@oh-my-pi/pi-catalog 18.2.5 → 18.2.7

package/src/compat/resolve.ts

@@ -311,6 +311,7 @@ function detectOpenAI(
 	reasoningCapable: boolean,
 ): OpenAIDetection {
 	const provider = spec.provider;
+	const backendProvider = spec.providerType ?? provider;
 	const baseUrl = spec.baseUrl;
 	const hostModel = { provider, baseUrl };
 	const isZai = modelMatchesHost(hostModel, "zai");
@@ -321,8 +322,8 @@ function detectOpenAI(
 	const isDeepseekFamily = modelMatchesHost(hostModel, "deepseekFamily") || facts.is("deepseek");
 	const isDeepseekReasoning = isDeepseekFamily && reasoningCapable;
 	const isLocalOpenAICompatBackend =
-		PROXY_OPENAI_COMPAT_PROVIDERS[provider] !== true &&
-		(LOCAL_OPENAI_COMPAT_PROVIDERS[provider] === true || hasLocalLoopbackBaseUrl(baseUrl));
+		PROXY_OPENAI_COMPAT_PROVIDERS[backendProvider] !== true &&
+		(LOCAL_OPENAI_COMPAT_PROVIDERS[backendProvider] === true || hasLocalLoopbackBaseUrl(baseUrl));
 	return {
 		facts,
 		isClinePass: provider === "cline-pass",
@@ -521,12 +522,8 @@ function detectOpenAICompat(
 		replayReasoningContent: d.isLocalOpenAICompatBackend,
 		qwenPreserveThinking:
 			(thinkingFormat === "qwen" || thinkingFormat === "qwen-chat-template") && d.isLocalOpenAICompatBackend,
-		qwenTemplateReasoningEffort:
-			(thinkingFormat === "qwen" || thinkingFormat === "qwen-chat-template") &&
-			d.isLocalOpenAICompatBackend &&
-			provider !== "ollama" &&
-			isQwen &&
-			facts.revGte("3.8"),
+		// Template effort support is a reviewed backend × model contract in KDL.
+		qwenTemplateReasoningEffort: false,
 		requiresAssistantContentForToolCalls: facts.is("kimi") || d.isDirectDeepseekReasoning,
 		cacheControlFormat:
 			(d.isClinePass && (isQwen || isAnthropicModel)) || (d.isOpenRouter && isAnthropicModel)
@@ -692,6 +689,7 @@ function resolveOpenAIResponsesPolicy(
 ): ResolvedOpenAIResponsesCompat {
 	const baseUrl = spec.baseUrl ?? "";
 	const provider = spec.provider;
+	const backendProvider = spec.providerType ?? provider;
 	const hostModel = { provider, baseUrl };
 	const isAzure = modelMatchesHost(hostModel, "azureOpenAI");
 	const isOpenRouter = modelMatchesHost(hostModel, "openrouter");
@@ -703,7 +701,8 @@ function resolveOpenAIResponsesPolicy(
 	const thinkingFormat: ResolvedOpenAISharedCompat["thinkingFormat"] = isOpenRouter ? "openrouter" : "openai";
 	const reasoningCapable = compatReasoning(spec, axes);
 	const isLocalServingBackend =
-		(PROXY_OPENAI_COMPAT_PROVIDERS[provider] !== true && LOCAL_OPENAI_COMPAT_PROVIDERS[provider] === true) ||
+		(PROXY_OPENAI_COMPAT_PROVIDERS[backendProvider] !== true &&
+			LOCAL_OPENAI_COMPAT_PROVIDERS[backendProvider] === true) ||
 		hasLocalLoopbackBaseUrl(baseUrl);
 	const isAnthropicModel = facts.is("anthropic");
 	const isDeepseekFamily = facts.is("deepseek");
@@ -1063,8 +1062,8 @@ function readRuleThinking(axes: ResolvedAxes): RuleThinking {
 
