pi-hypercharm-provider 1.1.4 → 1.2.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-hypercharm-provider",
-  "version": "1.1.4",
+  "version": "1.2.0",
   "description": "HyperCharm provider extension for pi - Access DeepSeek, GLM, Kimi, Qwen, MiniMax, Gemma, and GPT-OSS models through the Charm Hyper API",
   "type": "module",
   "main": "index.ts",
@@ -21,6 +21,12 @@
     "gemma",
     "gpt-oss"
   ],
+  "devDependencies": {
+    "@earendil-works/pi-ai": "0.84.1",
+    "@earendil-works/pi-coding-agent": "0.84.1",
+    "@types/node": "^24.10.1",
+    "typescript": "6.0.3"
+  },
   "author": "",
   "license": "MIT",
   "pi": {
@@ -31,7 +37,8 @@
   "scripts": {
     "clean": "echo 'nothing to clean'",
     "build": "echo 'nothing to build'",
-    "check": "echo 'nothing to check'",
+    "check": "tsc --noEmit && node tests/status.smoke.ts",
+    "smoke": "node tests/status.smoke.ts",
     "update-models": "node scripts/update-models.js"
   }
 }

package/patch.json

@@ -1,155 +1 @@
-{
-  "deepseek-v4-flash": {
-    "maxTokens": 384000,
-    "thinkingLevelMap": {
-      "minimal": null,
-      "low": null,
-      "medium": null,
-      "high": "high",
-      "max": "max"
-    },
-    "compat": {
-      "thinkingFormat": "deepseek",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true,
-      "requiresReasoningContentOnAssistantMessages": true
-    }
-  },
-  "deepseek-v4-pro": {
-    "maxTokens": 384000,
-    "thinkingLevelMap": {
-      "minimal": null,
-      "low": null,
-      "medium": null,
-      "high": "high",
-      "max": "max"
-    },
-    "compat": {
-      "thinkingFormat": "deepseek",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true,
-      "requiresReasoningContentOnAssistantMessages": true
-    }
-  },
-  "gemma-4-26b-a4b-it": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "glm-5": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "glm-5.1": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "gpt-oss-120b": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "kimi-k2.5": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "kimi-k2.6": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "minimax-m2.7": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "qwen3.6-flash": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "qwen3.6-max": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "qwen3.6-plus": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "qwen3.7-max": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "qwen3-coder-480b-a35b-instruct-int4-mixed-ar": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  },
-  "qwen3-next-80b-a3b-instruct": {
-    "compat": {
-      "thinkingFormat": "openai",
-      "maxTokensField": "max_tokens",
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "supportsReasoningEffort": true
-    }
-  }
-}
+{}

package/pnpm-workspace.yaml

@@ -0,0 +1,15 @@
+allowBuilds:
+  '@google/genai': set this to true or false
+  protobufjs: set this to true or false
+onlyBuiltDependencies:
+  - '@google/genai'
+  - protobufjs
+
+minimumReleaseAgeExclude:
+  - '@earendil-works/pi-agent-core@0.84.0 || 0.84.1'
+  - '@earendil-works/pi-ai@0.84.0 || 0.84.1'
+  - '@earendil-works/pi-client@0.84.0 || 0.84.1'
+  - '@earendil-works/pi-coding-agent@0.84.0 || 0.84.1'
+  - '@earendil-works/pi-protocol@0.84.0 || 0.84.1'
+  - '@earendil-works/pi-telemetry@0.84.0 || 0.84.1'
+  - '@earendil-works/pi-tui@0.84.0 || 0.84.1'

package/scripts/update-models.js

@@ -1,31 +1,133 @@
 #!/usr/bin/env node
 
 /**
- * Update HyperCharm models from API
+ * Update HyperCharm models from Charm's official typed provider catalog
  *
- * Fetches models from https://hyper.charm.land/v1/models and updates:
- * - models.json: Model definitions with curated reasoning/vision flags + API pricing
- * - README.md: Model table with patch.json overrides applied
+ * Fetches models from https://hyper.charm.land/v1/provider and updates:
+ * - models.json: canonical API-owned metadata used by @charmland/pi-hyper-provider
+ * - README.md: model table with patch.json overrides applied
  *
- * The HyperCharm API provides: id, display_name, supports_reasoning,
- * supports_reasoning_effort, supports_attachments, context_window,
- * max_output_tokens, and cost.usd pricing.
- *
- * Note: supports_reasoning is unreliable for some models (reports true for
- * Llama 3.3 70B which doesn't support extended thinking). models.json
- * curates reasoning flags based on known model capabilities; patch.json
- * adds compat flags and corrections.
+ * The endpoint provides canonical names, $/M pricing, context/output limits,
+ * can_reason, optional reasoning levels, and attachment support. models.json is
+ * pure API data. patch.json is reserved for verified endpoint regressions and
+ * currently contains no overrides.
  *
  * Merge order for README: models.json → apply patch.json → merge custom-models.json
+ *
+ * API key: the stored `hypercharm` credential in ~/.pi/agent/auth.json wins, then
+ * the HYPERCHARM_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));
 
