pi-io-provider 1.0.3

package/scripts/update-models.js

@@ -0,0 +1,393 @@
+#!/usr/bin/env node
+
+/**
+ * Script to update IO Intelligence models from the API
+ *
+ * Fetches the model list from https://api.intelligence.io.solutions/api/v1/models
+ * and regenerates models.json and the README model table.
+ *
+ * Requires IOINTELLIGENCE_API_KEY environment variable.
+ * Usage: IOINTELLIGENCE_API_KEY=your-key node scripts/update-models.js
+ */
+
+import https from 'https';
+import fs from 'fs';
+import path from 'path';
+import { fileURLToPath } from 'url';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = path.dirname(__filename);
+
+const API_BASE = 'https://api.intelligence.io.solutions/api/v1';
+const MODELS_PATH = path.join(process.cwd(), 'models.json');
+const PATCH_PATH = path.join(process.cwd(), 'patch.json');
+const CUSTOM_MODELS_PATH = path.join(process.cwd(), 'custom-models.json');
+
+// ─── HTTP helpers ───────────────────────────────────────────────────────────
+
+function fetchJSON(url, headers = {}) {
+  return new Promise((resolve, reject) => {
+    const req = https.get(url, { headers }, (res) => {
+      let data = '';
+      res.on('data', (chunk) => (data += chunk));
+      res.on('end', () => {
+        try {
+          resolve(JSON.parse(data));
+        } catch (e) {
+          reject(new Error(`Failed to parse JSON from ${url}: ${e.message}`));
+        }
+      });
+    });
+    req.on('error', reject);
+  });
+}
+
+// ─── Model transformation ──────────────────────────────────────────────────
+
+/** Clean up the display name from the API. */
+function cleanName(apiName, apiId) {
+  // API returns names like "MoonshotAI: Kimi K2.6" or "MiniMaxAI/MiniMax-M2.5"
+  let name = apiName;
+  const colonIdx = name.indexOf(': ');
+  if (colonIdx > 0 && colonIdx < 25) {
+    name = name.substring(colonIdx + 2);
+  }
+  if (name.includes('/') && !name.includes(' ')) {
+    const parts = name.split('/');
+    name = parts[parts.length - 1].replace(/-/g, ' ');
+  }
+  if (name === 'R1 0528') name = 'DeepSeek R1 0528';
+  return name;
+}
+
+function convertModel(apiModel, existingModelsMap) {
+  const id = apiModel.id;
+
+  // Preserve existing curated data (reasoning, compat, etc.)
+  if (existingModelsMap[id]) {
+    const existing = { ...existingModelsMap[id] };
+    // Update mutable fields from API
+    const ctx = apiModel.context_window || 0;
+    const maxTok = apiModel.max_tokens || ctx;
+    const priceIn = (apiModel.input_token_price || 0) * 1_000_000;
+    const priceOut = (apiModel.output_token_price || 0) * 1_000_000;
+    const cacheRead = (apiModel.cache_read_token_price || 0) * 1_000_000;
+    const cacheWrite = (apiModel.cache_write_token_price || 0) * 1_000_000;
+    if (ctx > 0) existing.contextWindow = ctx;
+    if (maxTok > 0) existing.maxTokens = maxTok;
+    // Round to 6 decimals of $/M: normalizes float noise from the ×1e6 multiply
+    // and preserves sub-cent cache prices like 0.003 (cent rounding erased them).