 /**
  * Compat-time reasoning capability. `axes.reasoning` also promotes targets on
- * any exact `thinking-efforts` rule (the cascade's thinking-axis gate), but
- * compat may only be repaired where the provider contract opted in with
+ * reviewed effort corrections (the cascade's thinking-axis gate), but compat
+ * may only be repaired where the matching contract opted in with
  * `thinking-upgrade-neutral`; everywhere else a spec that reports no reasoning
  * stays the authoritative capability surface.
  */
@@ -1104,7 +1103,7 @@ function resolveThinkingPolicy<TApi extends Api>(
 	// reasoning (e.g. Synthetic's `none`-only off-switch): reviewed KDL must
 	// not re-expand it into an unadvertised ladder. Absent metadata is
 	// repaired only where KDL opts in with `thinking-upgrade-neutral`
-	// alongside an exact `thinking-efforts` ladder (the cascade upgrade for
+	// alongside a reviewed `thinking-efforts` ladder (the cascade upgrade for
 	// stale source capability data); otherwise the neutral default holds.
 	if (!spec.reasoning && (explicitThinking !== undefined || rule.upgradeNeutral !== true)) return undefined;
 	if (
@@ -1210,9 +1209,13 @@ function fillExplicitThinking<TApi extends Api>(
 // Entry
 // ---------------------------------------------------------------------------
 
-function buildResolveTarget<TApi extends Api>(spec: ModelSpec<TApi>, identity: ModelIdentity): ResolveTarget {
+function buildResolveTarget<TApi extends Api>(
+	spec: ModelSpec<TApi>,
+	identity: ModelIdentity,
+	providerType = spec.providerType ?? spec.provider,
+): ResolveTarget {
 	const target: ResolveTarget = {
-		provider: spec.provider,
+		provider: providerType,
 		api: spec.api,
 		class: identity.class,
 		model: spec.id,
@@ -1227,6 +1230,13 @@ function specUsesApi<TApi extends Api>(spec: ModelSpec<Api>, api: TApi): spec is
 	return spec.api === api;
 }
 
+/** Resolve the request adapter assigned to a discovery backend before materialization. */
+export function resolveDiscoveryApi(spec: ModelSpec<Api>, providerType: string): Api {
+	const identity = resolveIdentity(spec);
+	const discoveryApi = resolveCascade(buildResolveTarget(spec, identity, providerType)).catalog.discoveryApi;
+	return typeof discoveryApi === "string" ? discoveryApi : spec.api;
+}
+
 /**
  * Resolves the full policy surface for one model spec: structured identity,
  * complete compat record, thinking metadata, and catalog-data corrections.

package/src/compat/rules/README.md

@@ -89,7 +89,7 @@ revision skip-bare "o1" "o3" "o4"
 
 ### Reviewed identity overrides
 
-`override` has properties only and no child block. Required string properties are `id` (stable, globally unique review ID), `model` (exact bare model identifier, compared case-insensitively), `rationale`, and `provenance`.
+`override` has properties only and no child block. Required string properties are `id` (stable, globally unique review ID), `rationale`, and `provenance`, plus exactly one selector: `model` (exact bare identifier) or `glob` (anchored `*` wildcard over the bare identifier). Both selectors compare case-insensitively; namespace prefixes before the final `/` are ignored.
 
 Optional properties are:
 
@@ -104,7 +104,7 @@ Optional properties are:
 | `thinking-variant` | Boolean marker for a separately exposed thinking sibling. |
 | `expires-at-ms` | Non-negative Unix time in milliseconds. The override is inactive when the observation time is at or after this value. |
 
-The pair `(provider, model)` must also be unique, including provider-agnostic pairs. When no observation time is supplied, an expiring override remains active.
+The tuple `(provider, selector-kind, selector)` must also be unique, including provider-agnostic selectors. Active exact overrides take precedence over every glob. Within each selector kind, provider-scoped overrides precede provider-agnostic ones; matching globs rank by non-wildcard byte count. Equal-ranked globs are an ambiguity error, never resolved by declaration order. When no observation time is supplied, an expiring override remains active.
 
 ### Suffix collapse
 
@@ -194,6 +194,8 @@ A `models` string without `*` is an exact, case-sensitive match against the prov
 
 `priority=N` is an optional signed integer property on the block that owns axis assignments. Its default is zero. Use it only to resolve an intentional equal-specificity overlap; do not use it to encode declaration order.
 
+`buildDiscoveredModel(spec, providerType)` resolves the catalog `discovery-api` axis before materializing compatibility. It preserves the credential-bearing provider ID and records `providerType` as the backend used for provider selectors on subsequent rebuilds. Ordinary `buildModel` preserves its input API. This lets custom-named llama.cpp deployments reuse the same rules without model-specific discovery code.
+
 ### Axis vocabulary and value shapes
 
 The directive vocabulary is closed and lives in **`src/compat/axes.ts`** — one table mapping each kebab-case directive to its resolved camelCase field, namespace (`wire` / `thinking` / `catalog`), value shape, applicable compat records, and (for enums) accepted values. The compiler rejects unknown directives and out-of-vocabulary values against that table; consult it rather than a duplicated table here.
@@ -262,7 +264,7 @@ The highest-ranked matching assignment wins for that axis. Two distinct rules th
 
