Add Meta AI (Llama) as BYOAI provider, fix glass theme message popover

- Add "meta" provider alias (OpenAI-compatible, base URL auto-configured)
- Add Meta AI to Settings dropdown and AI Profiles support
- Fix Llama API empty arguments rejection (pad with first schema property)
- Trim whitespace on profile base URLs to prevent silent failures
- Add Meta AI section to BYOAI docs (models, key format, rate limits)
- Fix glass theme transparency on Messages panel popover

Author: Kevin Hopper

.env.example

@@ -102,7 +102,7 @@ CROW_FILES_PATH=/home
 # ── AI PROVIDER (BYOAI — Bring Your Own AI) ──────────────
 
 # Configure an AI provider to enable the built-in AI Chat in the Crow's Nest.
-# Supports: openai, anthropic, google, ollama, openrouter
+# Supports: openai, anthropic, google, ollama, openrouter, meta
 # AI_PROVIDER=openai
 # AI_API_KEY=sk-...
 # AI_MODEL=gpt-4o

docs/guide/ai-providers.md

@@ -50,6 +50,7 @@ No gateway restart needed — the config is hot-reloaded.
 | Google Gemini | `google` | `gemini-2.5-flash` | Yes | [Get key](https://aistudio.google.com/app/apikey) |
 | Ollama | `ollama` | `llama3.1` | No | Fully local, no API key needed |
 | OpenRouter | `openrouter` | `openai/gpt-4o` | Yes | [Get key](https://openrouter.ai/keys) — access 100+ models |
+| Meta AI (Llama) | `meta` | `Llama-4-Maverick-17B-128E-Instruct-FP8` | Yes | [Get key](https://llama.com/) — Llama 4 & 3.3 models |
 | DashScope Coding | `openai` | `qwen3.5-plus` | Yes | [Get key](https://dashscope.console.aliyun.com/apiKey) — Qwen, GLM, Kimi, MiniMax ([guide](/guide/dashscope-coding)) |
 | Z.AI Coding | `openai` | `glm-5` | Yes | [Get key](https://z.ai) — GLM models ([guide](/guide/zai-coding)) |
 
@@ -107,6 +108,33 @@ AI_MODEL=openai/gpt-4o
 
 OpenRouter gives you access to 100+ models from multiple providers through a single API key. Great for trying different models without signing up for each provider separately. Many models have free tiers.
 
