pi-wafer-provider 1.1.4 → 1.1.6

package/README.md

@@ -76,13 +76,13 @@ pi
 
 | Model | Type | Context | Max Output | Input Cost | Output Cost | Cached Input |
 |-------|------|---------|------------|------------|-------------|--------------|
+| DeepSeek-V4-Flash-0731-Fast | Text | 1M | 1M | $0.28 | $0.56 | $0.07 |
 | 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 | Text + Image | 1M | 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

@@ -1 +1,27 @@
-{}
+{
+  "MiniMax-M3": {
+    "id": "MiniMax-M3",
+    "name": "MiniMax-M3",
+    "reasoning": true,
+    "input": [
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 0.33,
+      "output": 1.32,
+      "cacheRead": 0.07,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 16384,
+    "compat": {
+      "maxTokensField": "max_completion_tokens",
+      "supportsDeveloperRole": false,
+      "supportsStore": false,
+      "supportsZdr": false,
+      "supportsReasoningEffort": true
+    },
+    "deprecatedAt": "2026-08-10T02:00:21.380Z"
+  }
+}

package/index.ts

@@ -97,6 +97,7 @@ function applyPatch(model: JsonModel, patch: PatchEntry): JsonModel {
   if (patch.contextWindow !== undefined) result.contextWindow = patch.contextWindow;
   if (patch.maxTokens !== undefined) result.maxTokens = patch.maxTokens;
   if (patch.providers !== undefined) result.providers = patch.providers;
+  if (patch.thinkingLevelMap !== undefined) result.thinkingLevelMap = { ...patch.thinkingLevelMap };
 
   if (patch.cost) {
     result.cost = {

package/models.json

@@ -1,4 +1,27 @@
 [
+  {
+    "id": "DeepSeek-V4-Flash-0731-Fast",
+    "name": "DeepSeek-V4-Flash-0731-Fast",
+    "reasoning": true,
+    "input": [
+      "text"
+    ],
+    "cost": {
+      "input": 0.28,
+      "output": 0.56,
+      "cacheRead": 0.07,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 1048576,
+    "compat": {
+      "maxTokensField": "max_completion_tokens",
+      "supportsDeveloperRole": false,
+      "supportsStore": false,
+      "supportsZdr": true,
+      "supportsReasoningEffort": true
+    }
+  },
   {
     "id": "GLM-5.1",
     "name": "GLM-5.1",
@@ -106,7 +129,7 @@
       "cacheRead": 0.3,
       "cacheWrite": 0
     },
-    "contextWindow": 912384,
+    "contextWindow": 1048576,
     "maxTokens": 16384,
     "compat": {
       "maxTokensField": "max_completion_tokens",
@@ -140,30 +163,6 @@
       "supportsReasoningEffort": true
     }
   },
-  {
-    "id": "MiniMax-M3",
-    "name": "MiniMax-M3",
-    "reasoning": true,
-    "input": [
-      "text",
-      "image"
-    ],
-    "cost": {
-      "input": 0.33,
-      "output": 1.32,
-      "cacheRead": 0.07,
-      "cacheWrite": 0
-    },
-    "contextWindow": 1048576,
-    "maxTokens": 16384,
-    "compat": {
-      "maxTokensField": "max_completion_tokens",
-      "supportsDeveloperRole": false,
-      "supportsStore": false,
-      "supportsZdr": false,
-      "supportsReasoningEffort": true
-    }
-  },
   {
     "id": "Qwen3.5-397B-A17B",
     "name": "Qwen 3.5 397B (A17B)",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-wafer-provider",
-  "version": "1.1.4",
+  "version": "1.1.6",
   "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

@@ -1,4 +1,15 @@
 {
+  "kimi-k3-fast": {
+    "thinkingLevelMap": {
+      "off": null,
+      "minimal": null,
+      "low": "low",
+      "medium": null,
+      "high": "high",
+      "xhigh": null,
+      "max": "max"
+    }
+  },
   "Qwen3.5-397B-A17B": {
     "providers": [
       "wafer-serverless"

package/scripts/update-models.js

@@ -12,15 +12,119 @@
  *
  * patch.json is applied at runtime by the provider — not baked into models.json.
  *
- * Requires WAFER_SERVERLESS_API_KEY (or WAFER_API_KEY) environment variable.
+ * API key: the stored `wafer-serverless` credential in ~/.pi/agent/auth.json wins, then
+ * the WAFER_SERVERLESS_API_KEY (or WAFER_API_KEY) environment variable. The script refuses to run without one.
  */
 
 import fs from 'fs';
+import os from 'os';
+import { execSync } from 'child_process';
 import path from 'path';
 import { fileURLToPath } from 'url';
 