-const MODELS_API_URL = 'https://hyper.charm.land/v1/models';
+// 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 `hypercharm` credential in ~/.pi/agent/auth.json wins, then
+ * the HYPERCHARM_API_KEY environment variable.
+ */
+function resolveApiKey() {
+  try {
+    const auth = JSON.parse(fs.readFileSync(AUTH_JSON_PATH, 'utf8'));
+    const credential = auth?.hypercharm;
+    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.HYPERCHARM_API_KEY || undefined;
+}
+
+const MODELS_API_URL = 'https://hyper.charm.land/v1/provider';
 const MODELS_JSON_PATH = path.join(__dirname, '..', 'models.json');
 const PATCH_JSON_PATH = path.join(__dirname, '..', 'patch.json');
 const CUSTOM_MODELS_JSON_PATH = path.join(__dirname, '..', 'custom-models.json');
@@ -76,6 +178,9 @@ function applyPatch(model, patch) {
   if (!result.reasoning && result.compat?.thinkingFormat) {
     delete result.compat.thinkingFormat;
   }
+  if (!result.reasoning && result.thinkingLevelMap) {
+    delete result.thinkingLevelMap;
+  }
   if (result.compat && Object.keys(result.compat).length === 0) {
     delete result.compat;
   }
@@ -102,91 +207,68 @@ function buildModels(baseModels, customModels, patchData) {
 
 // ─── Model transformation ─────────────────────────────────────────────────────
 
-// Known non-reasoning models (API incorrectly reports supports_reasoning: true)
-const NON_REASONING_IDS = new Set([
-  'llama-3.3-70b-instruct',
-  'llama-4-maverick-17b-128e-instruct-fp8',
-]);
-
-// Known vision models
-const VISION_IDS = new Set([
-  'kimi-k2.5',
-  'kimi-k2.6',
-  'glm-5.1',
-  'gemma-4-26b-a4b-it',
-  'qwen3.6-flash',
-  'qwen3.6-max',
-  'qwen3.6-plus',
-  'qwen3.7-max',
-]);
-
-function transformModel(apiModel, existingModelsMap) {
-  const modelId = apiModel.id;
-
-  // Preserve existing curated data
-  if (existingModelsMap[modelId]) {
-    const existing = { ...existingModelsMap[modelId] };
-
-    // Update fields from API that may change
-    const cost = apiModel.cost?.usd || {};
-    const inputCost = convertPricing(cost['1m_in']);
-    const outputCost = convertPricing(cost['1m_out']);
-    const cacheReadCost = convertPricing(cost['1m_in_cache']);
-    const cacheWriteCost = convertPricing(cost['1m_out_cache']);
-
-    if (inputCost > 0) existing.cost.input = inputCost;
-    if (outputCost > 0) existing.cost.output = outputCost;
-    if (cacheReadCost > 0) existing.cost.cacheRead = cacheReadCost;
-    if (cacheWriteCost > 0) existing.cost.cacheWrite = cacheWriteCost;
-    if (apiModel.context_window) existing.contextWindow = apiModel.context_window;
-    // Don't override maxTokens from API for DeepSeek — it reports 8000 but the
-    // actual max is 384K (set in models.json / patch.json)
-    if (apiModel.max_output_tokens && !/^deepseek-v/.test(modelId)) {
-      existing.maxTokens = apiModel.max_output_tokens;
-    }
-
-    return existing;
+const PI_THINKING_LEVELS = ['minimal', 'low', 'medium', 'high', 'xhigh', 'max'];
+
+// Charm's official extension treats a reasoning-capable model with no levels as
+// a boolean on/off model: Pi's max level selects the single on state.
+const ON_OFF_THINKING_LEVEL_MAP = {
+  off: 'off',
+  minimal: null,
+  low: null,
+  medium: null,
+  high: null,
+  xhigh: null,
+  max: 'max',
+};
+
+function buildThinkingLevelMap(levels) {
+  if (levels.length === 0) return undefined;
+  const available = new Set(levels);
+  const result = {
+    // The provider enum uses "none" for the off state on newer deployments;
+    // the official extension looked only for the older "off" spelling.
+    off: available.has('off') ? 'off' : available.has('none') ? 'none' : null,
+  };
+  for (const level of PI_THINKING_LEVELS) {
+    result[level] = available.has(level) ? level : null;
   }
+  return result;
+}
 