+### Meta AI (Llama)
+
+```env
+AI_PROVIDER=meta
+AI_API_KEY=LLM|...
+AI_MODEL=Llama-4-Maverick-17B-128E-Instruct-FP8
+```
+
+Meta's Llama API provides direct access to Llama models. Available models:
+
+| Model | RPM | TPM |
+|---|---|---|
+| `Llama-4-Maverick-17B-128E-Instruct-FP8` | 10 | 250,000 |
+| `Llama-4-Scout-17B-16E-Instruct-FP8` | 10 | 250,000 |
+| `Llama-3.3-70B-Instruct` | 10 | 250,000 |
+| `Llama-3.3-8B-Instruct` | 10 | 250,000 |
+
+The API is OpenAI-compatible — no custom base URL needed.
+
+::: tip API Key Format
+Meta API keys start with `LLM|` (e.g., `LLM|953656...|8vKG-...`). Get one at [llama.com](https://llama.com/).
+:::
+
+::: warning Semantic Search
+Meta's API does not support embeddings. Semantic search is not available when using Meta as your AI provider — Crow falls back to keyword search (FTS5) automatically.
+:::
+
 ### DashScope Coding Plan (Alibaba Cloud)
 
 ```env
@@ -146,7 +174,7 @@ This works with vLLM, LM Studio, text-generation-webui, and other OpenAI-compati
 
 | Variable | Required | Description |
 |---|---|---|
-| `AI_PROVIDER` | Yes | Provider name: `openai`, `anthropic`, `google`, `ollama`, `openrouter` |
+| `AI_PROVIDER` | Yes | Provider name: `openai`, `anthropic`, `google`, `ollama`, `openrouter`, `meta` |
 | `AI_API_KEY` | Depends | API key (not needed for Ollama) |
 | `AI_MODEL` | No | Model name (uses provider default if blank) |
 | `AI_BASE_URL` | No | Custom API endpoint (for Ollama, OpenRouter, or self-hosted) |
@@ -250,7 +278,7 @@ LocalAI provides an OpenAI-compatible API running entirely on your hardware —
 Set `AI_PROVIDER` in Settings or `.env`. At minimum you need the provider name.
 
 ### "API key is invalid (401)"
-Double-check your `AI_API_KEY`. For Anthropic, keys start with `sk-ant-`. For OpenAI, `sk-`. For Google, `AIza`.
+Double-check your `AI_API_KEY`. For Anthropic, keys start with `sk-ant-`. For OpenAI, `sk-`. For Google, `AIza`. For Meta, keys start with `LLM|`.
 
 ### "Model not found (404)"
 The model name is provider-specific. Check the provider's docs for available models. For Ollama, run `ollama pull <model>` first.

servers/gateway/ai/adapters/openai.js

@@ -26,8 +26,23 @@ function mcpToolsToOpenAI(tools) {
 
 /**
  * Convert chat messages to OpenAI message format.
+ * @param {Array} messages
+ * @param {Array} [tools] - Tool schemas, used to fix up empty arguments for
+ *   providers (e.g. Meta Llama) that reject "{}" as "parameters missing".
  */
-function toOpenAIMessages(messages) {
+function toOpenAIMessages(messages, tools) {
+  // Build a map of tool name → first schema property for empty-args fixup
+  const toolFirstProp = new Map();
+  if (tools) {
+    for (const t of tools) {
+      const props = t.inputSchema?.properties;
+      if (props) {
+        const firstKey = Object.keys(props)[0];
+        if (firstKey) toolFirstProp.set(t.name, firstKey);
+      }
+    }
+  }
+
   return messages.map((m) => {
     if (m.role === "tool") {
       return {
@@ -41,14 +56,21 @@ function toOpenAIMessages(messages) {
       return {
         role: "assistant",
         content: m.content || null,
-        tool_calls: toolCalls.map((tc) => ({
-          id: tc.id,
-          type: "function",
-          function: {
-            name: tc.name,
-            arguments: typeof tc.arguments === "string" ? tc.arguments : JSON.stringify(tc.arguments),
-          },
-        })),
+        tool_calls: toolCalls.map((tc) => {
+          let args = typeof tc.arguments === "string" ? tc.arguments : JSON.stringify(tc.arguments);
+          // Fix empty arguments for providers that reject "{}" (e.g. Meta Llama API)
+          if (args === "{}" || args === "") {
+            const firstProp = toolFirstProp.get(tc.name);
+            if (firstProp) {
+              args = JSON.stringify({ [firstProp]: "" });
+            }
+          }
+          return {
+            id: tc.id,
+            type: "function",
+            function: { name: tc.name, arguments: args },
+          };
+        }),
       };
     }
     return { role: m.role, content: m.content || "" };
@@ -67,15 +89,15 @@ export default function createOpenAIAdapter(config) {
       const temperature = options.temperature ?? 0.7;
       const maxTokens = options.maxTokens || 4096;
 
+      const openaiTools = mcpToolsToOpenAI(tools);
+
       const body = {
         model,
-        messages: toOpenAIMessages(messages),
+        messages: toOpenAIMessages(messages, tools),
         temperature,
         max_tokens: maxTokens,
         stream: true,
       };
-
-      const openaiTools = mcpToolsToOpenAI(tools);
       if (openaiTools) {
