 | |
| 594 | // The classifier needs route capabilities, not credentials, endpoint |
| 595 | // configuration, or provider error text. Filter to runnable candidates |
| 596 | // and project only non-secret routing facts before serializing. |
| 597 | // |
| 598 | // Scope (#4411): without the persisted `[auto] cross_provider` opt-in, |
| 599 | // the payload names only the active provider's routes. Which other |
| 600 | // providers a user has credentials for is not something Auto discloses |
| 601 | // to a classifier by default. |
| 602 | let candidates = self |
| 603 | .candidates |
| 604 | .iter() |
| 605 | .filter(|candidate| { |
| 606 | candidate.readiness.can_attempt() |
| 607 | && self.auto_scope_allows(candidate.identity.key.as_str()) |
| 608 | }) |
| 609 | .map(|candidate| RouterCandidateContext { |
| 610 | provider: &candidate.provider_tag, |
| 611 | provider_name: &candidate.provider_name, |
| 612 | provider_display_name: &candidate.provider_display_name, |
| 613 | model: &candidate.model, |
| 614 | context_window: candidate.context_window, |
| 615 | context_window_unverified: candidate.context_window_unverified, |
| 616 | max_output: candidate.max_output, |
| 617 | thinking_supported: candidate.thinking_supported, |
| 618 | cache_telemetry_supported: candidate.cache_telemetry_supported, |
| 619 | default_for_provider: candidate.default_for_provider, |
| 620 | tags: &candidate.tags, |
| 621 | }) |
| 622 | .collect(); |
| 623 | serde_json::to_string(&RouterInventoryContext { |
| 624 | active_provider: self.active_identity.key.as_str(), |
| 625 | candidates, |
| 626 | }) |
| 627 | .unwrap_or_else(|_| "{}".to_string()) |
| 628 | } |
| 629 | } |
| 630 | |
| 631 | fn push_model(models: &mut Vec<String>, provider: ProviderKind, model: &str) { |
| 632 | if provider == ProviderKind::Ollama && crate::config::is_unresolved_local_ollama_model(model) { |
| 633 | return; |
| 634 | } |
| 635 | let Some(model) = normalize_model_name_for_provider(provider, model) |
| 636 | .or_else(|| crate::config::normalize_custom_model_id(model)) |
| 637 | else { |
| 638 | return; |
| 639 | }; |
| 640 | if !models |
| 641 | .iter() |
| 642 | .any(|existing| existing.eq_ignore_ascii_case(&model)) |
| 643 | { |
| 644 | models.push(model); |
| 645 | } |
| 646 | } |
| 647 | |
| 648 | fn configured_model_for_provider(config: &Config, identity: &ProviderIdentity) -> Option<String> { |
| 649 | config |
| 650 | .provider_config_for(identity) |
| 651 | .and_then(|entry| entry.model.clone()) |
| 652 | .map(|model| model.trim().to_string()) |
| 653 | .filter(|model| !model.is_empty()) |
| 654 | } |
| 655 | |
| 656 | pub(crate) fn provider_default_model(config: &Config, identity: &ProviderIdentity) -> String { |
| 657 | let provider = identity.provider; |
| 658 | let configured = configured_model_for_provider(config, identity).or_else(|| { |
| 659 | (config.active_provider_identity().ok().as_ref() == Some(identity) |
| 660 | && config.default_text_model.is_some()) |
| 661 | .then(|| config.default_model()) |
| 662 | }); |
| 663 | let selector = configured.as_deref().filter(|model| { |
| 664 | !model.trim().eq_ignore_ascii_case("auto") |
| 665 | && !(provider == ProviderKind::Ollama |
| 666 | && crate::config::is_unresolved_local_ollama_model(model)) |
| 667 | }); |
| 668 | // Inventory labels must use the executable route's exact endpoint default, |
| 669 | // not whichever provider-wide snapshot happened to refresh most recently. |
| 670 | crate::route_runtime::resolve_runtime_route_for_identity(config, identity, selector) |
| 671 | .map(|route| route.model) |
| 672 | .unwrap_or_else(|_| { |
| 673 | configured.unwrap_or_else(|| { |
| 674 | identity |
| 675 | .compatibility() |
| 676 | .map_or("", |row| row.default_model) |
| 677 | .to_string() |
| 678 | }) |
| 679 | }) |
| 680 | } |
| 681 | |
| 682 | fn auth_source_for_provider( |
| 683 | config: &Config, |
| 684 | identity: &ProviderIdentity, |
| 685 | ) -> Option<ModelAuthSource> { |
| 686 | let provider = identity.provider; |
| 687 | let credential_state = |
| 688 | crate::provider_readiness::credential_state_for_provider(config, identity); |
| 689 | match credential_state { |
| 690 | crate::provider_readiness::CredentialState::NoAuth => { |
| 691 | return Some(ModelAuthSource::NoAuth); |
| 692 | } |
| 693 | crate::provider_readiness::CredentialState::Local => { |
| 694 | return Some(ModelAuthSource::KeylessLocal); |
| 695 | } |
| 696 | crate::provider_readiness::CredentialState::ImportedToken => { |
| 697 | return Some(ModelAuthSource::ImportedToken); |
| 698 | } |
| 699 | crate::provider_readiness::CredentialState::MissingKey |
| 700 | | crate::provider_readiness::CredentialState::MissingLogin |
| 701 | | crate::provider_readiness::CredentialState::ExternalConsent |
| 702 | | crate::provider_readiness::CredentialState::Legacy => return None, |
| 703 | crate::provider_readiness::CredentialState::Saved => {} |
| 704 | } |
| 705 | |
| 706 | if provider == ProviderKind::Custom { |
| 707 | let configured = config.provider_config_for(identity)?; |
| 708 | if configured |
| 709 | .api_key_env |
| 710 | .as_deref() |
| 711 | .map(str::trim) |
| 712 | .filter(|name| !name.is_empty()) |
| 713 | .is_some_and(|name| std::env::var(name).is_ok_and(|value| !value.trim().is_empty())) |
| 714 | { |
| 715 | return Some(ModelAuthSource::Env); |
| 716 | } |
| 717 | return (configured.api_key.as_deref().is_some_and(|value| { |
| 718 | crate::config::classify_config_api_key_value(value) |
| 719 | == crate::config::ConfigApiKeyValueKind::Literal |
| 720 | }) || crate::config::explicit_cli_api_key_override().is_some()) |
| 721 | .then_some(ModelAuthSource::Config); |
| 722 | } |
| 723 | if provider_uses_oauth_cli(config, identity) { |
| 724 | return Some(ModelAuthSource::OAuthCli); |
| 725 | } |
| 726 | if config |
| 727 | .provider_config_for(identity) |
| 728 | .and_then(|entry| entry.api_key_env.as_deref()) |
| 729 | .map(str::trim) |
| 730 | .filter(|name| !name.is_empty()) |
| 731 | .is_some_and(|name| std::env::var(name).is_ok_and(|value| !value.trim().is_empty())) |
| 732 | { |
| 733 | return Some(ModelAuthSource::Env); |
| 734 | } |
