@pentoshi/clai 4.2.2 → 4.4.0

package/dist/llm/provider-profile-layers.js

@@ -1,4 +1,4 @@
-import { classifyBynaraModel, classifyNvidiaModel } from "./model-families.js";
+import { CHAT_COMPLETIONS_TERMINAL_PROOFS } from "./provider-profile.js";
 export const codeFact = (detail) => ({
     source: "builtin",
     confidence: "exact",
@@ -163,23 +163,7 @@ export const FAMILY_LAYERS = {
             reasoningOutput: ["usage.completion_tokens_details.reasoning_tokens"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
-            naturalEofAccepted: false,
-        },
-    },
-    groq: {
-        evidence: codeFact("groq-route"),
-        capabilities: { tools: "supported", streamOptions: "supported" },
-        reasoning: {
-            control: {
-                dialect: "groq-model-specific",
-                status: "unknown",
-                evidence: codeFact("groq-model-specific-controls"),
-            },
-        },
-        cache: { kind: "unknown", cacheAffectingFields: [] },
-        terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -195,13 +179,14 @@ export const FAMILY_LAYERS = {
             acceptedEfforts: ["low", "medium", "high"],
             outputShapes: ["reasoning-content", "structured-details"],
             replayScope: "tool-turn",
+            replayOptIn: "openrouter-reasoning-context",
         },
         cache: { kind: "unknown", cacheAffectingFields: [] },
         usage: {
             cachedInput: ["usage.prompt_tokens_details.cached_tokens"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -217,7 +202,7 @@ export const FAMILY_LAYERS = {
             outputShapes: ["reasoning-content"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -233,23 +218,7 @@ export const FAMILY_LAYERS = {
             outputShapes: ["reasoning-content"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
-            naturalEofAccepted: false,
-        },
-    },
-    kimchi: {
-        evidence: codeFact("kimchi-gateway"),
-        capabilities: { tools: "supported" },
-        reasoning: {
-            control: {
-                dialect: "none",
-                status: "unknown",
-                evidence: viaGateway("kimi-contract-via-gateway"),
-            },
-            outputShapes: ["reasoning-content"],
-        },
-        terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -265,7 +234,7 @@ export const FAMILY_LAYERS = {
             outputShapes: ["reasoning-content"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -286,7 +255,7 @@ export const FAMILY_LAYERS = {
             finalTurnPreservation: "supported",
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -316,7 +285,7 @@ export const FAMILY_LAYERS = {
             outputShapes: ["reasoning-content"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -325,15 +294,15 @@ export const FAMILY_LAYERS = {
         capabilities: { tools: "supported", images: "unknown" },
         reasoning: {
             control: {
-                dialect: "none",
-                status: "unknown",
-                evidence: codeFact("tokenrouter-contract-unknown"),
+                dialect: "openai-effort",
+                status: "supported",
+                evidence: providerDoc("tokenrouter-reasoning-effort-validated"),
             },
             outputShapes: ["reasoning-content"],
         },
         cache: { kind: "unknown", cacheAffectingFields: [] },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -363,7 +332,7 @@ export const FAMILY_LAYERS = {
             ],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -379,7 +348,7 @@ export const FAMILY_LAYERS = {
             outputShapes: ["reasoning-content"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -400,7 +369,7 @@ export const FAMILY_LAYERS = {
             cachedInput: ["usage.prompt_tokens_details.cached_tokens"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -417,7 +386,7 @@ export const FAMILY_LAYERS = {
             outputShapes: ["reasoning-content"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
@@ -433,20 +402,15 @@ export const FAMILY_LAYERS = {
             outputShapes: ["reasoning-content"],
         },
         terminal: {
-            proofs: ["done-sentinel", "finish-reason"],
+            proofs: CHAT_COMPLETIONS_TERMINAL_PROOFS,
             naturalEofAccepted: false,
         },
     },
 };
-const kimiMandatoryLayer = {
+export const kimiMandatoryLayer = {
     evidence: viaGateway("kimi-k3-k2p7-official-contract"),
     reasoning: {
         generation: "mandatory",
-        control: {