+    if (priceIn > 0) existing.cost.input = Math.round(priceIn * 1e6) / 1e6;
+    if (priceOut > 0) existing.cost.output = Math.round(priceOut * 1e6) / 1e6;
+    if (cacheRead > 0) existing.cost.cacheRead = Math.round(cacheRead * 1e6) / 1e6;
+    if (cacheWrite > 0) existing.cost.cacheWrite = Math.round(cacheWrite * 1e6) / 1e6;
+    if (apiModel.supports_images_input && !existing.input.includes('image')) {
+      existing.input = ['text', 'image'];
+    }
+    return existing;
+  }
+
+  // New model — build from API data + sensible defaults
+  const ctx = apiModel.context_window || 0;
+  const maxTok = apiModel.max_tokens || ctx;
+  const input = ['text'];
+  if (apiModel.supports_images_input) input.push('image');
+
+  const priceIn = (apiModel.input_token_price || 0) * 1_000_000;
+  const priceOut = (apiModel.output_token_price || 0) * 1_000_000;
+  const cacheRead = (apiModel.cache_read_token_price || 0) * 1_000_000;
+  const cacheWrite = (apiModel.cache_write_token_price || 0) * 1_000_000;
+
+  return {
+    id,
+    name: cleanName(apiModel.name, id),
+    reasoning: false,
+    input,
+    cost: {
+      input: Math.round(priceIn * 1e6) / 1e6,
+      output: Math.round(priceOut * 1e6) / 1e6,
+      cacheRead: Math.round(cacheRead * 1e6) / 1e6,
+      cacheWrite: Math.round(cacheWrite * 1e6) / 1e6,
+    },
+    contextWindow: ctx,
+    maxTokens: maxTok,
+  };
+}
+
+// ─── Patch & Custom Models ──────────────────────────────────────────────────
+
+function applyPatch(model, patch) {
+  const result = { ...model };
+  if (patch.name !== undefined) result.name = patch.name;
+  if (patch.reasoning !== undefined) result.reasoning = patch.reasoning;
+  if (patch.input !== undefined) result.input = patch.input;
+  if (patch.contextWindow !== undefined) result.contextWindow = patch.contextWindow;
+  if (patch.maxTokens !== undefined) result.maxTokens = patch.maxTokens;
+  if (patch.cost) {
+    result.cost = {
+      input: patch.cost.input ?? result.cost.input,
+      output: patch.cost.output ?? result.cost.output,
+      cacheRead: patch.cost.cacheRead ?? result.cost.cacheRead,
+      cacheWrite: patch.cost.cacheWrite ?? result.cost.cacheWrite,
+    };
+  }
+  if (patch.compat) {
+    result.compat = { ...(result.compat || {}), ...patch.compat };
+  }
+  if (!result.reasoning && result.compat?.thinkingFormat) {
+    delete result.compat.thinkingFormat;
+  }
+  if (result.compat && Object.keys(result.compat).length === 0) {
+    delete result.compat;
+  }
+  return result;
+}
+
+function buildModels(baseModels, customModels, patchData) {
+  const modelMap = new Map();
+  for (const model of baseModels) {
+    modelMap.set(model.id, model);
+  }
+  for (const [id, patchEntry] of Object.entries(patchData)) {
+    const existing = modelMap.get(id);
+    if (existing) {
+      modelMap.set(id, applyPatch(existing, patchEntry));
+    }
+  }
+  for (const model of customModels) {
+    const existing = modelMap.get(model.id);
+    const patchEntry = patchData[model.id];
+    if (existing && patchEntry) {
+      modelMap.set(model.id, applyPatch(model, patchEntry));
+    } else if (existing) {
+      modelMap.set(model.id, model);
+    } else if (patchEntry) {
+      modelMap.set(model.id, applyPatch(model, patchEntry));
+    } else {
+      modelMap.set(model.id, model);
+    }
+  }
+  return Array.from(modelMap.values());
+}
+
+// ─── README generation ──────────────────────────────────────────────────────
+
+function formatCost(cost) {
+  if (cost === 0) return 'Free';
+  if (cost < 0.01) return `<$0.01`;