 ### Capability gating
 
-Wire axes are considered for every matching target. Thinking axes are considered only when the structured resolve target sets `reasoning` — except that an exact model selector declaring `thinking-efforts` upgrades the target (a reviewed correction to stale source capability metadata). Family and revision selectors never match targets missing that rank. An unmatched target resolves to empty maps; the cascade does not infer negative capabilities from absence.
+Wire axes are considered for every matching target. Thinking axes require `reasoning`, except that an exact model selector declaring `thinking-efforts` opens the gate for reviewed corrections. A winning `thinking-upgrade-neutral #true` scoped to a recognized class, family, revision, or model selector also opens it when a matching effort ladder exists; provider-wide opt-in alone does not. Materializing a neutral spec as reasoning-capable requires that opt-in and no explicit thinking vocabulary. Family and revision selectors never match targets missing that rank. An unmatched target resolves to empty maps; the cascade does not infer negative capabilities from absence.
 
 ## Runtime behavior grammar
 

package/src/compat/rules/auth/local.kdl

@@ -0,0 +1,4 @@
+auth "local" {
+	name "Local models"
+	allows-missing-api-key #true
+}

package/src/compat/rules/auth/typesafe.kdl

@@ -1,9 +1,6 @@
-// TypeSafe (System One judgments) is not a chat-model provider: it answers
-// typed questions (choice / yes-no / score) over JSON state via
-// `POST /v1/systemone`, consumed through `@oh-my-pi/pi-ai`'s `TypeSafeJudge`.
-// No `providers/typesafe.kdl` entry exists on purpose — it has no chat models,
-// so it never appears in the model picker; the credential alone is what the
-// judgment resolver checks.
+// TypeSafe (System One judgments) answers typed questions over JSON state via
+// `POST /v1/systemone`. Its judgment models are catalogued under the `judge`
+// kind; this auth entry supplies both live `/v1/models` discovery and requests.
 auth "typesafe" {
 	name "TypeSafe"
 	env "TYPESAFE_API_KEY"

package/src/compat/rules/auth/web.kdl

@@ -0,0 +1,4 @@
+auth "web" {
+	name "Web search engines"
+	allows-missing-api-key #true
+}

package/src/compat/rules/classes/qwen.kdl

@@ -33,11 +33,12 @@ class "qwen" {
 	}
 	// Qwen 3.8+ open-weight chat templates steer thinking depth through the
 	// `reasoning_effort` template kwarg (wire-exact low/medium/xhigh, template
-	// default xhigh) and raise on `enable_thinking: false`, so local
-	// llama.cpp-style hosts route the selected effort onto the template instead
-	// of always reasoning at xhigh. vLLM ignores top-level
-	// `enable_thinking`, so it rides the chat-template-kwargs dialect.
-	on "llama.cpp" "lm-studio" {
+	// default xhigh) and raise on `enable_thinking: false`. LM Studio routes
+	// the selected effort onto the template instead of always reasoning at
+	// xhigh; llama.cpp owns its equivalent policy in the provider rule. vLLM
+	// ignores top-level `enable_thinking`, so it rides the
+	// chat-template-kwargs dialect.
+	on "lm-studio" {
 		revision ">=3.8" {
 			supports-reasoning-effort #true
 			template-reasoning-effort #true

package/src/compat/rules/providers/anthropic.kdl

@@ -4,6 +4,7 @@ provider "anthropic" {
 	default-model "claude-opus-4-8"
 	env "ANTHROPIC_API_KEY"
 	discovery label="Anthropic"
+	web-search "anthropic"
 
 	// Curated Anthropic models that are live or limited-availability on the
 	// first-party `/v1/models` endpoint but that stencil.so has not catalogued

package/src/compat/rules/providers/deepinfra.kdl

@@ -5,4 +5,32 @@ provider "deepinfra" {
 	env "DEEPINFRA_API_KEY"
 	dynamic-models-authoritative #true
 	discovery label="DeepInfra" allow-unauthenticated=#true
+	kind-apis {
+		image "openai-images"
+		tts "openai-speech"
+	}
+
+	seed base-url="https://api.deepinfra.com/v1/openai" bundle="always" {
+		model "black-forest-labs/FLUX-2-pro" name="FLUX.2 Pro" api="openai-images" {
+			reasoning #false
+			input "text" "image"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "hexgrad/Kokoro-82M" name="Kokoro-82M" api="openai-speech" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+	}
+
+	models "black-forest-labs/FLUX-2-pro" {
+		kind "image"
+	}
+	models "hexgrad/Kokoro-82M" {
+		kind "tts"
+	}
 }