| 735 | if !config.should_skip_secret_store_for_provider(identity) && env_has_key_for(provider) { |
| 736 | return Some(ModelAuthSource::Env); |
| 737 | } |
| 738 | Some(ModelAuthSource::Config) |
| 739 | } |
| 740 | |
| 741 | fn provider_uses_oauth_cli(config: &Config, identity: &ProviderIdentity) -> bool { |
| 742 | let provider = identity.provider; |
| 743 | if config.provider_uses_custom_endpoint(identity) { |
| 744 | return false; |
| 745 | } |
| 746 | match provider { |
| 747 | ProviderKind::OpenaiCodex => true, |
| 748 | ProviderKind::Xai => config |
| 749 | .provider_config_for(identity) |
| 750 | .and_then(|entry| entry.auth_mode.as_deref()) |
| 751 | .is_some_and(crate::oauth::auth_mode_uses_xai_oauth), |
| 752 | _ => false, |
| 753 | } |
| 754 | } |
| 755 | |
| 756 | fn env_has_key_for(provider: ProviderKind) -> bool { |
| 757 | env_keys_for_provider(provider) |
| 758 | .iter() |
| 759 | .any(|key| std::env::var(key).is_ok_and(|value| !value.trim().is_empty())) |
| 760 | } |
| 761 | |
| 762 | fn env_keys_for_provider(provider: ProviderKind) -> &'static [&'static str] { |
| 763 | provider.provider().env_vars() |
| 764 | } |
| 765 | |
| 766 | #[cfg(test)] |
| 767 | mod tests { |
| 768 | use super::*; |
| 769 | |
| 770 | #[test] |
| 771 | fn inventory_env_keys_follow_provider_metadata() { |
| 772 | for provider in ProviderKind::all() { |
| 773 | assert_eq!( |
| 774 | env_keys_for_provider(*provider), |
| 775 | provider.provider().env_vars() |
| 776 | ); |
| 777 | } |
| 778 | } |
| 779 | |
| 780 | #[test] |
| 781 | fn inventory_includes_only_usable_authenticated_providers() { |
| 782 | let _env_lock = crate::test_support::lock_test_env(); |
| 783 | let _deepseek = crate::test_support::EnvVarGuard::set("DEEPSEEK_API_KEY", "ds-key"); |
| 784 | let _zai = crate::test_support::EnvVarGuard::set("ZAI_API_KEY", "zai-key"); |
| 785 | let _minimax = crate::test_support::EnvVarGuard::remove("MINIMAX_API_KEY"); |
| 786 | let config = Config { |
| 787 | provider: Some("zai".to_string()), |
| 788 | default_text_model: Some("deepseek-v4-pro".to_string()), |
| 789 | ..Default::default() |
| 790 | }; |
| 791 | |
| 792 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 793 | |
| 794 | // A DeepSeek key alone no longer elects a network classifier: with no |
| 795 | // explicit `[auto.router]`, legacy Auto stays local/free. |
| 796 | assert!(!inventory.router_configured); |
| 797 | assert!(!inventory.router_available); |
| 798 | assert!( |
| 799 | inventory |
| 800 | .candidate(ProviderKind::Zai.as_str(), crate::config::ZAI_GLM_5_2_MODEL) |
| 801 | .is_some() |
| 802 | ); |
| 803 | assert!( |
| 804 | inventory |
| 805 | .candidates |
| 806 | .iter() |
| 807 | .all(|candidate| candidate.provider != ProviderKind::Minimax) |
| 808 | ); |
| 809 | } |
| 810 | |
| 811 | #[test] |
| 812 | fn inventory_marks_local_providers_keyless() { |
| 813 | let _env_lock = crate::test_support::lock_test_env(); |
| 814 | let _deepseek = crate::test_support::EnvVarGuard::remove("DEEPSEEK_API_KEY"); |
| 815 | let mut config = Config::default(); |
| 816 | config |
| 817 | .set_provider_model_override( |
| 818 | &config.test_identity_for_kind(ProviderKind::Ollama), |
| 819 | Some("local-tag:latest".into()), |
| 820 | ) |
| 821 | .unwrap(); |
| 822 | |
| 823 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 824 | |
| 825 | assert!( |
| 826 | inventory |
| 827 | .candidates |
| 828 | .iter() |
| 829 | .any(|candidate| candidate.provider == ProviderKind::Ollama |
| 830 | && candidate.auth_source == ModelAuthSource::KeylessLocal) |
| 831 | ); |
| 832 | } |
| 833 | |
| 834 | #[test] |
| 835 | fn inventory_never_marks_ollama_cloud_keyless_or_local() { |
| 836 | let _env_lock = crate::test_support::lock_test_env(); |
| 837 | let _cloud_env = crate::test_support::EnvVarGuard::remove("OLLAMA_CLOUD_API_KEY"); |
| 838 | let _official_env = crate::test_support::EnvVarGuard::remove("OLLAMA_API_KEY"); |
| 839 | let config = Config { |
| 840 | provider: Some("ollama-cloud".to_string()), |
| 841 | providers: Some(crate::config::ProvidersConfig { |
| 842 | ollama_cloud: crate::config::ProviderConfig { |
| 843 | api_key: Some("cloud-key".to_string()), |
| 844 | ..Default::default() |
| 845 | }, |
| 846 | ..Default::default() |
| 847 | }), |
| 848 | ..Default::default() |
| 849 | }; |
| 850 | |
| 851 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 852 | let candidate = inventory |
| 853 | .candidate( |
| 854 | ProviderKind::OllamaCloud.as_str(), |
| 855 | crate::config::DEFAULT_OLLAMA_CLOUD_MODEL, |
| 856 | ) |
| 857 | .expect("authenticated Ollama Cloud candidate"); |
| 858 | assert_eq!(candidate.auth_source, ModelAuthSource::Config); |
| 859 | assert!(!candidate.tags.contains(&"local")); |
| 860 | assert_ne!(candidate.readiness.label(), "local · not checked"); |
| 861 | } |
| 862 | |
| 863 | #[test] |
| 864 | fn inventory_never_admits_kimi_cli_oauth_import() { |
| 865 | let _env_lock = crate::test_support::lock_test_env(); |
| 866 | let temp = tempfile::tempdir().expect("Kimi import fixture root"); |
| 867 | let kimi_home = temp.path().join("kimi-code"); |
| 868 | std::fs::create_dir_all(kimi_home.join("credentials")).expect("Kimi credential directory"); |
| 869 | let expires_at = std::time::SystemTime::now() |
| 870 | .duration_since(std::time::UNIX_EPOCH) |
| 871 | .expect("clock after epoch") |
| 872 | .as_secs_f64() |
| 873 | + 3600.0; |
| 874 | let credential_path = kimi_home.join("credentials/kimi-code.json"); |
| 875 | let credential_raw = serde_json::json!({ |
| 876 | "access_token": "unexpired-user-owned-token", |
| 877 | "refresh_token": "must-not-be-used", |
| 878 | "expires_at": expires_at, |
| 879 | }) |
| 880 | .to_string(); |
| 881 | std::fs::write(&credential_path, &credential_raw).expect("write Kimi import fixture"); |
| 882 | let _kimi_home = crate::test_support::EnvVarGuard::set( |
| 883 | "KIMI_CODE_HOME", |
| 884 | kimi_home.to_str().expect("utf8 path"), |
| 885 | ); |
| 886 | let config = Config { |
| 887 | provider: Some("moonshot".to_string()), |
| 888 | providers: Some(crate::config::ProvidersConfig { |
| 889 | moonshot: crate::config::ProviderConfig { |
| 890 | auth_mode: Some("kimi_oauth".to_string()), |