+  return `$${cost.toFixed(2)}`;
+}
+
+function formatCtx(num) {
+  if (num >= 1_000_000) return `${(num / 1_000_000).toFixed(1)}M`;
+  if (num >= 1_000) return `${Math.round(num / 1_000)}K`;
+  return num.toString();
+}
+
+function generateReadme(models) {
+  const sorted = [...models].sort((a, b) => {
+    // Sort by: reasoning first, then vision, then by cost
+    if (a.reasoning !== b.reasoning) return b.reasoning - a.reasoning;
+    const aVis = a.input.includes('image') ? 1 : 0;
+    const bVis = b.input.includes('image') ? 1 : 0;
+    if (aVis !== bVis) return bVis - aVis;
+    return a.id.localeCompare(b.id);
+  });
+
+  const rows = sorted.map(m => {
+    const vision = m.input.includes('image') ? '✅' : '❌';
+    const reasoning = m.reasoning ? '✅' : '❌';
+    const cache = m.cost.cacheRead > 0 ? '✅' : '❌';
+    const displayName = m.name;
+    return `| ${displayName} | \`${m.id}\` | ${formatCtx(m.contextWindow)} | ${formatCtx(m.maxTokens)} | ${vision} | ${reasoning} | ${cache} | ${formatCost(m.cost.input)} | ${formatCost(m.cost.output)} |`;
+  }).join('\n');
+
+  const readme = `# pi-io-provider
+
+A [pi](https://github.com/badlogic/pi-mono) extension that adds [IO Intelligence](https://io.net) as a custom model provider.
+
+## Features
+
+- **OpenAI-compatible API** — Uses IO Intelligence's \`/api/v1/chat/completions\` endpoint
+- **23+ AI models** — DeepSeek, Kimi, GLM, Llama, Qwen, Mistral, and more
+- **Reasoning models** — DeepSeek R1, Kimi K2 Thinking with extended reasoning
+- **Vision models** — Kimi K2.5/K2.6, Llama 4 Maverick, Llama 3.2 Vision, Qwen2.5 VL, Mistral Large
+- **Prompt caching** — Cache read/write support on most models
+- **Confidential inference** — Verifiable TEE inference with attestation (via /private/ endpoints)
+- **Streaming** — Real-time token streaming
+
+## Available Models
+
+| Model | ID | Context | Max Output | Vision | Reasoning | Cache | Input $/M | Output $/M |
+|-------|----|---------|------------|--------|-----------|-------|-----------|------------|
+${rows}
+
+*Costs are per million tokens. Cache read/write pricing available on most models.*
+
+## Installation
+
+### Option 1: Using \`pi install\` (Recommended)
+
+Install directly from GitHub:
+
+\`\`\`bash
+pi install git:github.com/monotykamary/pi-io-provider
+\`\`\`
+
+Then set your API key and run pi:
+\`\`\`bash
+# Recommended: add to auth.json
+# See Authentication section below
+
+# Or set as environment variable
+export IOINTELLIGENCE_API_KEY=your-api-key-here
+
+pi
+\`\`\`
+
+Get your API key from [io.net](https://io.net).
+
+### Option 2: Manual Clone
+
+1. Clone this repository:
+   \`\`\`bash
+   git clone https://github.com/monotykamary/pi-io-provider.git
+   cd pi-io-provider
+   \`\`\`
+
+2. Set your IO Intelligence API key:
+   \`\`\`bash
+   # Recommended: add to auth.json
+   # See Authentication section below
+
+   # Or set as environment variable
+   export IOINTELLIGENCE_API_KEY=your-api-key-here
+   \`\`\`
+
+3. Run pi with the extension:
+   \`\`\`bash
+   pi -e /path/to/pi-io-provider
+   \`\`\`
+
+## Authentication
+
+The IO Intelligence API key can be configured in multiple ways (resolved in this order):
+
+1. **\`auth.json\`** (recommended) — Add to \`~/.pi/agent/auth.json\`:
+   \`\`\`json
+   { "io-intelligence": { "type": "api_key", "key": "your-api-key" } }
+   \`\`\`
+   The \`key\` field supports literal values, env var names, and shell commands (prefix with \`!\`). See [pi's auth file docs](https://github.com/badlogic/pi-mono) for details.