package/src/compat/rules/providers/google-antigravity.kdl

@@ -2,6 +2,20 @@
 
 provider "google-antigravity" {
 	default-model "gemini-3.1-pro"
+	kind-apis {
+		image "google-gemini-cli"
+	}
+	web-search "gemini"
+
+	seed api="google-gemini-cli" base-url="https://daily-cloudcode-pa.googleapis.com" bundle="always" {
+		model "gemini-3-pro-image" name="Gemini 3 Pro Image" {
+			reasoning #false
+			input "text" "image"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+	}
 
 	class "anthropic" {
 		family "opus" {
@@ -45,6 +59,9 @@ provider "google-antigravity" {
 	models "gemini-3-pro" {
 		thinking-mode "google-level"
 	}
+	models "gemini-3-pro-image" {
+		kind "image"
+	}
 	// residue: taxonomy ranks and exact globs do not isolate these models.
 	models "gemini-3.1-flash-lite" {
 		thinking-mode "google-level"

package/src/compat/rules/providers/google.kdl

@@ -3,6 +3,10 @@
 provider "google" {
 	default-model "gemini-3.1-pro-preview"
 	env "GEMINI_API_KEY"
+	kind-apis {
+		image "google-generative-ai"
+	}
+	web-search "gemini"
 
 	class "unknown" {
 		thinking-efforts "minimal" "low" "medium" "high"
@@ -17,4 +21,9 @@ provider "google" {
 	models "gemini-2.5-computer-use-preview-10-2025" "gemini-robotics-er-1.6-preview" {
 		thinking-efforts "minimal" "low" "medium" "high"
 	}
+	// The proxy conservatively classifies tool-capable text+image output as chat;
+	// this SKU is a native image generator on the Gemini image transport.
+	models "gemini-3.1-flash-lite-image" {
+		kind "image"
+	}
 }

package/src/compat/rules/providers/llama.cpp.kdl

@@ -1,6 +1,29 @@
 // Provider-wire compat for "llama.cpp".
 
 provider "llama.cpp" {
+	supports-store #false
+	supports-developer-role #false
+	supports-reasoning-effort #false
 	// Replaces the string-only named-tool-choice provider table.
 	supports-named-tool-choice #false
+
+	class "qwen" {
+		// Responses cannot encode the template's thinking controls.
+		discovery-api "openai-completions"
+		// llama-server exposes the model's bundled chat template. Qwen templates
+		// preserve prior reasoning and accept the chat-template thinking controls.
+		thinking-upgrade-neutral #true
+		thinking-mode "effort"
+		thinking-efforts "low" "medium" "high" "max"
+		thinking-format "qwen"
+		reasoning-disable-mode "qwen-enable-thinking-false"
+		qwen-preserve-thinking #true
+
+		revision ">=3.8" {
+			supports-reasoning-effort #true
+			template-reasoning-effort #true
+			thinking-efforts "low" "medium" "xhigh"
+			thinking-requires-effort #true
+		}
+	}
 }

package/src/compat/rules/providers/local.kdl

@@ -0,0 +1,110 @@
+// On-device inference workers exposed as catalog models.
+
+provider "local" {
+	default-model "lfm2.5-230m"
+	allow-unauthenticated #true
+
+	seed api="local-inference" base-url="local://inference" bundle="always" {
+		model "kokoro" name="Kokoro-82M" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "parakeet-tdt-0.6b-v3" name="Parakeet TDT 0.6B v3" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "whisper-base" name="Whisper Base" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "whisper-small" name="Whisper Small" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "whisper-large-v3-turbo" name="Whisper Large v3 Turbo" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "lfm2.5-230m" name="LFM2.5 230M" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "lfm2.5-350m" name="LFM2.5 350M" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "falcon-h1-90m" name="Falcon H1 Tiny 90M" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "qwen3-1.7b" name="Qwen3 1.7B" {
+			reasoning #true
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "llama3.2:3b" name="Llama 3.2 3B" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "gemma-3-1b" name="Gemma 3 1B" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "qwen2.5-1.5b" name="Qwen2.5 1.5B" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "lfm2-1.2b" name="LFM2 1.2B" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+	}
+
+	models "kokoro" {
+		kind "tts"
+	}
+	models "parakeet-tdt-0.6b-v3" "whisper-*" {
+		kind "stt"
+	}
+	models "lfm*" "falcon-*" "qwen*" "llama3.2:3b" "gemma-3-1b" {
+		kind "tiny"
+	}
+}

package/src/compat/rules/providers/openai-codex.kdl

@@ -3,6 +3,24 @@
 provider "openai-codex" {
 	default-model "gpt-5.5"
 	env "OPENAI_CODEX_OAUTH_TOKEN"
+	kind-apis {
+		image "openai-codex-responses"
+	}
+	web-search "codex"
+
+	seed api="openai-codex-responses" base-url="https://chatgpt.com/backend-api" bundle="always" {
+		model "gpt-image-1" name="GPT Image 1" {
+			reasoning #false
+			input "text" "image"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+	}
+
+	models "gpt-image-1" {
+		kind "image"
+	}
 
