pi-wafer-provider 1.1.2 → 1.1.4

package/AGENTS.md

@@ -7,6 +7,7 @@ The following files are **idempotent** and regenerated by `scripts/update-models
 | File | Why it's auto-generated |
 |------|------------------------|
 | `models.json` | Built from the provider API. `update-models.js` fetches models, preserves curated data for known IDs, and writes this file. |
+| `deprecated-models.json` | Graveyard for models the API delisted. update-models.js stamps them with deprecatedAt and pi keeps serving them for a 2-week grace period, then evicts them. Never edit by hand. |
 | `README.md` (model table) | The table under `## Available Models` is replaced in-place by `update-models.js` after merging base models → patch → custom models. |
 
 ## Correct Files to Edit

package/README.md

@@ -76,12 +76,14 @@ pi
 
 | Model | Type | Context | Max Output | Input Cost | Output Cost | Cached Input |
 |-------|------|---------|------------|------------|-------------|--------------|
-| GLM 5.1 | Text | 203K | 33K | $1.50 | $4.50 | $0.15 |
-| GLM 5.2 | Text | 1M | 16K | Free | Free | Free |
-| Glm5.2 Fast | Text | 1M | 16K | Free | Free | Free |
-| Kimi K2.6 | Text | 262K | 33K | $1.10 | $4.80 | $0.11 |
-| MiniMax M3 | Text | 1M | 16K | Free | Free | Free |
-| Qwen 3.5 397B (A17B) | Text + Image | 262K | 33K | $0.60 | $3.60 | $0.06 |
+| GLM-5.1 | Text | 203K | 33K | $1.00 | $3.20 | $0.10 |
+| GLM-5.2 | Text | 1M | 16K | $1.26 | $3.96 | $0.23 |
+| GLM5.2-Fast | Text | 1M | 16K | $2.10 | $6.60 | $0.21 |
+| Kimi-K2.6 | Text + Image | 262K | 33K | $1.14 | $4.80 | $0.19 |
+| Kimi-K3 | Text + Image | 912K | 16K | $3.00 | $15.00 | $0.30 |
+| Kimi-K3-Fast | Text + Image | 1M | 16K | $4.50 | $22.50 | $0.45 |
+| MiniMax-M3 | Text + Image | 1M | 16K | $0.33 | $1.32 | $0.07 |
+| Qwen 3.5 397B (A17B) | Text | 262K | 33K | Free | Free | Free |
 
 *Costs are per million tokens. Prices based on official provider pricing.*
 

package/deprecated-models.json

@@ -0,0 +1 @@
+{}

package/index.ts

@@ -35,6 +35,7 @@ import { getAgentDir, type ExtensionAPI, type ModelRegistry } from "@earendil-wo
 import modelsData from "./models.json" with { type: "json" };
 import customModelsData from "./custom-models.json" with { type: "json" };
 import patchData from "./patch.json" with { type: "json" };
+import deprecatedData from "./deprecated-models.json" with { type: "json" };
 import fs from "fs";
 import path from "path";
 
@@ -123,7 +124,10 @@ function applyPatch(model: JsonModel, patch: PatchEntry): JsonModel {
 function buildModels(base: JsonModel[], custom: JsonModel[], patch: PatchData): JsonModel[] {
   const modelMap = new Map<string, JsonModel>();
 
-  for (const model of base) {
+  // Seed with the base list plus grace-period deprecated models so patch.json
+  // entries apply to deprecated models exactly as while the model was live
+  // (withDeprecated keeps live data on id conflicts).
+  for (const model of withDeprecated(base)) {
     modelMap.set(model.id, model);
   }
 
@@ -185,17 +189,29 @@ interface ProviderConfig {
 
 /** Transform a model from the Wafer /v1/models API. */
 function transformApiModel(apiModel: any): JsonModel | null {
+  const details = apiModel.wafer || {};
+  const capabilities = details.capabilities || {};
+  const pricing = details.pricing || {};
+  const reasoning = capabilities.reasoning === true;
   return {
     id: apiModel.id,
-    name: apiModel.id,
-    reasoning: false,
-    input: ["text"],
-    cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-    contextWindow: apiModel.max_model_len || 0,
-    maxTokens: 0,
+    name: details.display_name || apiModel.id,
+    reasoning,
+    input: capabilities.vision === true ? ["text", "image"] : ["text"],
+    cost: {
+      input: (pricing.input_cents_per_million || 0) / 100,
+      output: (pricing.output_cents_per_million || 0) / 100,
+      cacheRead: (pricing.cache_read_cents_per_million || 0) / 100,
+      cacheWrite: 0,
+    },
+    contextWindow: details.context_length || apiModel.max_model_len || 0,
+    maxTokens: details.context_length || apiModel.max_model_len || 0,
     compat: {
+      supportsStore: false,
+      supportsDeveloperRole: false,
+      maxTokensField: "max_completion_tokens",
       supportsZdr: apiModel.zdr_supported ?? undefined,
-      supportsReasoningEffort: true,
+      ...(reasoning ? { supportsReasoningEffort: true } : {}),
     },
   };
 }
@@ -275,6 +291,35 @@ function mergeWithEmbedded(liveModels: JsonModel[], embeddedModels: JsonModel[])
   return result;
 }
 