| 891 | ..Default::default() |
| 892 | }, |
| 893 | ..Default::default() |
| 894 | }), |
| 895 | ..Default::default() |
| 896 | }; |
| 897 | |
| 898 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 899 | assert!( |
| 900 | inventory |
| 901 | .candidates |
| 902 | .iter() |
| 903 | .all(|candidate| candidate.provider != ProviderKind::Moonshot), |
| 904 | "unsupported Kimi CLI OAuth must not enter the routing inventory" |
| 905 | ); |
| 906 | assert_eq!( |
| 907 | std::fs::read_to_string(credential_path).expect("Kimi file remains untouched"), |
| 908 | credential_raw |
| 909 | ); |
| 910 | } |
| 911 | |
| 912 | #[test] |
| 913 | fn inventory_uses_kimi_code_k3_route_context_not_generic_fallback() { |
| 914 | let config = Config { |
| 915 | provider: Some("moonshot".to_string()), |
| 916 | providers: Some(crate::config::ProvidersConfig { |
| 917 | moonshot: crate::config::ProviderConfig { |
| 918 | api_key: Some("test-kimi-key".to_string()), |
| 919 | base_url: Some(crate::config::DEFAULT_KIMI_CODE_BASE_URL.to_string()), |
| 920 | model: Some(crate::config::KIMI_CODE_K3_MODEL.to_string()), |
| 921 | ..Default::default() |
| 922 | }, |
| 923 | ..Default::default() |
| 924 | }), |
| 925 | ..Default::default() |
| 926 | }; |
| 927 | |
| 928 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 929 | let candidate = inventory |
| 930 | .candidate( |
| 931 | ProviderKind::Moonshot.as_str(), |
| 932 | crate::config::KIMI_CODE_K3_MODEL, |
| 933 | ) |
| 934 | .expect("configured Kimi Code K3 route"); |
| 935 | |
| 936 | assert_eq!(candidate.context_window, 262_144); |
| 937 | assert!(candidate.thinking_supported); |
| 938 | assert!(candidate.tags.contains(&"thinking")); |
| 939 | assert!(!candidate.tags.contains(&"long_context")); |
| 940 | } |
| 941 | |
| 942 | /// #5441: the auto-router inventory carries a `_Nk` name-suffix window |
| 943 | /// together with the fact that nobody verified it, so a classifier never |
| 944 | /// reads a naming convention as a route capability. |
| 945 | #[test] |
| 946 | fn router_inventory_marks_name_suffix_windows_unverified() { |
| 947 | let config = Config { |
| 948 | provider: Some("vllm".to_string()), |
| 949 | providers: Some(crate::config::ProvidersConfig { |
| 950 | vllm: crate::config::ProviderConfig { |
| 951 | base_url: Some("http://localhost:8000/v1".to_string()), |
| 952 | model: Some("qwen3-32b-256k".to_string()), |
| 953 | ..Default::default() |
| 954 | }, |
| 955 | ..Default::default() |
| 956 | }), |
| 957 | ..Default::default() |
| 958 | }; |
| 959 | |
| 960 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 961 | let candidate = inventory |
| 962 | .candidate(ProviderKind::Vllm.as_str(), "qwen3-32b-256k") |
| 963 | .expect("configured self-hosted route"); |
| 964 | assert_eq!(candidate.context_window, 256_000); |
| 965 | assert!( |
| 966 | candidate.context_window_unverified, |
| 967 | "a name-suffix window must not enter the router payload as a fact" |
| 968 | ); |
| 969 | |
| 970 | let payload = inventory.router_context_json(); |
| 971 | assert!( |
| 972 | payload.contains("\"context_window_unverified\":true"), |
| 973 | "payload must serialize the marker: {payload}" |
| 974 | ); |
| 975 | } |
| 976 | |
| 977 | #[test] |
| 978 | fn inventory_includes_custom_api_key_env_route() { |
| 979 | let _env_lock = crate::test_support::lock_test_env(); |
| 980 | let _custom_key = crate::test_support::EnvVarGuard::set("ACME_CUSTOM_KEY", "custom-key"); |
| 981 | let config = Config { |
| 982 | provider: Some("acme".to_string()), |
| 983 | providers: Some(crate::config::ProvidersConfig { |
| 984 | custom: std::collections::HashMap::from([( |
| 985 | "acme".to_string(), |
| 986 | crate::config::ProviderConfig { |
| 987 | kind: Some("openai-compatible".to_string()), |
| 988 | base_url: Some("https://api.acme.test/v1".to_string()), |
| 989 | model: Some("acme-coder".to_string()), |
| 990 | api_key_env: Some("ACME_CUSTOM_KEY".to_string()), |
| 991 | ..Default::default() |
| 992 | }, |
| 993 | )]), |
| 994 | ..Default::default() |
| 995 | }), |
| 996 | ..Default::default() |
| 997 | }; |
| 998 | |
| 999 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 1000 | assert!( |
| 1001 | inventory |
| 1002 | .candidates |
| 1003 | .iter() |
| 1004 | .any(|candidate| candidate.provider == ProviderKind::Custom |
| 1005 | && candidate.model == "acme-coder" |
| 1006 | && candidate.auth_source == ModelAuthSource::Env) |
| 1007 | ); |
| 1008 | } |
| 1009 | |
| 1010 | #[test] |
| 1011 | fn inventory_router_timeout_secs_respects_config_with_clamp() { |
| 1012 | let _env_lock = crate::test_support::lock_test_env(); |
| 1013 | |
| 1014 | // Unset: the legacy default (4 s) survives. |
| 1015 | let config = Config { |
| 1016 | ..Default::default() |
| 1017 | }; |
| 1018 | assert_eq!( |
| 1019 | ModelInventory::from_config(&config) |
| 1020 | .unwrap() |
| 1021 | .router_timeout_secs, |
| 1022 | 4 |
| 1023 | ); |
| 1024 | |
| 1025 | // Explicit value is honored. |
| 1026 | let config = Config { |
| 1027 | auto: Some(crate::config::AutoConfig { |
| 1028 | router: Some(crate::config::AutoRouterConfig { |
| 1029 | provider: Some("custom".to_string()), |
| 1030 | model: Some("local-router".to_string()), |
| 1031 | thinking: None, |
| 1032 | timeout_secs: Some(15), |
| 1033 | ..Default::default() |
| 1034 | }), |
| 1035 | ..Default::default() |
| 1036 | }), |
| 1037 | ..Default::default() |
| 1038 | }; |
| 1039 | assert_eq!( |
| 1040 | ModelInventory::from_config(&config) |
| 1041 | .unwrap() |
| 1042 | .router_timeout_secs, |
| 1043 | 15 |
| 1044 | ); |
| 1045 | |
| 1046 | // Out-of-range values clamp to the safety ceiling, never to zero. |
| 1047 | let config = Config { |
| 1048 | auto: Some(crate::config::AutoConfig { |
| 1049 | router: Some(crate::config::AutoRouterConfig { |
| 1050 | provider: Some("custom".to_string()), |
| 1051 | model: Some("local-router".to_string()), |
| 1052 | thinking: None, |
| 1053 | timeout_secs: Some(9_999), |
| 1054 | ..Default::default() |
| 1055 | }), |
| 1056 | ..Default::default() |
| 1057 | }), |
| 1058 | ..Default::default() |
| 1059 | }; |
| 1060 | assert_eq!( |