-            dialect: "none",
-            status: "unknown",
-            evidence: viaGateway("kimi-toggle-not-sent"),
-        },
         replayScope: "all-history",
         finalTurnPreservation: "required",
     },
@@ -456,7 +420,7 @@ const kimiMandatoryLayer = {
         mandatoryReasoningReserveTokens: 4096,
     },
 };
-const kimiConfigurableLayer = {
+export const kimiConfigurableLayer = {
     evidence: viaGateway("kimi-k2p6-official-contract"),
     reasoning: {
         generation: "optional",
@@ -464,7 +428,7 @@ const kimiConfigurableLayer = {
         finalTurnPreservation: "supported",
     },
 };
-const kimiNoPreservationLayer = {
+export const kimiNoPreservationLayer = {
     evidence: viaGateway("kimi-k2p5-official-contract"),
     reasoning: {
         generation: "optional",
@@ -472,7 +436,7 @@ const kimiNoPreservationLayer = {
         finalTurnPreservation: "unsupported",
     },
 };
-const deepseekV4DefaultOn = {
+export const deepseekV4DefaultOn = {
     evidence: viaGateway("deepseek-v4-default-thinking"),
     reasoning: {
         generation: "default-on",
@@ -480,482 +444,4 @@ const deepseekV4DefaultOn = {
         outputShapes: ["reasoning-content"],
     },
 };
-function nvidiaModelLayer(model) {
-    switch (classifyNvidiaModel(model)) {
-        case "kimi-thinking":
-            return {
-                evidence: codeFact("nim-kimi-thinking"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "kimi-template-thinking",
-                        status: "supported",
-                        evidence: codeFact("nim-kimi-thinking"),
-                    },
-                    disable: "supported",
-                    disableForm: "template-thinking-false",
-                    replayScope: "tool-turn",
-                },
-            };
-        case "deepseek-v4":
-            return {
-                evidence: codeFact("nim-deepseek-v4"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "chat-template-thinking",
-                        status: "supported",
-                        evidence: codeFact("nim-deepseek-v4"),
-                    },
-                    acceptedEfforts: ["none", "high"],
-                    disable: "supported",
-                    disableForm: "template-thinking-false",
-                    replayScope: "tool-turn",
-                    outputShapes: ["reasoning-content"],
-                },
-            };
-        case "thinking":
-            return {
-                evidence: codeFact("nim-chat-template-thinking"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "chat-template-thinking",
-                        status: "supported",
-                        evidence: codeFact("nim-chat-template-thinking"),
-                    },
-                    disable: "supported",
-                    disableForm: "template-thinking-false",
-                },
-            };
-        case "nemotron-3":
-            return {
-                evidence: codeFact("nim-nemotron-3"),
-                reasoning: {
-                    control: {
-                        dialect: "nemotron-reasoning-budget",
-                        status: "supported",
-                        evidence: codeFact("nim-nemotron-3"),
-                    },
-                    disable: "supported",
-                    disableForm: "template-enable-thinking-false",
-                },
-            };
-        case "glm-thinking":
-            return {
-                evidence: codeFact("nim-glm-thinking"),
-                reasoning: {
-                    control: {
-                        dialect: "glm-enable-thinking",
-                        status: "supported",
-                        evidence: codeFact("nim-glm-thinking"),
-                    },
-                    disable: "supported",
-                    disableForm: "template-enable-thinking-false",
-                },
-            };
-        case "enable-thinking":
-            return {
-                evidence: codeFact("nim-gemma-enable-thinking"),
-                reasoning: {
-                    control: {
-                        dialect: "glm-enable-thinking",
-                        status: "supported",
-                        evidence: codeFact("nim-gemma-enable-thinking"),
-                    },
-                    disable: "supported",
-                    disableForm: "template-enable-thinking-false",
-                },
-            };
-        case "effort-only":
-            if (/gpt-oss/i.test(model)) {
-                return {
-                    evidence: codeFact("nim-gpt-oss-effort"),