+// Grace period for delisted models. When the provider API stops listing a
+// model, update-models.js moves its last-known definition into
+// deprecated-models.json (stamped with deprecatedAt) instead of dropping it.
+// For 14 days the model keeps working here so in-flight sessions and saved
+// model settings do not break; afterwards it is evicted permanently.
+const DEPRECATED_MODEL_TTL_MS = 14 * 24 * 60 * 60 * 1000;
+
+// Grace-period deprecated models with deprecation metadata stripped.
+function activeDeprecatedModels(): JsonModel[] {
+  const now = Date.now();
+  const result: JsonModel[] = [];
+  for (const entry of Object.values(deprecatedData as Record<string, JsonModel & { deprecatedAt?: string }>)) {
+    if (!entry?.id) continue;
+    const removedAt = Date.parse(entry.deprecatedAt ?? "");
+    if (Number.isNaN(removedAt) || now - removedAt > DEPRECATED_MODEL_TTL_MS) continue;
+    const model = { ...entry } as JsonModel & { deprecatedAt?: string };
+    delete model.deprecatedAt;
+    result.push(model);
+  }
+  return result;
+}
+
+// Append grace-period deprecated models the list does not already have (live data wins).
+function withDeprecated(models: JsonModel[]): JsonModel[] {
+  const seen = new Set(models.map((m) => m.id));
+  const extras = activeDeprecatedModels().filter((m) => !seen.has(m.id));
+  return extras.length > 0 ? [...models, ...extras] : models;
+}
+
 function loadStaleModels(providerId: string, embeddedModels: JsonModel[]): JsonModel[] {
   const cached = loadCachedModels(providerId);
   if (!cached || cached.length === 0) return embeddedModels;