| 1061 | ModelInventory::from_config(&config) |
| 1062 | .unwrap() |
| 1063 | .router_timeout_secs, |
| 1064 | crate::config::MAX_AUTO_ROUTER_TIMEOUT_SECS |
| 1065 | ); |
| 1066 | |
| 1067 | // Zero means "use the default", not an instant timeout. |
| 1068 | let config = Config { |
| 1069 | auto: Some(crate::config::AutoConfig { |
| 1070 | router: Some(crate::config::AutoRouterConfig { |
| 1071 | provider: Some("custom".to_string()), |
| 1072 | model: Some("local-router".to_string()), |
| 1073 | thinking: None, |
| 1074 | timeout_secs: Some(0), |
| 1075 | ..Default::default() |
| 1076 | }), |
| 1077 | ..Default::default() |
| 1078 | }), |
| 1079 | ..Default::default() |
| 1080 | }; |
| 1081 | assert_eq!( |
| 1082 | ModelInventory::from_config(&config) |
| 1083 | .unwrap() |
| 1084 | .router_timeout_secs, |
| 1085 | 4 |
| 1086 | ); |
| 1087 | } |
| 1088 | |
| 1089 | #[test] |
| 1090 | fn inventory_ignores_unresolved_command_and_secret_auth_metadata() { |
| 1091 | let _env_lock = crate::test_support::lock_test_env(); |
| 1092 | let temp = tempfile::tempdir().expect("isolated credential home"); |
| 1093 | let _home = crate::test_support::EnvVarGuard::set("CODEWHALE_HOME", temp.path()); |
| 1094 | let _backend = crate::test_support::EnvVarGuard::set("CODEWHALE_SECRET_BACKEND", "file"); |
| 1095 | let _deepseek = crate::test_support::EnvVarGuard::remove("DEEPSEEK_API_KEY"); |
| 1096 | let _openai = crate::test_support::EnvVarGuard::remove("OPENAI_API_KEY"); |
| 1097 | let _xai = crate::test_support::EnvVarGuard::remove("XAI_API_KEY"); |
| 1098 | let mut providers = crate::config::ProvidersConfig::default(); |
| 1099 | providers.openai.auth = Some(codewhale_config::ProviderAuthSourceToml { |
| 1100 | source: codewhale_config::AuthSourceKind::Command, |
| 1101 | command: vec!["secret-tool".to_string(), "lookup".to_string()], |
| 1102 | timeout_ms: Some(2000), |
| 1103 | secret_id: None, |
| 1104 | }); |
| 1105 | providers.xai.auth = Some(codewhale_config::ProviderAuthSourceToml { |
| 1106 | source: codewhale_config::AuthSourceKind::Secret, |
| 1107 | command: Vec::new(), |
| 1108 | timeout_ms: None, |
| 1109 | secret_id: Some("codewhale/xai".to_string()), |
| 1110 | }); |
| 1111 | let config = Config { |
| 1112 | provider: Some("openai".to_string()), |
| 1113 | providers: Some(providers), |
| 1114 | ..Default::default() |
| 1115 | }; |
| 1116 | |
| 1117 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 1118 | assert!(inventory.candidates.iter().all(|candidate| !matches!( |
| 1119 | candidate.provider, |
| 1120 | ProviderKind::Openai | ProviderKind::Xai |
| 1121 | ))); |
| 1122 | } |
| 1123 | |
| 1124 | #[test] |
| 1125 | fn auto_router_config_overrides_default_classifier_route() { |
| 1126 | let config = Config { |
| 1127 | auto: Some(crate::config::AutoConfig { |
| 1128 | cost_saving: None, |
| 1129 | cross_provider: None, |
| 1130 | router: Some(crate::config::AutoRouterConfig { |
| 1131 | provider: Some("zai".to_string()), |
| 1132 | model: Some("glm-5-turbo".to_string()), |
| 1133 | thinking: Some("low".to_string()), |
| 1134 | timeout_secs: None, |
| 1135 | ..Default::default() |
| 1136 | }), |
| 1137 | }), |
| 1138 | ..Default::default() |
| 1139 | }; |
| 1140 | |
| 1141 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 1142 | assert!(inventory.router_configured); |
| 1143 | assert_eq!(inventory.router_provider, ProviderKind::Zai); |
| 1144 | assert_eq!(inventory.router_model, "glm-5-turbo"); |
| 1145 | assert_eq!(inventory.router_thinking.as_deref(), Some("low")); |
| 1146 | } |
| 1147 | |
| 1148 | /// A DeepSeek key must never, on its own, turn on a network classifier. |
| 1149 | /// `[auto.router]` stays legacy `model = auto` configuration; absent it, |
| 1150 | /// legacy Auto is local/free. |
| 1151 | #[test] |
| 1152 | fn a_deepseek_key_alone_never_elects_an_implicit_flash_classifier() { |
| 1153 | let _env_lock = crate::test_support::lock_test_env(); |
| 1154 | let _deepseek = crate::test_support::EnvVarGuard::set("DEEPSEEK_API_KEY", "ds-key"); |
| 1155 | let config = Config { |
| 1156 | provider: Some("deepseek".to_string()), |
| 1157 | ..Default::default() |
| 1158 | }; |
| 1159 | |
| 1160 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 1161 | |
| 1162 | assert!( |
| 1163 | !inventory.router_configured, |
| 1164 | "no [auto.router] means no configured classifier" |
| 1165 | ); |
| 1166 | assert!( |
| 1167 | !inventory.router_available, |
| 1168 | "holding a DeepSeek key must not silently select deepseek-v4-flash as a classifier" |
| 1169 | ); |
| 1170 | } |
| 1171 | |
| 1172 | #[test] |
| 1173 | fn an_explicit_legacy_auto_router_still_works_when_its_key_is_present() { |
| 1174 | let _env_lock = crate::test_support::lock_test_env(); |
| 1175 | let _zai = crate::test_support::EnvVarGuard::set("ZAI_API_KEY", "zai-key"); |
| 1176 | let config = Config { |
| 1177 | auto: Some(crate::config::AutoConfig { |
| 1178 | cost_saving: None, |
| 1179 | router: Some(crate::config::AutoRouterConfig { |
| 1180 | provider: Some("zai".to_string()), |
| 1181 | model: Some("glm-5-turbo".to_string()), |
| 1182 | thinking: None, |
| 1183 | timeout_secs: None, |
| 1184 | ..Default::default() |
| 1185 | }), |
| 1186 | cross_provider: None, |
| 1187 | }), |
| 1188 | ..Default::default() |
| 1189 | }; |
| 1190 | |
| 1191 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 1192 | |
| 1193 | assert!(inventory.router_configured); |
| 1194 | assert!(inventory.router_available); |
| 1195 | assert_eq!(inventory.router_model, "glm-5-turbo"); |
| 1196 | } |
| 1197 | |
| 1198 | #[test] |
| 1199 | fn inventory_marks_explicit_no_auth_separately_from_keyless_local() { |
| 1200 | let mut providers = crate::config::ProvidersConfig::default(); |
| 1201 | providers.vllm.auth_mode = Some("none".to_string()); |
| 1202 | providers.vllm.model = Some("local-model".to_string()); |
| 1203 | let config = Config { |
| 1204 | provider: Some("vllm".to_string()), |
| 1205 | providers: Some(providers), |
| 1206 | ..Default::default() |
| 1207 | }; |
| 1208 | |
| 1209 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 1210 | let candidate = inventory |
| 1211 | .candidates |
| 1212 | .iter() |