-  // New model — build from API data + curated defaults
-  const cost = apiModel.cost?.usd || {};
-  const isReasoning = apiModel.supports_reasoning === true && !NON_REASONING_IDS.has(modelId);
-  const isVision = VISION_IDS.has(modelId);
-  const isDeepSeek = /^deepseek-v/.test(modelId);
-
-  const model = {
-    id: modelId,
-    name: apiModel.display_name || modelId,
-    reasoning: isReasoning,
-    input: isVision ? ['text', 'image'] : ['text'],
+function transformModel(apiModel) {
+  const reasoningLevels = Array.isArray(apiModel.reasoning_levels)
+    ? apiModel.reasoning_levels.filter(level => typeof level === 'string')
+    : [];
+  const supportsReasoningEffort = reasoningLevels.length > 0;
+  const thinkingLevelMap = supportsReasoningEffort
+    ? buildThinkingLevelMap(reasoningLevels)
+    : apiModel.can_reason === true
+      ? ON_OFF_THINKING_LEVEL_MAP
+      : undefined;
+
+  return {
+    id: apiModel.id,
+    name: apiModel.name,
+    reasoning: apiModel.can_reason === true,
+    ...(thinkingLevelMap ? { thinkingLevelMap } : {}),
+    input: apiModel.supports_attachments === true ? ['text', 'image'] : ['text'],
     cost: {
-      input: convertPricing(cost['1m_in']),
-      output: convertPricing(cost['1m_out']),
-      cacheRead: convertPricing(cost['1m_in_cache']),
-      cacheWrite: convertPricing(cost['1m_out_cache']),
+      input: typeof apiModel.cost_per_1m_in === 'number' ? apiModel.cost_per_1m_in : 0,
+      output: typeof apiModel.cost_per_1m_out === 'number' ? apiModel.cost_per_1m_out : 0,
+      cacheRead: typeof apiModel.cost_per_1m_in_cached === 'number' ? apiModel.cost_per_1m_in_cached : 0,
+      // Hyper exposes discounted cached-output pricing, not cache-write pricing;
+      // pi's cost.cacheWrite field remains unused. This matches Charm's official extension.
+      cacheWrite: 0,
     },
     contextWindow: apiModel.context_window || 0,
-    maxTokens: apiModel.max_output_tokens || 0,
-  };
-
-  // DeepSeek models: override maxTokens (API reports 8000, actual is 384K)
-  // and add thinkingLevelMap + deepseek compat
-  if (isDeepSeek && isReasoning) {
-    model.maxTokens = 384000;
-    model.thinkingLevelMap = {
-      minimal: null, low: null, medium: null, high: 'high', max: 'max',
-    };
-    model.compat = {
+    maxTokens: apiModel.default_max_tokens || apiModel.context_window || 0,
+    compat: {
+      supportsStore: false,
+      supportsReasoningEffort,
       thinkingFormat: 'deepseek',
       maxTokensField: 'max_tokens',
-      supportsDeveloperRole: true,
-      supportsStore: false,
-      supportsReasoningEffort: true,
-      requiresReasoningContentOnAssistantMessages: true,
-    };
-  }
-
-  return model;
+    },
+  };
 }
 
 // ─── README generation ────────────────────────────────────────────────────────
@@ -299,9 +381,9 @@ function updateDeprecatedModels(modelsJsonPath, newModels) {
 }
 
 async function main() {
-  const apiKey = process.env.HYPERCHARM_API_KEY;
+  const apiKey = resolveApiKey();
   if (!apiKey) {
-    console.error('Error: HYPERCHARM_API_KEY environment variable is required');
+    console.error('Error: No API key found: no `hypercharm` credential resolved from ' + AUTH_JSON_PATH + ' and HYPERCHARM_API_KEY is not set');
     console.error('Usage: HYPERCHARM_API_KEY=your-key node scripts/update-models.js');
     process.exit(1);
   }
@@ -317,7 +399,9 @@ async function main() {
     }
 
     const apiResponse = await response.json();
-    const apiModels = apiResponse.data || apiResponse;
+    const apiModels = Array.isArray(apiResponse)
+      ? apiResponse
+      : (apiResponse.models || apiResponse.data || []);
 
     if (!Array.isArray(apiModels)) {
       throw new Error('API response does not contain an array of models');
@@ -338,16 +422,12 @@ async function main() {
     }
 
     // Transform models from API, preserving existing curated data
-    let apiTransformed = apiModels.map(m => transformModel(m, existingModelsMap));
+    let apiTransformed = apiModels.map(m => transformModel(m));
     apiTransformed.sort((a, b) => a.name.localeCompare(b.name));
 
-    // Log new models (not in patch.json)
+    // Load patch overrides for README rendering. Canonical metadata already
+    // comes from /v1/provider, so new models do not require a patch entry.
     const patch = loadJson(PATCH_JSON_PATH);
-    for (const m of apiTransformed) {
-      if (!patch[m.id]) {
-        console.log(`  🆕 New model: ${m.id} (${m.name}) — add to patch.json for compat overrides`);
-      }
-    }
 
     // Update models.json — curated API data
     // Move delisted models to deprecated-models.json BEFORE models.json is overwritten