package/models.json

@@ -1,15 +1,15 @@
 [
   {
     "id": "GLM-5.1",
-    "name": "GLM 5.1",
-    "reasoning": false,
+    "name": "GLM-5.1",
+    "reasoning": true,
     "input": [
       "text"
     ],
     "cost": {
-      "input": 1.5,
-      "output": 4.5,
-      "cacheRead": 0.15,
+      "input": 1,
+      "output": 3.2,
+      "cacheRead": 0.1,
       "cacheWrite": 0
     },
     "contextWindow": 202752,
@@ -17,20 +17,22 @@
     "compat": {
       "maxTokensField": "max_completion_tokens",
       "supportsDeveloperRole": false,
-      "supportsZdr": true
+      "supportsZdr": true,
+      "supportsStore": false,
+      "supportsReasoningEffort": true
     }
   },
   {
     "id": "GLM-5.2",
-    "name": "GLM 5.2",
-    "reasoning": false,
+    "name": "GLM-5.2",
+    "reasoning": true,
     "input": [
       "text"
     ],
     "cost": {
-      "input": 0,
-      "output": 0,
-      "cacheRead": 0,
+      "input": 1.26,
+      "output": 3.96,
+      "cacheRead": 0.23,
       "cacheWrite": 0
     },
     "contextWindow": 1048576,
@@ -39,20 +41,21 @@
       "maxTokensField": "max_completion_tokens",
       "supportsDeveloperRole": false,
       "supportsStore": false,
-      "supportsZdr": true
+      "supportsZdr": true,
+      "supportsReasoningEffort": true
     }
   },
   {
     "id": "glm5.2-fast",
-    "name": "Glm5.2 Fast",
-    "reasoning": false,
+    "name": "GLM5.2-Fast",
+    "reasoning": true,
     "input": [
       "text"
     ],
     "cost": {
-      "input": 0,
-      "output": 0,
-      "cacheRead": 0,
+      "input": 2.1,
+      "output": 6.6,
+      "cacheRead": 0.21,
       "cacheWrite": 0
     },
     "contextWindow": 1048576,
@@ -61,20 +64,22 @@
       "maxTokensField": "max_completion_tokens",
       "supportsDeveloperRole": false,
       "supportsStore": false,
-      "supportsZdr": true
+      "supportsZdr": true,
+      "supportsReasoningEffort": true
     }
   },
   {
     "id": "Kimi-K2.6",
-    "name": "Kimi K2.6",
-    "reasoning": false,
+    "name": "Kimi-K2.6",
+    "reasoning": true,
     "input": [
-      "text"
+      "text",
+      "image"
     ],
     "cost": {
-      "input": 1.1,
+      "input": 1.14,
       "output": 4.8,
-      "cacheRead": 0.11,
+      "cacheRead": 0.19,
       "cacheWrite": 0
     },
     "contextWindow": 262144,
@@ -82,20 +87,71 @@
     "compat": {
       "maxTokensField": "max_completion_tokens",
       "supportsDeveloperRole": false,
-      "supportsZdr": false
+      "supportsZdr": false,
+      "supportsStore": false,
+      "supportsReasoningEffort": true
+    }
+  },
+  {
+    "id": "Kimi-K3",
+    "name": "Kimi-K3",
+    "reasoning": true,
+    "input": [
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 3,
+      "output": 15,
+      "cacheRead": 0.3,
+      "cacheWrite": 0
+    },
+    "contextWindow": 912384,
+    "maxTokens": 16384,
+    "compat": {
+      "maxTokensField": "max_completion_tokens",
+      "supportsDeveloperRole": false,
+      "supportsStore": false,
+      "supportsZdr": true,
+      "supportsReasoningEffort": true
+    }
+  },
+  {
+    "id": "kimi-k3-fast",
+    "name": "Kimi-K3-Fast",
+    "reasoning": true,
+    "input": [
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 4.5,
+      "output": 22.5,
+      "cacheRead": 0.45,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 16384,
+    "compat": {
+      "maxTokensField": "max_completion_tokens",
+      "supportsDeveloperRole": false,
+      "supportsStore": false,
+      "supportsZdr": true,
+      "supportsReasoningEffort": true
     }
   },
   {
     "id": "MiniMax-M3",
-    "name": "MiniMax M3",
-    "reasoning": false,
+    "name": "MiniMax-M3",
+    "reasoning": true,
     "input": [
-      "text"
+      "text",
+      "image"
     ],
     "cost": {
-      "input": 0,
-      "output": 0,
-      "cacheRead": 0,
+      "input": 0.33,
+      "output": 1.32,
+      "cacheRead": 0.07,
       "cacheWrite": 0
     },
     "contextWindow": 1048576,
@@ -104,7 +160,8 @@
       "maxTokensField": "max_completion_tokens",
       "supportsDeveloperRole": false,
       "supportsStore": false,
-      "supportsZdr": false
+      "supportsZdr": false,
+      "supportsReasoningEffort": true
     }
   },
   {
@@ -112,13 +169,12 @@
     "name": "Qwen 3.5 397B (A17B)",
     "reasoning": false,
     "input": [
-      "text",
-      "image"
+      "text"
     ],
     "cost": {
-      "input": 0.6,
-      "output": 3.6,
-      "cacheRead": 0.06,
+      "input": 0,
+      "output": 0,
+      "cacheRead": 0,
       "cacheWrite": 0
     },
     "contextWindow": 262144,
@@ -126,7 +182,8 @@
     "compat": {
       "maxTokensField": "max_completion_tokens",
       "supportsDeveloperRole": false,
-      "supportsZdr": false
+      "supportsZdr": false,
+      "supportsStore": false
     }
   }
 ]

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-wafer-provider",
-  "version": "1.1.2",
+  "version": "1.1.4",
   "description": "Wafer Serverless provider extension for pi - Access Qwen3.5-397B-A17B, GLM-5.1, Kimi K2.6, and DeepSeek-V4 through the Wafer Serverless API",
   "type": "module",
   "main": "index.ts",