+2. **Runtime override** — Use the \`--api-key\` CLI flag
+3. **Environment variable** — Set \`IOINTELLIGENCE_API_KEY\`
+
+Get your API key from [io.net](https://io.net).
+
+## Environment Variables
+
+| Variable | Required | Description |
+|----------|----------|-------------|
+| \`IOINTELLIGENCE_API_KEY\` | No | Your IO Intelligence API key (fallback if not in auth.json) |
+
+## Configuration
+
+Add to your pi configuration for automatic loading:
+
+\`\`\`json
+{
+  "extensions": [
+    "/path/to/pi-io-provider"
+  ]
+}
+\`\`\`
+
+## Usage
+
+Once loaded, select a model with:
+
+\`\`\`
+/model io-intelligence deepseek-ai/DeepSeek-R1-0528
+\`\`\`
+
+Or use \`/models\` to browse all available IO Intelligence models.
+
+## API Documentation
+
+- IO Intelligence Docs: https://io.net/docs/guides/confidential-inference/quick-start
+- OpenAI-compatible endpoint: \`https://api.intelligence.io.solutions/api/v1\`
+- Models endpoint: \`https://api.intelligence.io.solutions/api/v1/models\`
+- Confidential inference: \`https://api.intelligence.io.solutions/api/v1/private/completions\`
+
+## License
+
+MIT
+`;
+
+  return readme;
+}
+
+// ─── Main ────────────────────────────────────────────────────────────────────
+
+async function main() {
+  const apiKey = process.env.IOINTELLIGENCE_API_KEY;
+  if (!apiKey) {
+    console.error('Error: IOINTELLIGENCE_API_KEY environment variable is required');
+    console.error('Usage: IOINTELLIGENCE_API_KEY=your-key node scripts/update-models.js');
+    process.exit(1);
+  }
+
+  console.log('Fetching models from IO Intelligence API...\n');
+
+  try {
+    const data = await fetchJSON(`${API_BASE}/models`, {
+      Authorization: `Bearer ${apiKey}`,
+    });
+
+    const apiModels = data.data || [];
+    console.log(`Total models from API: ${apiModels.length}`);
+
+    // Load existing models.json — source of truth for curated specs
+    let existingModels = [];
+    try {
+      existingModels = JSON.parse(fs.readFileSync(MODELS_PATH, 'utf8'));
+    } catch (e) {
+      // File might not exist or be invalid
+    }
+    const existingModelsMap = {};
+    for (const m of existingModels) {
+      existingModelsMap[m.id] = m;
+    }
+
+    const models = apiModels.map(m => convertModel(m, existingModelsMap));
+    console.log(`Converted ${models.length} models`);
+
+    // Save models.json (pure API output, no patch/custom baked in)
+    fs.writeFileSync(MODELS_PATH, JSON.stringify(models, null, 2) + '\n');
+    console.log(`✓ Saved ${models.length} models to models.json`);
+
+    // Build full model list for README: base → patch → custom
+    let patchData = {};
+    let customModels = [];
+    try {
+      patchData = JSON.parse(fs.readFileSync(PATCH_PATH, 'utf8'));
+    } catch {}
+    try {
+      customModels = JSON.parse(fs.readFileSync(CUSTOM_MODELS_PATH, 'utf8'));
+      if (!Array.isArray(customModels)) customModels = [];
+    } catch {}
+    const readmeModels = buildModels(models, customModels, patchData);
+    readmeModels.sort((a, b) => a.name.localeCompare(b.name));
+
+    // Update README
+    const readme = generateReadme(readmeModels);
+    fs.writeFileSync(path.join(process.cwd(), 'README.md'), readme);
+    console.log(`✓ Updated README.md`);
+
+    console.log('\nDone!');
+  } catch (error) {
+    console.error('Error:', error.message);
+    process.exit(1);
+  }
+}
+
+main();