 	// Replaces the Codex provider's handwritten flex/priority pricing fallback.
 	service-tier-cost {

package/src/compat/rules/providers/openai.kdl

@@ -3,6 +3,9 @@
 provider "openai" {
 	default-model "gpt-5.5"
 	env "OPENAI_API_KEY"
+	kind-apis {
+		image "openai-responses"
+	}
 
 	// Daybreak models are approval-gated first-party Responses models that are
 	// not yet present in stencil.so. Seed the documented aliases and current

package/src/compat/rules/providers/openrouter.kdl

@@ -4,6 +4,25 @@ provider "openrouter" {
 	default-model "openai/gpt-5.5"
 	env "OPENROUTER_API_KEY"
 	discovery label="OpenRouter" allow-unauthenticated=#true
+	kind-apis {
+		image "openrouter-images"
+	}
+	web-search "openrouter"
+
+	// TypeSafe's Jev answers only through the Decisions API (`/api/alpha/decisions`,
+	// System One wire shape); the default `/models` roster omits `text->decisions`
+	// rows, so the alias is authored here and discovery refreshes the family.
+	seed api="openrouter-decisions" base-url="https://openrouter.ai/api/alpha" bundle="always" {
+		model "~typesafe/jev-latest" name="TypeSafe: Jev Latest" {
+			reasoning #false
+			input "text"
+			cost input=0.042 output=0 cache-read=0 cache-write=0
+			limits context=32000 max-tokens=28800
+		}
+	}
+	models "~typesafe/*" "typesafe/*" {
+		kind "judge"
+	}
 
 	// Replaces the OpenRouter provider wire-model-id dispatch branch.
 	wire-model-id-mode "openrouter"
@@ -59,6 +78,11 @@ provider "openrouter" {
 		thinking-efforts "minimal" "low" "medium" "high"
 		thinking-requires-effort #true
 	}
+	// The proxy conservatively classifies this tool-capable text+image output row
+	// as chat; OpenRouter's dedicated image roster confirms the image transport.
+	models "google/gemini-3-pro-image" {
+		kind "image"
+	}
 	// residue: taxonomy ranks and exact globs do not isolate these models.
 	models "qwen/qwen3-coder" {
 		thinking-format "openrouter"

package/src/compat/rules/providers/typesafe.kdl

@@ -0,0 +1,21 @@
+// TypeSafe System One judgment models.
+
+provider "typesafe" {
+	default-model "jev-latest"
+	env "TYPESAFE_API_KEY"
+
+	seed api="typesafe" base-url="https://api.typesafe.ai" bundle="always" {
+		model "jev-latest" name="TypeSafe jev" {
+			reasoning #false
+			input "text"
+			// Jev bills input only; matches the OpenRouter route seed.
+			cost input=0.042 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+	}
+
+	models "*" {
+		kind "judge"
+	}
+}

package/src/compat/rules/providers/web.kdl

@@ -0,0 +1,153 @@
+// Pure search engines exposed through the web-search runner.
+
+provider "web" {
+	default-model "public"
+	allow-unauthenticated #true
+
+	seed api="web-search" base-url="web://search" bundle="always" {
+		model "parallel" name="Parallel" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "perplexity" name="Perplexity" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "zai" name="Z.AI" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "exa" name="Exa" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "tinyfish" name="TinyFish" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "jina" name="Jina" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "kagi" name="Kagi" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "tavily" name="Tavily" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "firecrawl" name="Firecrawl" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "brave" name="Brave" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "kimi" name="Kimi" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "synthetic" name="Synthetic" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "ollama" name="Ollama" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "searxng" name="SearXNG" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "startpage" name="Startpage" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "duckduckgo" name="DuckDuckGo" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "ecosia" name="Ecosia" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "google" name="Google" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "mojeek" name="Mojeek" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "public" name="Public Web" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+	}
+
+	models "*" {
+		kind "search"
+	}
+}