package/patch.json

@@ -1,6 +1,8 @@
 {
   "Qwen3.5-397B-A17B": {
-    "providers": ["wafer-serverless"],
+    "providers": [
+      "wafer-serverless"
+    ],
     "reasoning": true,
     "compat": {
       "thinkingFormat": "qwen",
@@ -10,7 +12,9 @@
     }
   },
   "GLM-5.1": {
-    "providers": ["wafer-serverless"],
+    "providers": [
+      "wafer-serverless"
+    ],
     "reasoning": true,
     "compat": {
       "thinkingFormat": "zai",
@@ -20,67 +24,14 @@
     }
   },
   "Kimi-K2.6": {
-    "providers": ["wafer-serverless"],
+    "providers": [
+      "wafer-serverless"
+    ],
     "reasoning": true,
     "compat": {
       "maxTokensField": "max_completion_tokens",
       "supportsDeveloperRole": false,
       "supportsReasoningEffort": true
     }
-  },
-  "Qwen3.6-35B-A3B": {
-    "providers": ["wafer-serverless"],
-    "reasoning": true,
-    "compat": {
-      "thinkingFormat": "qwen",
-      "maxTokensField": "max_completion_tokens",
-      "supportsDeveloperRole": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "deepseek-v4-flash": {
-    "providers": ["wafer-serverless"],
-    "reasoning": true,
-    "compat": {
-      "thinkingFormat": "deepseek",
-      "maxTokensField": "max_completion_tokens",
-      "supportsDeveloperRole": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "deepseek-v4-pro": {
-    "providers": ["wafer-serverless"],
-    "reasoning": true,
-    "contextWindow": 1000000,
-    "maxTokens": 384000,
-    "thinkingLevelMap": {
-      "minimal": null,
-      "low": null,
-      "medium": null,
-      "high": "high",
-      "max": "max"
-    },
-    "compat": {
-      "thinkingFormat": "deepseek",
-      "maxTokensField": "max_completion_tokens",
-      "supportsDeveloperRole": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "qwen3.7-max": {
-    "providers": ["wafer-serverless"],
-    "reasoning": true,
-    "cost": {
-      "input": 5.0,
-      "output": 15.0,
-      "cacheRead": 0.5,
-      "cacheWrite": 0
-    },
-    "compat": {
-      "thinkingFormat": "qwen",
-      "maxTokensField": "max_completion_tokens",
-      "supportsDeveloperRole": false,
-      "supportsReasoningEffort": true
-    }
   }
 }

package/scripts/update-models.js

@@ -6,11 +6,9 @@
  * - models.json: Provider model definitions (enriched with pricing & compat)
  * - README.md: Model table in the Available Models section
  *
- * The Wafer /v1/models API returns basic model info (id, max_model_len)
- * but does NOT include pricing or max output tokens.
- * models.json is the source of truth for curated specs — the script preserves
- * existing data and only adds new models with sensible defaults.
- * Curate models.json manually after new model discovery.
+ * The Wafer /v1/models API returns model limits plus nested capability, pricing,
+ * modality, and ZDR metadata. models.json mirrors those API-owned fields while
+ * patch.json remains the source of truth for thinking controls and corrections.
  *
  * patch.json is applied at runtime by the provider — not baked into models.json.
  *
@@ -72,39 +70,51 @@ async function fetchModels() {
 function transformApiModel(apiModel, existingModelsMap) {
   const id = apiModel.id;
 
-  // Preserve existing curated data (pricing, reasoning, compat, etc.)
+  const details = apiModel.wafer || {};
+  const capabilities = details.capabilities || {};
+  const pricing = details.pricing || {};
+  const reasoning = capabilities.reasoning === true;
+  const input = capabilities.vision === true ? ['text', 'image'] : ['text'];
+  const cost = {
+    input: (pricing.input_cents_per_million || 0) / 100,
+    output: (pricing.output_cents_per_million || 0) / 100,
+    cacheRead: (pricing.cache_read_cents_per_million || 0) / 100,
+    cacheWrite: 0,
+  };
+
   if (existingModelsMap[id]) {
     const existing = { ...existingModelsMap[id] };
-    // Update API-derived fields if changed
-    if (apiModel.max_model_len) {
-      existing.contextWindow = apiModel.max_model_len;
-    }
-    if (apiModel.zdr_supported !== undefined) {
-      existing.compat = existing.compat || {};
-      existing.compat.supportsZdr = apiModel.zdr_supported;
-    }
+    existing.name = details.display_name || existing.name;
+    existing.reasoning = reasoning;
+    existing.input = input;
+    existing.cost = cost;
+    existing.contextWindow = details.context_length || apiModel.max_model_len || existing.contextWindow;
+    existing.compat = {
+      ...(existing.compat || {}),
+      supportsStore: false,
+      supportsDeveloperRole: false,
+      maxTokensField: 'max_completion_tokens',
+      supportsZdr: apiModel.zdr_supported,
+      ...(reasoning ? { supportsReasoningEffort: true } : {}),
+    };
     return existing;
   }
 
