wip

Author: khromov

AGENTS.md

@@ -53,6 +53,14 @@ MODEL=openrouter/google/gemini-pro
 MODEL=openrouter/meta-llama/llama-3.1-405b-instruct
 ```
 
+**LM Studio (Local models via OpenAI-compatible API):**
+
+```bash
+MODEL=lmstudio/model-name
+```
+
+LM Studio runs a local OpenAI-compatible API server on `http://localhost:1234/v1`. Make sure LM Studio is running with a model loaded before using this provider.
+
 ### MCP Server Configuration
 
 The `MCP_SERVER_URL` environment variable controls MCP (Model Context Protocol) integration. The tool automatically detects whether to use HTTP or StdIO transport based on the value format.
@@ -101,6 +109,7 @@ MCP_SERVER_URL=
 - `ANTHROPIC_API_KEY`: Required when using `anthropic/*` models
 - `OPENAI_API_KEY`: Required when using `openai/*` models (get at https://platform.openai.com/api-keys)
 - `OPENROUTER_API_KEY`: Required when using `openrouter/*` models (get at https://openrouter.ai/keys)
+- No API key required for `lmstudio/*` models (runs locally)
 
 ### Provider Routing
 
@@ -109,6 +118,7 @@ The benchmark tool automatically routes to the correct provider based on the `MO
 - `anthropic/*` → Direct Anthropic API
 - `openai/*` → Direct OpenAI API
 - `openrouter/*` → OpenRouter unified API
+- `lmstudio/*` → LM Studio local server (OpenAI-compatible)
 
 This allows switching models and providers without any code changes.
 
@@ -154,6 +164,7 @@ tests/
 - **Vercel AI SDK v5**: Agent framework with tool calling
 - **@ai-sdk/anthropic**: Anthropic provider for direct API access
 - **@ai-sdk/openai**: OpenAI provider for direct API access
+- **@ai-sdk/openai-compatible**: OpenAI-compatible provider for LM Studio and other local servers
 - **@openrouter/ai-sdk-provider**: OpenRouter provider for unified access to 300+ models
 - **@ai-sdk/mcp**: MCP client integration (with custom patch)
 - **Bun Runtime**: JavaScript runtime (not Node.js)
@@ -255,3 +266,4 @@ This naming convention allows you to:
 - MCP integration can be toggled via `MCP_SERVER_URL` environment variable without code changes
 - MCP status is clearly indicated in both the JSON metadata and HTML report with a visual badge
 - Exit code is 0 if all tests pass, 1 if any tests fail
+- LM Studio provider requires LM Studio to be running locally with a model loaded

README.md

@@ -33,13 +33,44 @@ MCP_SERVER_URL=https://mcp.svelte.dev/mcp
 
 **Required:**
 
-- `MODEL`: The AI model to use (e.g., `anthropic/claude-sonnet-4`, `openai/gpt-5`, `openrouter/anthropic/claude-sonnet-4`)
+- `MODEL`: The AI model to use (e.g., `anthropic/claude-sonnet-4`, `openai/gpt-5`, `openrouter/anthropic/claude-sonnet-4`, `lmstudio/model-name`)
 - Corresponding API key (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, or `OPENROUTER_API_KEY`)
+  - Note: No API key required for `lmstudio/*` models (runs locally)
 
 **Optional:**
 
 - `MCP_SERVER_URL`: MCP server URL (leave empty to disable MCP integration)
 
+### Supported Providers
+
+**Cloud Providers:**
+- `anthropic/*` - Direct Anthropic API (requires `ANTHROPIC_API_KEY`)
+- `openai/*` - Direct OpenAI API (requires `OPENAI_API_KEY`)
+- `openrouter/*` - OpenRouter unified API (requires `OPENROUTER_API_KEY`)
+
+**Local Providers:**
+- `lmstudio/*` - LM Studio local server (requires LM Studio running on `http://localhost:1234`)
+
+Example configurations:
+
+```bash
+# Anthropic
+MODEL=anthropic/claude-sonnet-4
+ANTHROPIC_API_KEY=sk-ant-...
+
+# OpenAI
+MODEL=openai/gpt-5
+OPENAI_API_KEY=sk-...
+
+# OpenRouter
+MODEL=openrouter/anthropic/claude-sonnet-4
+OPENROUTER_API_KEY=sk-or-...
+