| 1213 | .find(|candidate| { |
| 1214 | candidate.provider == ProviderKind::Vllm && candidate.model == "local-model" |
| 1215 | }) |
| 1216 | .expect("vLLM no-auth candidate"); |
| 1217 | |
| 1218 | assert_eq!(candidate.auth_source, ModelAuthSource::NoAuth); |
| 1219 | assert_eq!( |
| 1220 | candidate.readiness, |
| 1221 | crate::provider_readiness::ResolvedProviderReadiness::NoAuthUnchecked |
| 1222 | ); |
| 1223 | } |
| 1224 | |
| 1225 | #[test] |
| 1226 | fn unready_candidates_are_never_provider_defaults() { |
| 1227 | use crate::provider_readiness::ResolvedProviderReadiness; |
| 1228 | |
| 1229 | let candidate = ModelRouteCandidate { |
| 1230 | provider: ProviderKind::Openai, |
| 1231 | provider_tag: "openai".into(), |
| 1232 | identity: Config::default() |
| 1233 | .builtin_provider_identity(ProviderKind::Openai) |
| 1234 | .unwrap(), |
| 1235 | provider_name: "openai".into(), |
| 1236 | provider_display_name: "OpenAI".into(), |
| 1237 | model: "gpt-5.5".to_string(), |
| 1238 | context_window: 128_000, |
| 1239 | context_window_unverified: false, |
| 1240 | user_declared: false, |
| 1241 | max_output: Some(16_384), |
| 1242 | thinking_supported: true, |
| 1243 | cache_telemetry_supported: false, |
| 1244 | auth_source: ModelAuthSource::Config, |
| 1245 | readiness: ResolvedProviderReadiness::MissingLogin, |
| 1246 | default_for_provider: false, |
| 1247 | tags: vec!["unready"], |
| 1248 | }; |
| 1249 | assert!(!candidate.readiness.can_attempt()); |
| 1250 | assert!(!candidate.default_for_provider); |
| 1251 | assert!(candidate.tags.contains(&"unready")); |
| 1252 | } |
| 1253 | |
| 1254 | #[test] |
| 1255 | fn active_default_never_falls_back_to_unready_candidate() { |
| 1256 | let inventory = ModelInventory { |
| 1257 | active_provider: ProviderKind::Openai, |
| 1258 | active_identity: Config::default() |
| 1259 | .builtin_provider_identity(ProviderKind::Openai) |
| 1260 | .unwrap(), |
| 1261 | router_provider: ProviderKind::Deepseek, |
| 1262 | router_identity: None, |
| 1263 | router_model: "deepseek-v4-flash".to_string(), |
| 1264 | router_thinking: None, |
| 1265 | router_timeout_secs: 4, |
| 1266 | router_configured: false, |
| 1267 | router_available: false, |
| 1268 | router_kind: AutoRouterKind::Chat, |
| 1269 | router_decision_route: None, |
| 1270 | router_base_url: None, |
| 1271 | router_min_confidence_bp: 5_000, |
| 1272 | router_setup_issue: None, |
| 1273 | cross_provider_auto: false, |
| 1274 | candidates: vec![ModelRouteCandidate { |
| 1275 | provider: ProviderKind::Openai, |
| 1276 | provider_tag: "openai".into(), |
| 1277 | identity: Config::default() |
| 1278 | .builtin_provider_identity(ProviderKind::Openai) |
| 1279 | .unwrap(), |
| 1280 | provider_name: "openai".into(), |
| 1281 | provider_display_name: "OpenAI".into(), |
| 1282 | model: "unsupported-model".to_string(), |
| 1283 | context_window: 1, |
| 1284 | context_window_unverified: false, |
| 1285 | user_declared: false, |
| 1286 | max_output: Some(1), |
| 1287 | thinking_supported: false, |
| 1288 | cache_telemetry_supported: false, |
| 1289 | auth_source: ModelAuthSource::Config, |
| 1290 | readiness: crate::provider_readiness::ResolvedProviderReadiness::InvalidRoute, |
| 1291 | default_for_provider: false, |
| 1292 | tags: vec!["unready"], |
| 1293 | }], |
| 1294 | }; |
| 1295 | |
| 1296 | assert!(inventory.active_default().is_none()); |
| 1297 | } |
| 1298 | |
| 1299 | #[test] |
| 1300 | fn router_context_is_runnable_and_redacts_auth_and_failure_details() { |
| 1301 | let _env_lock = crate::test_support::lock_test_env(); |
| 1302 | let _deepseek = crate::test_support::EnvVarGuard::set("DEEPSEEK_API_KEY", "ds-key"); |
| 1303 | let mut inventory = ModelInventory::from_config(&Config::default()).unwrap(); |
| 1304 | let candidate = inventory |
| 1305 | .candidates |
| 1306 | .iter_mut() |
| 1307 | .find(|candidate| candidate.provider == ProviderKind::Deepseek) |
| 1308 | .expect("DeepSeek inventory candidate"); |
| 1309 | candidate.readiness = |
| 1310 | crate::provider_readiness::ResolvedProviderReadiness::SavedLastCheckFailed { |
| 1311 | category: crate::error_taxonomy::ErrorCategory::Authentication, |
| 1312 | message: "Bearer super-secret-router-token".to_string(), |
| 1313 | }; |
| 1314 | inventory.candidates.push(ModelRouteCandidate { |
| 1315 | provider: ProviderKind::Openai, |
| 1316 | provider_tag: "openai".into(), |
| 1317 | identity: Config::default() |
| 1318 | .builtin_provider_identity(ProviderKind::Openai) |
| 1319 | .unwrap(), |
| 1320 | provider_name: "openai".into(), |
| 1321 | provider_display_name: "OpenAI".into(), |
| 1322 | model: "unsupported-model".to_string(), |
| 1323 | context_window: 1, |
| 1324 | context_window_unverified: false, |
| 1325 | user_declared: false, |
| 1326 | max_output: Some(1), |
| 1327 | thinking_supported: false, |
| 1328 | cache_telemetry_supported: false, |
| 1329 | auth_source: ModelAuthSource::Config, |
| 1330 | readiness: crate::provider_readiness::ResolvedProviderReadiness::InvalidRoute, |
| 1331 | default_for_provider: false, |
| 1332 | tags: vec!["unready"], |
| 1333 | }); |
| 1334 | |
| 1335 | let json = inventory.router_context_json(); |
| 1336 | |
| 1337 | assert!(json.contains("deepseek-v4")); |
| 1338 | assert!(!json.contains("super-secret-router-token")); |
| 1339 | assert!(!json.contains("auth_source")); |
| 1340 | assert!(!json.contains("unsupported-model")); |
| 1341 | } |
| 1342 | |
| 1343 | #[test] |
| 1344 | fn router_context_names_only_the_active_provider_by_default() { |
| 1345 | // #4411: a Z.ai session with a DeepSeek key in the environment must |
| 1346 | // not disclose the DeepSeek routes — or the fact that a DeepSeek |
| 1347 | // credential exists — to the classifier. |
| 1348 | let _env_lock = crate::test_support::lock_test_env(); |
| 1349 | let _deepseek = crate::test_support::EnvVarGuard::set("DEEPSEEK_API_KEY", "ds-key"); |
| 1350 | let _zai = crate::test_support::EnvVarGuard::set("ZAI_API_KEY", "zai-key"); |
| 1351 | let config = Config { |
| 1352 | provider: Some("zai".to_string()), |
| 1353 | ..Default::default() |
| 1354 | }; |
| 1355 | |
| 1356 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 1357 | assert!( |