-                    reasoning: {
-                        generation: "default-on",
-                        control: {
-                            dialect: "openai-effort",
-                            status: "supported",
-                            evidence: codeFact("nim-gpt-oss-effort"),
-                        },
-                        acceptedEfforts: ["low", "medium", "high"],
-                        disable: "unsupported",
-                        disableForm: "effort-minimal-floor",
-                    },
-                };
-            }
-            return {
-                evidence: codeFact("nim-effort-only"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "openai-effort",
-                        status: "supported",
-                        evidence: codeFact("nim-effort-only"),
-                    },
-                    acceptedEfforts: ["none", "low", "medium", "high"],
-                    disable: "supported",
-                    disableForm: "effort-none",
-                },
-            };
-        default:
-            return {
-                evidence: codeFact("nim-no-thinking-knob"),
-                reasoning: {
-                    generation: "unknown",
-                    control: {
-                        dialect: "none",
-                        status: "unsupported",
-                        evidence: codeFact("nim-no-thinking-knob"),
-                    },
-                },
-            };
-    }
-}
-function bynaraModelLayer(model) {
-    switch (classifyBynaraModel(model)) {
-        case "kimi":
-            return {
-                evidence: codeFact("bynara-kimi"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "kimi-template-thinking",
-                        status: "supported",
-                        evidence: codeFact("bynara-kimi"),
-                    },
-                    acceptedEfforts: ["low", "high"],
-                    disable: "supported",
-                    disableForm: "template-thinking-false",
-                    replayScope: "tool-turn",
-                },
-            };
-        case "deepseek":
-        case "agnes":
-            return {
-                evidence: codeFact("bynara-effort"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "openai-effort",
-                        status: "supported",
-                        evidence: codeFact("bynara-effort"),
-                    },
-                    acceptedEfforts: ["none", "low", "medium", "high"],
-                    disable: "supported",
-                    disableForm: "effort-none",
-                },
-            };
-        case "stepfun":
-            return {
-                evidence: codeFact("bynara-stepfun"),
-                reasoning: {
-                    control: {
-                        dialect: "glm-enable-thinking",
-                        status: "supported",
-                        evidence: codeFact("bynara-stepfun"),
-                    },
-                    disable: "supported",
-                    disableForm: "template-enable-thinking-false",
-                },
-            };
-        default:
-            return undefined;
-    }
-}
-function agentrouterModelLayer(model) {
-    const m = model.toLowerCase();
-    if (/(?:^|\/)gpt-5|(?:^|\/)o[134](?:\b|-)/.test(m)) {
-        return {
-            evidence: codeFact("agentrouter-gpt-effort"),
-            reasoning: {
-                generation: "default-on",
-                control: {
-                    dialect: "openai-effort",
-                    status: "supported",
-                    evidence: codeFact("agentrouter-gpt-effort"),
-                },
-                acceptedEfforts: ["minimal", "low", "medium", "high"],
-                disable: "unsupported",
-                disableForm: "effort-minimal-floor",
-            },
-        };
-    }
-    if (/glm/.test(m)) {
-        return {
-            evidence: codeFact("agentrouter-glm-disabled"),
-            reasoning: {
-                generation: "default-on",
-                control: {
-                    dialect: "deepseek-thinking",
-                    status: "supported",
-                    evidence: codeFact("agentrouter-glm-disabled"),
-                },
-                disable: "supported",
-                disableForm: "thinking-disabled",
-            },
-        };
-    }
-    if (/claude/.test(m)) {
-        return {
-            evidence: codeFact("agentrouter-claude-effort"),
-            reasoning: {
-                generation: "optional",