         body.tools = openaiTools;
       }

servers/gateway/ai/provider.js

@@ -10,6 +10,7 @@
  *   - anthropic  — Anthropic Messages API
  *   - google     — Google Gemini API
  *   - ollama     — Ollama native /api/chat endpoint
+ *   - meta       — Meta Llama API (OpenAI-compatible)
  */
 
 import { readEnvFile, resolveEnvPath } from "../env-manager.js";
@@ -91,6 +92,7 @@ export async function createProviderAdapter() {
   // Resolve provider to adapter loader (support aliases)
   let adapterKey = provider;
   if (provider === "openrouter") adapterKey = "openai";
+  if (provider === "meta") adapterKey = "openai";
 
   const loader = ADAPTER_LOADERS[adapterKey];
   if (!loader) {
@@ -122,6 +124,9 @@ export async function createProviderAdapter() {
   if (provider === "openrouter" && !baseUrl) {
     adapterConfig.baseUrl = "https://openrouter.ai/api/v1";
   }
+  if (provider === "meta" && !baseUrl) {
+    adapterConfig.baseUrl = "https://api.llama.com/compat/v1/";
+  }
 
   const adapter = createAdapter(adapterConfig);
   return { adapter, config };
@@ -151,7 +156,7 @@ export async function getAiProfiles(db, { includeKeys = false } = {}) {
  * Returns { adapter, config } — same shape as createProviderAdapter().
  */
 export async function createAdapterFromProfile(profile, model) {
-  const adapterKey = profile.provider === "openrouter" ? "openai" : profile.provider;
+  const adapterKey = ["openrouter", "meta"].includes(profile.provider) ? "openai" : profile.provider;
   const loader = ADAPTER_LOADERS[adapterKey];
   if (!loader) {
     throw Object.assign(new Error(`Unknown provider: ${profile.provider}`), { code: "invalid_provider" });
@@ -173,6 +178,9 @@ export async function createAdapterFromProfile(profile, model) {
   if (profile.provider === "openrouter" && !profile.baseUrl) {
     adapterConfig.baseUrl = "https://openrouter.ai/api/v1";
   }
+  if (profile.provider === "meta" && !profile.baseUrl) {
+    adapterConfig.baseUrl = "https://api.llama.com/compat/v1/";
+  }
 
   const adapter = mod.default(adapterConfig);
   return { adapter, config: { provider: profile.provider, model: resolvedModel, baseUrl: profile.baseUrl } };

servers/gateway/dashboard/settings/sections/ai-profiles.js

@@ -196,7 +196,7 @@ export default {
         profiles[idx].name = profile_name;
         profiles[idx].provider = profile_provider;
         if (profile_api_key) profiles[idx].apiKey = profile_api_key;
-        profiles[idx].baseUrl = profile_base_url || "";
+        profiles[idx].baseUrl = (profile_base_url || "").trim();
         profiles[idx].models = models;
         profiles[idx].defaultModel = defaultModel;
       } else {
@@ -207,7 +207,7 @@ export default {
           name: profile_name,
           provider: profile_provider,
           apiKey: profile_api_key || "",
-          baseUrl: profile_base_url || "",
+          baseUrl: (profile_base_url || "").trim(),
           models,
           defaultModel,
         });

servers/gateway/dashboard/settings/sections/ai-provider.js

@@ -37,6 +37,7 @@ export default {
       { id: "google", name: "Google Gemini", defaultModel: "gemini-2.5-flash" },
       { id: "ollama", name: "Ollama (local)", defaultModel: "llama3.1" },
       { id: "openrouter", name: "OpenRouter", defaultModel: "openai/gpt-4o" },
+      { id: "meta", name: "Meta AI (Llama)", defaultModel: "Llama-4-Maverick-17B-128E-Instruct-FP8" },
     ];
 
     const currentProvider = aiProviderConfig?.provider || "";
@@ -88,7 +89,7 @@ export default {
       var p = document.getElementById('ai-provider').value;
       var urlField = document.getElementById('ai-base-url-field');
       urlField.style.display = (p === 'ollama' || p === 'openrouter' || p === '') ? 'block' : 'none';
-      var defaults = {openai:'gpt-4o',anthropic:'claude-sonnet-4-20250514',google:'gemini-2.5-flash',ollama:'llama3.1',openrouter:'openai/gpt-4o'};
+      var defaults = {openai:'gpt-4o',anthropic:'claude-sonnet-4-20250514',google:'gemini-2.5-flash',ollama:'llama3.1',openrouter:'openai/gpt-4o',meta:'Llama-4-Maverick-17B-128E-Instruct-FP8'};
       document.getElementById('ai-model').placeholder = defaults[p] || 'Model name';
     }
     async function saveAiProvider() {

servers/gateway/dashboard/shared/layout.js

@@ -729,7 +729,8 @@ function dashboardCss() {
   .theme-glass .header-dropdown,
   .theme-glass .crow-dropdown,
   .theme-glass #modal-content,
-  .theme-glass #crow-player-bar {
+  .theme-glass #crow-player-bar,
+  .theme-glass .msg-popover {
     background: var(--crow-bg-popup);
     backdrop-filter: var(--crow-glass-blur);
     -webkit-backdrop-filter: var(--crow-glass-blur);