| 1358 | inventory |
| 1359 | .candidates |
| 1360 | .iter() |
| 1361 | .any(|candidate| candidate.provider == ProviderKind::Deepseek), |
| 1362 | "the full inventory still knows about DeepSeek for pickers/explicit routes" |
| 1363 | ); |
| 1364 | |
| 1365 | let json = inventory.router_context_json(); |
| 1366 | let payload: serde_json::Value = |
| 1367 | serde_json::from_str(&json).expect("router context is JSON"); |
| 1368 | let providers: Vec<&str> = payload["candidates"] |
| 1369 | .as_array() |
| 1370 | .expect("candidate array") |
| 1371 | .iter() |
| 1372 | .map(|candidate| candidate["provider_name"].as_str().expect("provider name")) |
| 1373 | .collect(); |
| 1374 | |
| 1375 | assert!(!providers.is_empty(), "active provider routes must remain"); |
| 1376 | assert!( |
| 1377 | providers.iter().all(|provider| *provider == "zai"), |
| 1378 | "classifier payload leaked another provider: {json}" |
| 1379 | ); |
| 1380 | assert!(!json.contains("deepseek"), "{json}"); |
| 1381 | } |
| 1382 | |
| 1383 | #[test] |
| 1384 | fn router_context_includes_other_providers_under_persisted_opt_in() { |
| 1385 | let _env_lock = crate::test_support::lock_test_env(); |
| 1386 | let _deepseek = crate::test_support::EnvVarGuard::set("DEEPSEEK_API_KEY", "ds-key"); |
| 1387 | let _zai = crate::test_support::EnvVarGuard::set("ZAI_API_KEY", "zai-key"); |
| 1388 | let config = Config { |
| 1389 | provider: Some("zai".to_string()), |
| 1390 | auto: Some(crate::config::AutoConfig { |
| 1391 | cost_saving: None, |
| 1392 | cross_provider: Some(true), |
| 1393 | router: None, |
| 1394 | }), |
| 1395 | ..Default::default() |
| 1396 | }; |
| 1397 | |
| 1398 | let json = ModelInventory::from_config(&config) |
| 1399 | .unwrap() |
| 1400 | .router_context_json(); |
| 1401 | |
| 1402 | assert!(json.contains("\"zai\""), "{json}"); |
| 1403 | assert!(json.contains("deepseek"), "{json}"); |
| 1404 | } |
| 1405 | |
| 1406 | #[test] |
| 1407 | fn implicit_deepseek_classifier_is_out_of_scope_for_another_active_provider() { |
| 1408 | // #4411: the default classifier route is DeepSeek flash. Calling it |
| 1409 | // from a Z.ai session would send the turn's prompt to a second |
| 1410 | // provider, so it stays unavailable without an explicit opt-in. |
| 1411 | let _env_lock = crate::test_support::lock_test_env(); |
| 1412 | let _deepseek = crate::test_support::EnvVarGuard::set("DEEPSEEK_API_KEY", "ds-key"); |
| 1413 | let _zai = crate::test_support::EnvVarGuard::set("ZAI_API_KEY", "zai-key"); |
| 1414 | let zai = Config { |
| 1415 | provider: Some("zai".to_string()), |
| 1416 | ..Default::default() |
| 1417 | }; |
| 1418 | assert!(!ModelInventory::from_config(&zai).unwrap().router_available); |
| 1419 | |
| 1420 | // `cross_provider = true` widens which candidates Auto may pick; it is |
| 1421 | // NOT a classifier election. With the implicit DeepSeek-flash default |
| 1422 | // removed, no network classifier runs without an explicit |
| 1423 | // `[auto.router]` route — a scope opt-in alone stays local/free. |
| 1424 | let opted_in = Config { |
| 1425 | auto: Some(crate::config::AutoConfig { |
| 1426 | cost_saving: None, |
| 1427 | cross_provider: Some(true), |
| 1428 | router: None, |
| 1429 | }), |
| 1430 | ..zai.clone() |
| 1431 | }; |
| 1432 | let widened = ModelInventory::from_config(&opted_in).unwrap(); |
| 1433 | assert!(!widened.router_available); |
| 1434 | assert!(widened.auto_scope_allows(ProviderKind::Deepseek.as_str())); |
| 1435 | |
| 1436 | // An explicitly configured `[auto.router]` is itself a persisted |
| 1437 | // opt-in for that classifier route. |
| 1438 | let explicit_router = Config { |
| 1439 | auto: Some(crate::config::AutoConfig { |
| 1440 | cost_saving: None, |
| 1441 | cross_provider: None, |
| 1442 | router: Some(crate::config::AutoRouterConfig { |
| 1443 | provider: Some("deepseek".to_string()), |
| 1444 | model: Some("deepseek-v4-flash".to_string()), |
| 1445 | thinking: None, |
| 1446 | timeout_secs: None, |
| 1447 | ..Default::default() |
| 1448 | }), |
| 1449 | }), |
| 1450 | ..zai.clone() |
| 1451 | }; |
| 1452 | assert!( |
| 1453 | ModelInventory::from_config(&explicit_router) |
| 1454 | .unwrap() |
| 1455 | .router_available |
| 1456 | ); |
| 1457 | |
| 1458 | // A DeepSeek session gets no free classifier either: with the |
| 1459 | // implicit flash default removed, only an explicit `[auto.router]` |
| 1460 | // elects a network classifier, active provider or not. |
| 1461 | let deepseek = Config { |
| 1462 | provider: Some("deepseek".to_string()), |
| 1463 | ..Default::default() |
| 1464 | }; |
| 1465 | assert!( |
| 1466 | !ModelInventory::from_config(&deepseek) |
| 1467 | .unwrap() |
| 1468 | .router_available |
| 1469 | ); |
| 1470 | } |
| 1471 | |
| 1472 | #[test] |
| 1473 | fn declared_inventory_ids_remain_case_distinct() { |
| 1474 | let _env = crate::test_support::lock_test_env(); |
| 1475 | let home = tempfile::tempdir().unwrap(); |
| 1476 | let _home = crate::test_support::EnvVarGuard::set("CODEWHALE_HOME", home.path()); |
| 1477 | let mut config: Config = toml::from_str(include_str!( |
| 1478 | "../../config/tests/fixtures/custom_models.toml" |
| 1479 | )) |
| 1480 | .unwrap(); |
| 1481 | let declaration = config.custom_models.as_mut().unwrap().first_mut().unwrap(); |
| 1482 | declaration.id = "Preview-fixture".into(); |
| 1483 | let mut other = declaration.clone(); |
| 1484 | other.id = "preview-fixture".into(); |
| 1485 | other.limit.as_mut().unwrap().context = Some(128000); |
| 1486 | config.custom_models.as_mut().unwrap().push(other); |
| 1487 | config |
| 1488 | .set_provider_api_key_override( |
| 1489 | &config.test_identity_for_kind(ProviderKind::Deepseek), |
| 1490 | Some("fixture-key".into()), |
| 1491 | ) |
| 1492 | .unwrap(); |
| 1493 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 1494 | for (id, context) in [("Preview-fixture", 96000), ("preview-fixture", 128000)] { |
| 1495 | let candidate = inventory |
| 1496 | .candidate(ProviderKind::Deepseek.as_str(), id) |
| 1497 | .unwrap(); |
| 1498 | assert!(candidate.user_declared); |
| 1499 | assert_eq!(candidate.model, id); |
| 1500 | assert_eq!(candidate.context_window, context); |
