fix(models): drop GitHub Llama-3.1 (no tool-calling via Azure endpoint)

Audit (scripts/test-tools.mjs) of all 15 configured models against the
calculator tool found:
- 9 OK: gemini-2.5-flash, all 6 Groq, gpt-4o-mini, gpt-4o
- 4 quota-exhausted today (Gemini free-tier, retry tomorrow)
- 2 broken: github Meta-Llama-3.1-8B/405B-Instruct never emit tool-call,
they only output prose 'tôi sẽ dùng calculator', so agent flows fail.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

lib/providers.ts

@@ -32,12 +32,12 @@ export const PROVIDERS = {
     label: "GitHub Models",
     envKey: "GITHUB_TOKEN",
     // Verified active 2026-05-20 against https://models.inference.ai.azure.com/models.
-    // Phi/Mistral/Cohere are no longer listed for chat-completion via this endpoint.
+    // Llama-3.1-8B / 405B are listed but DO NOT support function-calling through the
+    // Azure inference endpoint (they reply in prose without emitting tool-calls), so
+    // they are unusable for an agent UI. Removed 2026-05-20.
     models: [
       "gpt-4o-mini",
       "gpt-4o",
-      "Meta-Llama-3.1-8B-Instruct",
-      "Meta-Llama-3.1-405B-Instruct",
     ],
     default: "gpt-4o-mini",
   },

scripts/test-tools.mjs

@@ -0,0 +1,149 @@
+// Test tool-calling support across all configured models.
+import { readFileSync } from "node:fs"
+import { streamText, tool } from "ai"
+import { createGoogleGenerativeAI } from "@ai-sdk/google"
+import { createGroq } from "@ai-sdk/groq"
+import { createOpenAI } from "@ai-sdk/openai"
+import { z } from "zod"
+
+// Tiny inline .env.local loader (avoids extra dep)
+try {
+  const env = readFileSync(".env.local", "utf8")
+  for (const line of env.split("\n")) {
+    const m = line.match(/^\s*([A-Z_][A-Z0-9_]*)\s*=\s*"?([^"\r\n]*)"?\s*$/)
+    if (m && !process.env[m[1]]) process.env[m[1]] = m[2]
+  }
+} catch {}
+
+const PROVIDERS = {
+  google: {
+    models: [
+      "gemini-2.5-flash-lite",
+      "gemini-2.5-flash",
+      "gemini-2.5-pro",
+      "gemini-2.0-flash",
+      "gemini-2.0-flash-lite",
+    ],
+  },
+  groq: {
+    models: [
+      "llama-3.3-70b-versatile",
+      "llama-3.1-8b-instant",
+      "openai/gpt-oss-120b",
+      "openai/gpt-oss-20b",
+      "qwen/qwen3-32b",
+      "meta-llama/llama-4-scout-17b-16e-instruct",
+    ],
+  },
+  github: {
+    models: [
+      "gpt-4o-mini",
+      "gpt-4o",
+      "Meta-Llama-3.1-8B-Instruct",
+      "Meta-Llama-3.1-405B-Instruct",
+    ],
+  },
+}
+
+const tools = {
+  calculator: tool({
+    description: "Evaluate a math expression. Use for arithmetic.",
+    parameters: z.object({ expression: z.string() }),
+    execute: async ({ expression }) => {
+      const r = Function(`"use strict";return (${expression})`)()
+      return { result: r }
+    },
+  }),
+  currentTime: tool({
+    description: "Get current ISO date-time.",
+    parameters: z.object({}),
+    execute: async () => ({ now: new Date().toISOString() }),
+  }),
+}
+
+function build(provider, modelId) {
+  if (provider === "google")
+    return createGoogleGenerativeAI({ apiKey: process.env.GOOGLE_GENERATIVE_AI_API_KEY })(modelId)
+  if (provider === "groq") return createGroq({ apiKey: process.env.GROQ_API_KEY })(modelId)
+  if (provider === "github")
+    return createOpenAI({
+      apiKey: process.env.GITHUB_TOKEN,
+      baseURL: "https://models.inference.ai.azure.com",
+    })(modelId)
+}
+
+async function testOne(provider, modelId) {
+  const t0 = Date.now()
+  try {
+    const result = streamText({
+      model: build(provider, modelId),
+      tools,
+      maxSteps: 3,
+      system:
+        "Khi user hỏi tính toán PHẢI gọi tool calculator. " +
+        "Khi user hỏi giờ PHẢI gọi tool currentTime. Không tự tính.",
+      messages: [{ role: "user", content: "Tính giúp: 137 * 29 + 8 = ?" }],
+    })
+
+    let toolCalled = false
+    let toolName = null
+    let text = ""
+    let chunks = 0
+    let firstChunkMs = null
+
+    for await (const part of result.fullStream) {
+      chunks++
+      if (firstChunkMs === null) firstChunkMs = Date.now() - t0
+      if (part.type === "tool-call") {
+        toolCalled = true
+        toolName = part.toolName
+      } else if (part.type === "text-delta") {
+        text += part.textDelta
+      } else if (part.type === "error") {
+        throw part.error
+      }
+    }
+    return {
+      ok: true,
+      toolCalled,
+      toolName,
+      streaming: chunks > 3,
+      chunks,
+      ttfbMs: firstChunkMs,
+      totalMs: Date.now() - t0,
+      textPreview: text.slice(0, 80).replace(/\s+/g, " "),
+    }
+  } catch (e) {
+    return { ok: false, error: (e?.message || String(e)).slice(0, 240), totalMs: Date.now() - t0 }
+  }
+}
+
+const results = []
+for (const [provider, info] of Object.entries(PROVIDERS)) {
+  for (const m of info.models) {
+    process.stdout.write(`Testing ${provider}/${m} ... `)
+    const r = await testOne(provider, m)
+    const status = r.ok
+      ? r.toolCalled
+        ? `OK tool=${r.toolName} stream=${r.streaming ? "y" : "n"} chunks=${r.chunks} ttfb=${r.ttfbMs}ms total=${r.totalMs}ms`
+        : `NO-TOOL stream=${r.streaming ? "y" : "n"} total=${r.totalMs}ms text="${r.textPreview}"`
+      : `ERR ${r.error}`
+    console.log(status)
+    results.push({ provider, model: m, ...r })
+    await new Promise((r) => setTimeout(r, 800))
+  }
+}
+
+console.log("\n=== SUMMARY ===")
+for (const r of results) {
+  const tool = r.ok ? (r.toolCalled ? "OK " : "NO ") : "ERR"
+  const stream = r.ok ? (r.streaming ? "y" : "n") : "-"
+  console.log(`${tool} stream=${stream}  ${r.provider.padEnd(8)} ${r.model}`)
+}
+const broken = results.filter((r) => !r.ok)
+const noTool = results.filter((r) => r.ok && !r.toolCalled)
+console.log(`\n${results.length} tested · ${broken.length} broken · ${noTool.length} no tool-call`)
+if (broken.length) {
+  console.log("\nBROKEN:")
+  for (const b of broken) console.log(`  ${b.provider}/${b.model}: ${b.error}`)
+}