+# LM Studio (local)
+MODEL=lmstudio/llama-3-8b
+# No API key needed - make sure LM Studio is running!
+```
+
 ## Usage
 
 To run the benchmark (automatically discovers and runs all tests):

bun.lock

@@ -8,6 +8,7 @@
         "@ai-sdk/anthropic": "^2.0.53",
         "@ai-sdk/mcp": "0.0.11",
         "@ai-sdk/openai": "^2.0.77",
+        "@ai-sdk/openai-compatible": "^1.0.28",
         "@openrouter/ai-sdk-provider": "^1.4.1",
         "@testing-library/svelte": "^5.2.9",
         "@testing-library/user-event": "^14.6.1",
@@ -39,6 +40,8 @@
 
     "@ai-sdk/openai": ["@ai-sdk/openai@2.0.77", "", { "dependencies": { "@ai-sdk/provider": "2.0.0", "@ai-sdk/provider-utils": "3.0.18" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-lEJ9vyWSU5VLo+6Msr6r32RnABf4SRxPSV3Hz1Yb5yt43bWYxbBzwaDNYGhJaDL6rCgfUVvcIf5TKiiEuVd4EQ=="],
 
+    "@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@1.0.28", "", { "dependencies": { "@ai-sdk/provider": "2.0.0", "@ai-sdk/provider-utils": "3.0.18" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-yKubDxLYtXyGUzkr9lNStf/lE/I+Okc8tmotvyABhsQHHieLKk6oV5fJeRJxhr67Ejhg+FRnwUOxAmjRoFM4dA=="],
+
     "@ai-sdk/provider": ["@ai-sdk/provider@2.0.0", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-6o7Y2SeO9vFKB8lArHXehNuusnpddKPk7xqL7T2/b+OvXMRIXUO1rR4wcv1hAFUAT9avGZshty3Wlua/XA7TvA=="],
 
     "@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@3.0.18", "", { "dependencies": { "@ai-sdk/provider": "2.0.0", "@standard-schema/spec": "^1.0.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-ypv1xXMsgGcNKUP+hglKqtdDuMg68nWHucPPAhIENrbFAI+xCHiqPVN8Zllxyv1TNZwGWUghPxJXU+Mqps0YRQ=="],

lib/providers.ts

@@ -1,6 +1,7 @@
 import { anthropic } from "@ai-sdk/anthropic";
 import { openai } from "@ai-sdk/openai";
 import { createOpenRouter } from "@openrouter/ai-sdk-provider";
+import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
 import type { LanguageModel } from "ai";
 
 interface ProviderConfig {
@@ -67,9 +68,22 @@ export function getModelProvider(config: ProviderConfig): LanguageModel {
     return openrouter.chat(modelPath);
   }
 
+  // Route to LM Studio provider (OpenAI-compatible)
+  if (modelString.startsWith("lmstudio/")) {
+    // Create LM Studio provider instance
+    const lmstudio = createOpenAICompatible({
+      name: "lmstudio",
+      baseURL: "http://localhost:1234/v1",
+    });
+
+    // Extract model name (e.g., "lmstudio/model-name" -> "model-name")
+    const modelName = modelString.replace("lmstudio/", "");
+    return lmstudio(modelName);
+  }
+
   // Invalid format
   throw new Error(
-    `Invalid MODEL format: "${modelString}". Must start with "anthropic/", "openai/", or "openrouter/"`,
+    `Invalid MODEL format: "${modelString}". Must start with "anthropic/", "openai/", "openrouter/", or "lmstudio/"`,
   );
 }
 
@@ -86,7 +100,7 @@ export function loadEnvConfig(): ProviderConfig {
   // Model is required
   if (!modelString) {
     throw new Error(
-      "MODEL environment variable is required. Format: 'anthropic/model-name', 'openai/model-name', or 'openrouter/provider/model-name'",
+      "MODEL environment variable is required. Format: 'anthropic/model-name', 'openai/model-name', 'openrouter/provider/model-name', or 'lmstudio/model-name'",
     );
   }
 

package.json

@@ -25,6 +25,7 @@
     "@ai-sdk/anthropic": "^2.0.53",
     "@ai-sdk/mcp": "0.0.11",
     "@ai-sdk/openai": "^2.0.77",
+    "@ai-sdk/openai-compatible": "^1.0.28",
     "@openrouter/ai-sdk-provider": "^1.4.1",
     "@testing-library/svelte": "^5.2.9",
     "@testing-library/user-event": "^14.6.1",