| 1501 | } |
| 1502 | assert!( |
| 1503 | inventory |
| 1504 | .candidate(ProviderKind::Deepseek.as_str(), "PREVIEW-FIXTURE") |
| 1505 | .is_none() |
| 1506 | ); |
| 1507 | } |
| 1508 | |
| 1509 | #[test] |
| 1510 | fn ollama_inventory_default_uses_only_the_fresh_exact_endpoint_roster() { |
| 1511 | use codewhale_config::catalog::{ |
| 1512 | CatalogOffering, CatalogRefreshError, CatalogSource, ProviderCatalogDelta, |
| 1513 | base_url_fingerprint, now_unix, |
| 1514 | }; |
| 1515 | |
| 1516 | let _env = crate::test_support::lock_test_env(); |
| 1517 | let _live = crate::provider_lake::lock_live_snapshot(); |
| 1518 | let home = tempfile::tempdir().unwrap(); |
| 1519 | let _home = crate::test_support::EnvVarGuard::set("CODEWHALE_HOME", home.path()); |
| 1520 | crate::provider_catalog_live::reset_cache_for_test(); |
| 1521 | crate::provider_lake::clear_live_snapshot(); |
| 1522 | let mut config = Config { |
| 1523 | provider: Some("ollama".to_string()), |
| 1524 | ..Default::default() |
| 1525 | }; |
| 1526 | let endpoint = "http://localhost:11445/v1"; |
| 1527 | config |
| 1528 | .provider_config_for_mut(&config.test_identity_for_kind(ProviderKind::Ollama)) |
| 1529 | .unwrap() |
| 1530 | .base_url = Some(endpoint.into()); |
| 1531 | assert_eq!( |
| 1532 | provider_default_model( |
| 1533 | &config, |
| 1534 | &(config).test_identity_for_kind(ProviderKind::Ollama) |
| 1535 | ), |
| 1536 | "unknown" |
| 1537 | ); |
| 1538 | assert!( |
| 1539 | ModelInventory::from_config(&config) |
| 1540 | .unwrap() |
| 1541 | .candidates |
| 1542 | .iter() |
| 1543 | .all(|row| { row.provider != ProviderKind::Ollama || row.model != "unknown" }) |
| 1544 | ); |
| 1545 | let fingerprint = base_url_fingerprint(endpoint); |
| 1546 | let now = now_unix(); |
| 1547 | let ticket = crate::provider_catalog_live::begin_refresh_for_identity( |
| 1548 | ProviderKind::Ollama, |
| 1549 | "ollama", |
| 1550 | endpoint, |
| 1551 | ); |
| 1552 | crate::provider_catalog_live::record_success_if_current( |
| 1553 | &ticket, |
| 1554 | ProviderCatalogDelta { |
| 1555 | provider: "ollama".into(), |
| 1556 | base_url_fingerprint: fingerprint.clone(), |
| 1557 | fetched_at: now, |
| 1558 | offerings: vec![CatalogOffering { |
| 1559 | provider: "ollama".into(), |
| 1560 | wire_model_id: "qwen2.5:0.5b".into(), |
| 1561 | endpoint_key: "chat".into(), |
| 1562 | source: CatalogSource::Live { |
| 1563 | base_url_fingerprint: fingerprint.clone(), |
| 1564 | fetched_at: now, |
| 1565 | }, |
| 1566 | ..Default::default() |
| 1567 | }], |
| 1568 | }, |
| 1569 | ); |
| 1570 | assert_eq!( |
| 1571 | provider_default_model( |
| 1572 | &config, |
| 1573 | &(config).test_identity_for_kind(ProviderKind::Ollama) |
| 1574 | ), |
| 1575 | "qwen2.5:0.5b" |
| 1576 | ); |
| 1577 | let inventory = ModelInventory::from_config(&config).unwrap(); |
| 1578 | assert!(inventory.candidates.iter().any(|row| { |
| 1579 | row.provider == ProviderKind::Ollama |
| 1580 | && row.model == "qwen2.5:0.5b" |
| 1581 | && row.default_for_provider |
| 1582 | })); |
| 1583 | let mut other = config.clone(); |
| 1584 | other |
| 1585 | .provider_config_for_mut(&other.test_identity_for_kind(ProviderKind::Ollama)) |
| 1586 | .unwrap() |
| 1587 | .base_url = Some("http://localhost:11446/v1".into()); |
| 1588 | assert_eq!( |
| 1589 | provider_default_model( |
| 1590 | &other, |
| 1591 | &(other).test_identity_for_kind(ProviderKind::Ollama) |
| 1592 | ), |
| 1593 | "unknown" |
| 1594 | ); |
| 1595 | crate::provider_catalog_live::record_failure_if_current( |
| 1596 | &ticket, |
| 1597 | "ollama", |
| 1598 | &fingerprint, |
| 1599 | CatalogRefreshError::Network, |
| 1600 | ); |
| 1601 | assert_eq!( |
| 1602 | provider_default_model( |
| 1603 | &config, |
| 1604 | &(config).test_identity_for_kind(ProviderKind::Ollama) |
| 1605 | ), |
| 1606 | "unknown" |
| 1607 | ); |
| 1608 | config |
| 1609 | .set_provider_model_override( |
| 1610 | &config.test_identity_for_kind(ProviderKind::Ollama), |
| 1611 | Some("chosen:tag".into()), |
| 1612 | ) |
| 1613 | .unwrap(); |
| 1614 | assert_eq!( |
| 1615 | provider_default_model( |
| 1616 | &config, |
| 1617 | &(config).test_identity_for_kind(ProviderKind::Ollama) |
| 1618 | ), |
| 1619 | "chosen:tag" |
| 1620 | ); |
| 1621 | crate::provider_catalog_live::reset_cache_for_test(); |
| 1622 | crate::provider_lake::clear_live_snapshot(); |
| 1623 | } |
| 1624 | } |
| 1625 | |
| 1626 | #[cfg(test)] |
| 1627 | mod decision_router_inventory_tests { |
| 1628 | use super::*; |
| 1629 | |
| 1630 | fn with_router(router: crate::config::AutoRouterConfig, openrouter_key: bool) -> Config { |
| 1631 | Config { |
| 1632 | provider: Some("deepseek".to_string()), |
| 1633 | providers: Some(crate::config::ProvidersConfig { |
| 1634 | openrouter: crate::config::ProviderConfig { |
| 1635 | api_key: openrouter_key.then(|| "or-test-key".to_string()), |
| 1636 | ..Default::default() |
| 1637 | }, |
| 1638 | ..Default::default() |
| 1639 | }), |
| 1640 | auto: Some(crate::config::AutoConfig { |
| 1641 | cost_saving: None, |
| 1642 | cross_provider: None, |
| 1643 | router: Some(router), |
| 1644 | }), |
| 1645 | ..Default::default() |
| 1646 | } |
| 1647 | } |
| 1648 | |
| 1649 | fn decision(provider: &str) -> crate::config::AutoRouterConfig { |
| 1650 | crate::config::AutoRouterConfig { |
| 1651 | kind: Some("decision".to_string()), |
| 1652 | provider: Some(provider.to_string()), |
| 1653 | model: Some("typesafe/jev-1.13".to_string()), |
| 1654 | ..Default::default() |
| 1655 | } |
| 1656 | } |
| 1657 | |
| 1658 | /// Tuple fields drop in order: restore the environment, then unlock. |
| 1659 | fn hermetic() -> ( |
| 1660 | crate::test_support::EnvVarGuard, |
| 1661 | crate::test_support::EnvVarGuard, |
| 1662 | crate::test_support::TestEnvLock, |
| 1663 | ) { |
| 1664 | let lock = crate::test_support::lock_test_env(); |
| 1665 | ( |
| 1666 | crate::test_support::EnvVarGuard::remove("OPENROUTER_API_KEY"), |
| 1667 | crate::test_support::EnvVarGuard::remove("TYPESAFE_API_KEY"), |
| 1668 | lock, |
| 1669 | ) |
| 1670 | } |
| 1671 | |
| 1672 | #[test] |
| 1673 | fn decision_router_on_openrouter_needs_the_openrouter_key() { |