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 
+// pi's agent directory: PI_CODING_AGENT_DIR (with ~ expansion) or ~/.pi/agent.
+function piAgentDir() {
+  const envDir = process.env.PI_CODING_AGENT_DIR;
+  if (envDir) {
+    return envDir.startsWith('~/') || envDir === '~'
+      ? path.join(os.homedir(), envDir.slice(1))
+      : envDir;
+  }
+  return path.join(os.homedir(), '.pi', 'agent');
+}
+
+const AUTH_JSON_PATH = path.join(piAgentDir(), 'auth.json');
+
+/**
+ * Resolve a configured value using pi's semantics (resolve-config-value.ts in
+ * pi-mono): "!command" runs via the shell (10s timeout) and uses trimmed
+ * stdout; "$VAR" / "${VAR}" interpolate environment variables ("$$" escapes a
+ * literal "$", "$!" a literal "!"); anything else is a literal. Returns
+ * undefined when a referenced env var is unset or a command fails.
+ */
+function resolveConfigValue(config, env) {
+  if (typeof config !== 'string' || config.length === 0) return undefined;
+  if (config.startsWith('!')) {
+    try {
+      const out = execSync(config.slice(1), {
+        encoding: 'utf8',
+        timeout: 10000,
+        stdio: ['ignore', 'pipe', 'ignore'],
+      });
+      return out.trim() || undefined;
+    } catch {
+      return undefined;
+    }
+  }
+  const ENV_NAME_RE = /^[A-Za-z_][A-Za-z0-9_]*$/;
+  let resolved = '';
+  let index = 0;
+  while (index < config.length) {
+    const dollar = config.indexOf('$', index);
+    if (dollar < 0) {
+      resolved += config.slice(index);
+      break;
+    }
+    resolved += config.slice(index, dollar);
+    const next = config[dollar + 1];
+    let name;
+    if (next === '$' || next === '!') {
+      resolved += next;
+      index = dollar + 2;
+      continue;
+    } else if (next === '{') {
+      const end = config.indexOf('}', dollar + 2);
+      if (end < 0) {
+        resolved += '$';
+        index = dollar + 1;
+        continue;
+      }
+      const inner = config.slice(dollar + 2, end);
+      if (!ENV_NAME_RE.test(inner)) {
+        resolved += config.slice(dollar, end + 1);
+        index = end + 1;
+        continue;
+      }
+      name = inner;
+      index = end + 1;
+    } else {
+      const match = config.slice(dollar + 1).match(/^[A-Za-z_][A-Za-z0-9_]*/);
+      if (!match) {
+        resolved += '$';
+        index = dollar + 1;
+        continue;
+      }
+      name = match[0];
+      index = dollar + 1 + name.length;
+    }
+    const value = (env && env[name]) || process.env[name] || undefined;
+    if (value === undefined) return undefined;
+    resolved += value;
+  }
+  return resolved;
+}
+
+/**
+ * The API key, resolved the way pi itself resolves it for this provider: the
+ * stored `wafer-serverless` credential in ~/.pi/agent/auth.json wins, then
+ * the WAFER_SERVERLESS_API_KEY (or WAFER_API_KEY) environment variable.
+ */
+function resolveApiKey() {
+  try {
+    const auth = JSON.parse(fs.readFileSync(AUTH_JSON_PATH, 'utf8'));
+    const credential = auth?.['wafer-serverless'];
+    if (credential && credential.type === 'api_key' && typeof credential.key === 'string') {
+      const key = resolveConfigValue(credential.key, credential.env);
+      if (key) return key;
+    }
+  } catch {
+    // Missing or unparseable auth.json: fall through to the env var.
+  }
+  return process.env.WAFER_SERVERLESS_API_KEY || process.env.WAFER_API_KEY || undefined;
+}
+
 const MODELS_API_URL = 'https://pass.wafer.ai/v1/models';
 const MODELS_JSON_PATH = path.join(__dirname, '..', 'models.json');
 const PATCH_JSON_PATH = path.join(__dirname, '..', 'patch.json');
@@ -45,9 +149,9 @@ function saveJson(filePath, data) {
 // ─── API fetch ───────────────────────────────────────────────────────────────
 
 async function fetchModels() {
-  const apiKey = process.env.WAFER_SERVERLESS_API_KEY || process.env.WAFER_API_KEY;
+  const apiKey = resolveApiKey();
   if (!apiKey) {
-    throw new Error('WAFER_SERVERLESS_API_KEY (or WAFER_API_KEY) environment variable is required');
+    throw new Error('No API key found: no `wafer-serverless` credential resolved from ' + AUTH_JSON_PATH + ' and WAFER_SERVERLESS_API_KEY (or WAFER_API_KEY) is not set');
   }
 
   console.log(`Fetching models from ${MODELS_API_URL}...`);