-                control: {
-                    dialect: "anthropic-thinking",
-                    status: "supported",
-                    evidence: codeFact("agentrouter-claude-effort"),
-                },
-                acceptedEfforts: ["low", "medium", "high"],
-                disable: "supported",
-                disableForm: "omit-control",
-            },
-        };
-    }
-    return undefined;
-}
-const OPENROUTER_REASONING_PATTERN = /:thinking|deepseek-v4|deepseek-r1|qwen3|kimi-k[23]|claude-(?:opus|sonnet|haiku)-4|gpt-5|(?:^|\/)o[134]|grok.*reasoner/i;
-function openrouterModelLayer(model) {
-    if (/kimi-k3|kimi-k2\.7/.test(model))
-        return kimiMandatoryLayer;
-    if (/kimi-k2\.6/.test(model))
-        return kimiConfigurableLayer;
-    if (/kimi-k2\.5|kimi-k2(?:\b|-)/.test(model))
-        return kimiNoPreservationLayer;
-    if (!OPENROUTER_REASONING_PATTERN.test(model))
-        return undefined;
-    return {
-        evidence: providerDoc("openrouter-reasoning"),
-        reasoning: {
-            generation: "default-on",
-            control: {
-                dialect: "openai-nested-reasoning",
-                status: "supported",
-                evidence: providerDoc("openrouter-reasoning"),
-            },
-            acceptedEfforts: ["low", "medium", "high"],
-            replayScope: "tool-turn",
-            outputShapes: ["reasoning-content", "structured-details"],
-        },
-    };
-}
-const MODEL_RULES = {
-    anthropic: [
-        {
-            pattern: /claude-(?:opus|sonnet|haiku)-(?:3-7|4|5)|claude-3-7/i,
-            layer: {
-                evidence: providerDoc("anthropic-extended-thinking"),
-                reasoning: { generation: "optional" },
-            },
-        },
-        {
-            pattern: /claude/i,
-            layer: {
-                evidence: providerDoc("anthropic-no-thinking"),
-                reasoning: {
-                    generation: "none",
-                    control: {
-                        dialect: "none",
-                        status: "unsupported",
-                        evidence: providerDoc("anthropic-no-thinking"),
-                    },
-                },
-            },
-        },
-    ],
-    gemini: [
-        {
-            pattern: /gemini-(?:2\.5|3|3\.5)/i,
-            layer: {
-                evidence: providerDoc("gemini-thinking"),
-                reasoning: { generation: "optional" },
-            },
-        },
-    ],
-    ollama: [
-        {
-            pattern: /deepseek-r1|qwen3|qwq/i,
-            layer: {
-                evidence: codeFact("ollama-think-supported-families"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "ollama-think",
-                        status: "supported",
-                        evidence: codeFact("ollama-think-supported-families"),
-                    },
-                    disable: "supported",
-                    disableForm: "omit-control",
-                },
-            },
-        },
-    ],
-    openai: [
-        {
-            pattern: /gpt-5|(?:^|\/)o[134]/i,
-            layer: {
-                evidence: providerDoc("openai-reasoning-models"),
-                reasoning: {
-                    generation: "optional",
-                    acceptedEfforts: ["minimal", "low", "medium", "high"],
-                    disable: "unsupported",
-                    disableForm: "effort-minimal-floor",
-                },
-            },
-        },
-        {
-            pattern: /gpt-4o|gpt-4\.1|gpt-4-turbo/i,
-            layer: {
-                evidence: providerDoc("openai-non-reasoning"),
-                reasoning: {
-                    generation: "none",
-                    control: {
-                        dialect: "none",
-                        status: "unsupported",
-                        evidence: providerDoc("openai-non-reasoning"),
-                    },
-                },
-            },
-        },
-    ],
-    groq: [
-        {
-            pattern: /qwen\/qwen3-32b/i,
-            layer: {
-                evidence: codeFact("groq-qwen3-effort"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "openai-effort",
-                        status: "supported",
-                        evidence: codeFact("groq-qwen3-effort"),
-                    },
-                    acceptedEfforts: ["default", "none"],
-                    disable: "supported",
-                    disableForm: "effort-none",
-                },
-            },
-        },
-        {