-  // New model — sensible defaults; curate models.json manually after discovery
+  const contextWindow = details.context_length || apiModel.max_model_len || 131072;
   const model = {
     id,
-    name: generateDisplayName(id),
-    reasoning: false,
-    input: ['text'],
-    cost: {
-      input: 0,
-      output: 0,
-      cacheRead: 0,
-      cacheWrite: 0,
-    },
-    contextWindow: apiModel.max_model_len || 131072,
-    maxTokens: 16384,
+    name: details.display_name || generateDisplayName(id),
+    reasoning,
+    input,
+    cost,
+    contextWindow,
+    maxTokens: contextWindow,
     compat: {
       maxTokensField: 'max_completion_tokens',
       supportsDeveloperRole: false,
       supportsStore: false,
       supportsZdr: apiModel.zdr_supported,
+      ...(reasoning ? { supportsReasoningEffort: true } : {}),
     },
   };
 
@@ -226,6 +236,67 @@ function updateReadme(models) {
 
 // ─── Main ────────────────────────────────────────────────────────────────────
 
+// Grace period for delisted models: update-models.js moves models the API no
+// longer lists into deprecated-models.json (stamped with deprecatedAt) instead
+// of dropping them; the runtime appends them back so sessions and saved model
+// settings keep working, and after 14 days they are evicted permanently.
+const DEPRECATED_MODEL_TTL_MS = 14 * 24 * 60 * 60 * 1000;
+
+/**
+ * Reconcile deprecated-models.json against the freshly fetched model list.
+ * - in old models.json but not the API: moved into the deprecated file
+ *   (deprecatedAt = now; preserved on repeat runs so the grace clock is not reset)
+ * - back in the API: resurrected (dropped from the deprecated file)
+ * - deprecatedAt older than 14 days: evicted permanently
+ * Must run BEFORE the new models.json is written; it reads the old file itself.
+ */
+function updateDeprecatedModels(modelsJsonPath, newModels) {
+  const deprecatedPath = path.join(path.dirname(modelsJsonPath), 'deprecated-models.json');
+
+  let oldModels = [];
+  try {
+    const parsed = JSON.parse(fs.readFileSync(modelsJsonPath, 'utf8'));
+    if (Array.isArray(parsed)) oldModels = parsed;
+  } catch { /* first run: no previous models.json */ }
+
+  let deprecated = {};
+  try {
+    const parsed = JSON.parse(fs.readFileSync(deprecatedPath, 'utf8'));
+    if (parsed && typeof parsed === 'object' && !Array.isArray(parsed)) deprecated = parsed;
+  } catch { /* no graveyard yet */ }
+
+  const currentIds = new Set(newModels.map(m => m.id));
+  const now = new Date().toISOString();
+  const added = [];
+  const resurrected = [];
+  const evicted = [];
+
+  for (const old of oldModels) {
+    if (old && old.id && !currentIds.has(old.id) && !deprecated[old.id]) {
+      deprecated[old.id] = { ...old, deprecatedAt: now };
+      added.push(old.id);
+    }
+  }
+
+  for (const [id, entry] of Object.entries(deprecated)) {
+    if (currentIds.has(id)) {
+      delete deprecated[id];
+      resurrected.push(id);
+      continue;
+    }
+    const removedAt = Date.parse(entry && entry.deprecatedAt ? entry.deprecatedAt : '');
+    if (Number.isNaN(removedAt) || Date.now() - removedAt > DEPRECATED_MODEL_TTL_MS) {
+      delete deprecated[id];
+      evicted.push(id);
+    }
+  }
+
+  if (added.length > 0 || resurrected.length > 0 || evicted.length > 0) {
+    fs.writeFileSync(deprecatedPath, JSON.stringify(deprecated, null, 2) + '\n');
+    console.log('Updated deprecated-models.json ' + JSON.stringify({ added, resurrected, evicted }));
+  }
+}
+
 async function main() {
   try {
     const apiModels = await fetchModels();
@@ -249,6 +320,8 @@ async function main() {
     models.sort((a, b) => a.name.localeCompare(b.name));
 
     // Save models.json (pure API output, no patch/custom baked in)
+    // Move delisted models to deprecated-models.json BEFORE models.json is overwritten
+    updateDeprecatedModels(MODELS_JSON_PATH, models);
     saveJson(MODELS_JSON_PATH, models);
 
     // Build full model list for README: base → patch → custom