| 1674 | let _env = hermetic(); |
| 1675 | let with_key = |
| 1676 | ModelInventory::from_config(&with_router(decision("openrouter"), true)).unwrap(); |
| 1677 | assert!(with_key.router_configured); |
| 1678 | assert!(with_key.router_available); |
| 1679 | assert_eq!(with_key.router_kind, AutoRouterKind::Decision); |
| 1680 | assert_eq!( |
| 1681 | with_key.router_decision_route, |
| 1682 | Some(DecisionRouterRoute::Openrouter) |
| 1683 | ); |
| 1684 | assert_eq!( |
| 1685 | with_key.router_thinking, None, |
| 1686 | "decision models ignore thinking" |
| 1687 | ); |
| 1688 | assert_eq!(with_key.router_min_confidence_bp, 5_000); |
| 1689 | assert_eq!(with_key.router_setup_issue, None); |
| 1690 | |
| 1691 | let without_key = |
| 1692 | ModelInventory::from_config(&with_router(decision("openrouter"), false)).unwrap(); |
| 1693 | assert!(without_key.router_configured); |
| 1694 | assert!(!without_key.router_available); |
| 1695 | assert_eq!( |
| 1696 | without_key.router_setup_issue, |
| 1697 | Some(AutoRouterSetupIssue::MissingKey) |
| 1698 | ); |
| 1699 | } |
| 1700 | |
| 1701 | #[test] |
| 1702 | fn unknown_kind_or_unsupported_decision_provider_is_not_configured() { |
| 1703 | let _env = hermetic(); |
| 1704 | let bogus = crate::config::AutoRouterConfig { |
| 1705 | kind: Some("bogus".to_string()), |
| 1706 | ..decision("openrouter") |
| 1707 | }; |
| 1708 | let inventory = ModelInventory::from_config(&with_router(bogus, true)).unwrap(); |
| 1709 | assert!(!inventory.router_configured); |
| 1710 | assert!(!inventory.router_available); |
| 1711 | assert_eq!( |
| 1712 | inventory.router_setup_issue, |
| 1713 | Some(AutoRouterSetupIssue::UnknownKind) |
| 1714 | ); |
| 1715 | |
| 1716 | let zai = ModelInventory::from_config(&with_router(decision("zai"), true)).unwrap(); |
| 1717 | assert!(!zai.router_configured); |
| 1718 | assert_eq!( |
| 1719 | zai.router_setup_issue, |
| 1720 | Some(AutoRouterSetupIssue::UnsupportedDecisionProvider) |
| 1721 | ); |
| 1722 | } |
| 1723 | |
| 1724 | #[test] |
| 1725 | fn a_chat_router_with_an_unknown_thinking_tier_is_not_configured() { |
| 1726 | let _env = hermetic(); |
| 1727 | let chat = |thinking: &str| crate::config::AutoRouterConfig { |
| 1728 | kind: Some("chat".to_string()), |
| 1729 | provider: Some("openrouter".to_string()), |
| 1730 | model: Some("openai/gpt-5-mini".to_string()), |
| 1731 | thinking: Some(thinking.to_string()), |
| 1732 | ..Default::default() |
| 1733 | }; |
| 1734 | let typo = ModelInventory::from_config(&with_router(chat("hgih"), true)).unwrap(); |
| 1735 | assert!(!typo.router_configured); |
| 1736 | assert!(!typo.router_available); |
| 1737 | assert_eq!( |
| 1738 | typo.router_setup_issue, |
| 1739 | Some(AutoRouterSetupIssue::InvalidThinking) |
| 1740 | ); |
| 1741 | |
| 1742 | let valid = ModelInventory::from_config(&with_router(chat("low"), true)).unwrap(); |
| 1743 | assert!(valid.router_available); |
| 1744 | assert_eq!(valid.router_setup_issue, None); |
| 1745 | assert_eq!(valid.router_thinking.as_deref(), Some("low")); |
| 1746 | } |
| 1747 | |
| 1748 | #[test] |
| 1749 | fn a_chat_router_whose_provider_cannot_serve_the_model_is_not_configured() { |
| 1750 | let _env = hermetic(); |
| 1751 | let chat = |provider: &str, model: &str| crate::config::AutoRouterConfig { |
| 1752 | kind: Some("chat".to_string()), |
| 1753 | provider: Some(provider.to_string()), |
| 1754 | model: Some(model.to_string()), |
| 1755 | ..Default::default() |
| 1756 | }; |
| 1757 | // A model from another provider's namespace, either direction. |
| 1758 | for (provider, model) in [("deepseek", "gpt-5-mini"), ("zai", "deepseek-v4-flash")] { |
| 1759 | let wrong = |
| 1760 | ModelInventory::from_config(&with_router(chat(provider, model), true)).unwrap(); |
| 1761 | assert!(!wrong.router_configured, "{provider}/{model}"); |
| 1762 | assert_eq!( |
| 1763 | wrong.router_setup_issue, |
| 1764 | Some(AutoRouterSetupIssue::InvalidModel), |
| 1765 | "{provider}/{model}" |
| 1766 | ); |
| 1767 | } |
| 1768 | |
| 1769 | let valid = |
| 1770 | ModelInventory::from_config(&with_router(chat("deepseek", "deepseek-v4-flash"), true)) |
| 1771 | .unwrap(); |
| 1772 | assert!(valid.router_configured); |
| 1773 | assert_ne!( |
| 1774 | valid.router_setup_issue, |
| 1775 | Some(AutoRouterSetupIssue::InvalidModel) |
| 1776 | ); |
| 1777 | } |
| 1778 | |
| 1779 | #[test] |
| 1780 | fn a_router_table_with_only_tuning_keys_is_declared_and_incomplete() { |
| 1781 | let _env = hermetic(); |
| 1782 | let tuning_only = [ |
| 1783 | crate::config::AutoRouterConfig { |
| 1784 | timeout_secs: Some(3), |
| 1785 | ..Default::default() |
| 1786 | }, |
| 1787 | crate::config::AutoRouterConfig { |
| 1788 | min_confidence: Some(0.6), |
| 1789 | ..Default::default() |
| 1790 | }, |
| 1791 | crate::config::AutoRouterConfig { |
| 1792 | thinking: Some("off".to_string()), |
| 1793 | ..Default::default() |
| 1794 | }, |
| 1795 | crate::config::AutoRouterConfig { |
| 1796 | base_url: Some("https://api.typesafe.ai/v1".to_string()), |
| 1797 | ..Default::default() |
| 1798 | }, |
| 1799 | ]; |
| 1800 | for router in tuning_only { |
| 1801 | let inventory = |
| 1802 | ModelInventory::from_config(&with_router(router.clone(), true)).unwrap(); |
| 1803 | assert!(!inventory.router_available, "{router:?}"); |
| 1804 | assert_eq!( |
| 1805 | inventory.router_setup_issue, |
| 1806 | Some(AutoRouterSetupIssue::Incomplete), |
| 1807 | "{router:?}" |
| 1808 | ); |
| 1809 | } |
| 1810 | } |
| 1811 | |
| 1812 | #[test] |
| 1813 | fn min_confidence_is_clamped() { |
| 1814 | let _env = hermetic(); |
| 1815 | for (configured, expected) in [ |
| 1816 | (Some(2.0), 10_000), |
| 1817 | (Some(-1.0), 0), |
| 1818 | (Some(0.35), 3_500), |
| 1819 | (None, 5_000), |
| 1820 | ] { |
| 1821 | let router = crate::config::AutoRouterConfig { |
| 1822 | min_confidence: configured, |
| 1823 | ..decision("openrouter") |
| 1824 | }; |
| 1825 | assert_eq!( |
| 1826 | ModelInventory::from_config(&with_router(router, true)) |
| 1827 | .unwrap() |
| 1828 | .router_min_confidence_bp, |
| 1829 | expected, |
| 1830 | "{configured:?}" |
| 1831 | ); |
| 1832 | } |
| 1833 | } |
| 1834 | } |
| 1835 |