-            pattern: /openai\/gpt-oss-(?:20b|120b)/i,
-            layer: {
-                evidence: codeFact("groq-gpt-oss-effort"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "openai-effort",
-                        status: "supported",
-                        evidence: codeFact("groq-gpt-oss-effort"),
-                    },
-                    acceptedEfforts: ["low", "medium", "high"],
-                    disable: "unsupported",
-                    disableForm: "effort-minimal-floor",
-                },
-            },
-        },
-    ],
-    fireworks: [
-        {
-            pattern: /deepseek-v4/i,
-            layer: {
-                evidence: providerDoc("fireworks-deepseek-v4"),
-                reasoning: {
-                    generation: "default-on",
-                    acceptedEfforts: ["none", "high", "max"],
-                    disable: "supported",
-                    disableForm: "effort-none",
-                    replayScope: "tool-turn",
-                    outputShapes: ["reasoning-content"],
-                },
-            },
-        },
-        {
-            pattern: /kimi-k3|kimi-k2\.7/i,
-            layer: kimiMandatoryLayer,
-        },
-        {
-            pattern: /kimi-k2\.6/i,
-            layer: kimiConfigurableLayer,
-        },
-        {
-            pattern: /kimi/i,
-            layer: kimiNoPreservationLayer,
-        },
-    ],
-    "qwen-cloud": [
-        {
-            pattern: /qwen3/i,
-            layer: {
-                evidence: providerDoc("alibaba-qwen3-thinking"),
-                reasoning: { generation: "default-on" },
-            },
-        },
-    ],
-    hetzner: [
-        {
-            pattern: /qwen/i,
-            layer: {
-                evidence: codeFact("hetzner-qwen-thinking"),
-                reasoning: {
-                    generation: "default-on",
-                    control: {
-                        dialect: "glm-enable-thinking",
-                        status: "supported",
-                        evidence: codeFact("hetzner-qwen-thinking"),
-                    },
-                    disable: "supported",
-                    disableForm: "template-enable-thinking-false",
-                },
-            },
-        },
-    ],
-    kimchi: [
-        { pattern: /kimi-k3|kimi-k2\.7/i, layer: kimiMandatoryLayer },
-        { pattern: /kimi-k2\.6/i, layer: kimiConfigurableLayer },
-        { pattern: /kimi/i, layer: kimiNoPreservationLayer },
-    ],
-    tokenrouter: [
-        { pattern: /kimi-k3|kimi-k2\.7/i, layer: kimiMandatoryLayer },
-        { pattern: /kimi-k2\.6/i, layer: kimiConfigurableLayer },
-        { pattern: /kimi/i, layer: kimiNoPreservationLayer },
-        { pattern: /deepseek/i, layer: deepseekV4DefaultOn },
-    ],
-    orcarouter: [
-        { pattern: /kimi-k3|kimi-k2\.7/i, layer: kimiMandatoryLayer },
-        { pattern: /kimi-k2\.6/i, layer: kimiConfigurableLayer },
-        { pattern: /kimi/i, layer: kimiNoPreservationLayer },
-        { pattern: /deepseek/i, layer: deepseekV4DefaultOn },
-    ],
-    modal: [
-        { pattern: /kimi-k3|kimi-k2\.7/i, layer: kimiMandatoryLayer },
-        { pattern: /kimi-k2\.6/i, layer: kimiConfigurableLayer },
-        { pattern: /kimi/i, layer: kimiNoPreservationLayer },
-        { pattern: /deepseek/i, layer: deepseekV4DefaultOn },
-    ],
-    lightning: [
-        { pattern: /kimi-k3|kimi-k2\.7/i, layer: kimiMandatoryLayer },
-        { pattern: /kimi-k2\.6/i, layer: kimiConfigurableLayer },
-        { pattern: /kimi/i, layer: kimiNoPreservationLayer },
-        { pattern: /deepseek/i, layer: deepseekV4DefaultOn },
-    ],
-    free: [{ pattern: /deepseek/i, layer: deepseekV4DefaultOn }],
-    openrouter: [],
-};
-const AWS_MANTLE_ANTHROPIC_MODEL = /(?:^|[./-])(?:anthropic|claude)(?:[./-]|$)/i;
-export function modelLayerFor(provider, model) {
-    if (provider === "nvidia")
-        return nvidiaModelLayer(model);
-    if (provider === "bynara")
-        return bynaraModelLayer(model);
-    if (provider === "agentrouter")
-        return agentrouterModelLayer(model);
-    if (provider === "openrouter")
-        return openrouterModelLayer(model);
-    if (provider === "aws-mantle") {
-        return AWS_MANTLE_ANTHROPIC_MODEL.test(model)
-            ? MODEL_RULES.anthropic?.find((rule) => rule.pattern.test(model))?.layer
-            : nvidiaModelLayer(model);
-    }
-    return MODEL_RULES[provider]?.find((rule) => rule.pattern.test(model))?.layer;
-}
 //# sourceMappingURL=provider-profile-layers.js.map