Added OpenAI compatible "input" field support in /v1/responses API (#69)

Changes to /v1/responses API
- Support for "endpoint" field for LiteLLM
- Support for array of messages in the "input" field

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

## Release Notes

**New Features**
* Added LiteLLM provider support with per-model endpoint configuration
* Enhanced multi-turn conversation support with conversation history
preservation
* Implemented system prompt override capability for personalized chat
interactions
* Extended input validation to support OpenAI-compatible conversation
formats

**Documentation**
* Updated API documentation with LiteLLM configuration examples and
endpoint handling guidance

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Author: olesho

agent-server/README.md

@@ -75,7 +75,7 @@ The server will start:
 
 Send a task to a connected browser agent and get response.
 
-**Request:**
+**Request (OpenAI):**
 ```json
 {
   "input": "Click the submit button",
@@ -101,6 +101,39 @@ Send a task to a connected browser agent and get response.
 }
 ```
 
+**Request (LiteLLM):**
+```json
+{
+  "input": "Navigate to google.com",
+  "url": "about:blank",
+  "wait_timeout": 5000,
+  "model": {
+    "main_model": {
+      "provider": "litellm",
+      "model": "qwen3:14b",
+      "endpoint": "http://localhost:4000",
+      "api_key": "sk-litellm-key"
+    },
+    "mini_model": {
+      "provider": "litellm",
+      "model": "qwen3:14b",
+      "endpoint": "http://localhost:4000"
+    },
+    "nano_model": {
+      "provider": "litellm",
+      "model": "qwen3:14b",
+      "endpoint": "http://localhost:4000"
+    }
+  }
+}
+```
+
+**Endpoint Configuration:**
+- `endpoint` can be specified per model tier (e.g., `main_model.endpoint`)
+- Or at top-level `model.endpoint` to apply to all tiers
+- Falls back to `LITELLM_ENDPOINT` environment variable if not provided
+- Required for LiteLLM provider unless set via environment variable
+
 **Response:**
 ```json
 [

agent-server/nodejs/CLAUDE.md

@@ -70,7 +70,7 @@ The eval-server is a **thin HTTP API wrapper for Browser Operator**. It provides
 
 Primary endpoint for sending tasks to browser agents.
 
-**Request:**
+**Request (OpenAI):**
 ```json
 {
   "input": "Click the submit button",
@@ -84,6 +84,79 @@ Primary endpoint for sending tasks to browser agents.
 }
 ```
 
+**Request (LiteLLM with endpoint):**
+```json
+{
+  "input": "Navigate to google.com",
+  "url": "about:blank",
+  "wait_timeout": 5000,
+  "model": {
+    "main_model": {
+      "provider": "litellm",
+      "model": "qwen3:14b",
+      "endpoint": "http://localhost:4000"
+    },
+    "mini_model": {
+      "provider": "litellm",
+      "model": "qwen3:14b",
+      "endpoint": "http://localhost:4000"
+    },
+    "nano_model": {
+      "provider": "litellm",
+      "model": "qwen3:14b",
+      "endpoint": "http://localhost:4000"
+    }
+  }
+}
+```
+
+**Note:** The `endpoint` field can be:
+- Specified per model tier (e.g., `main_model.endpoint`)
+- Specified at top-level (`model.endpoint`) to apply to all tiers
+- Omitted to use `LITELLM_ENDPOINT` environment variable
+
+**Request (Conversation State - OpenAI Responses API format):**
+```json
+{
+  "input": [
+    {
+      "role": "system",
+      "content": "You are a web automation expert."
+    },
+    {
+      "role": "user",
+      "content": "Navigate to bloomberg.com"
+    },
+    {
+      "role": "assistant",
+      "content": "I've navigated to bloomberg.com. I can see the homepage."
+    },
+    {
+      "role": "user",
+      "content": "Summarize todays news"
+    }
+  ],
+  "url": "https://bloomberg.com",
+  "model": {
+    "main_model": {
+      "provider": "litellm",
+      "model": "gemma3:12b",
+      "endpoint": "http://localhost:4000"
+    }
+  }
+}
+```
+
+**Input Format Options:**
+1. **String format**: `"input": "Your message"` (simple, single message)
+2. **Conversation array**: `"input": [{role, content}, ...]` (multi-turn with history)
+
+**Message Requirements:**
+- Each message needs `role` (`system`, `user`, or `assistant`) and `content` (string)
+- At least one `user` message required
+- System messages are extracted as system prompt
+- Maximum 100 messages, 10,000 characters each
+
 **Response (OpenAI-compatible format):**
 ```json
 [

agent-server/nodejs/README.md

@@ -365,16 +365,105 @@ The `/v1/responses` endpoint provides an OpenAI-compatible interface for chat re
 ```json
 {
   "input": "Your question or prompt here",
+  "url": "about:blank",
+  "wait_timeout": 5000,
   "model": {
     "main_model": {
       "provider": "openai",
       "model": "gpt-4",
       "api_key": "sk-..."
+    },
+    "mini_model": {
+      "provider": "openai",
+      "model": "gpt-4-mini",
+      "api_key": "sk-..."
+    },
+    "nano_model": {
+      "provider": "openai",
+      "model": "gpt-3.5-turbo",
+      "api_key": "sk-..."
     }
   }
 }
 ```
 
+**LiteLLM Provider with Endpoint:**
+```json
+{
+  "input": "Your question or prompt here",
+  "url": "about:blank",
+  "model": {
+    "main_model": {
+      "provider": "litellm",
+      "model": "qwen3:14b",
+      "endpoint": "http://localhost:4000"
+    },
+    "mini_model": {
+      "provider": "litellm",
+      "model": "qwen3:14b",
+      "endpoint": "http://localhost:4000"
+    },
+    "nano_model": {
+      "provider": "litellm",
+      "model": "qwen3:14b",
+      "endpoint": "http://localhost:4000"
+    }
+  }
+}
+```
+
+**Endpoint Configuration Priority:**
+1. Per-tier endpoint (e.g., `main_model.endpoint`) - highest priority
+2. Top-level endpoint (e.g., `model.endpoint`) - applies to all tiers
+3. Environment variable `LITELLM_ENDPOINT` - fallback
+4. Default: `http://localhost:4000` (LiteLLMProvider built-in default)
+
+**Conversation State Format (OpenAI Responses API):**
+
+You can also provide conversation history using an array of messages:
+
+```json
+{
+  "input": [
+    {
+      "role": "system",
+      "content": "You are a web automation expert."
+    },
+    {
+      "role": "user",
+      "content": "Navigate to bloomberg.com"
+    },
+    {
+      "role": "assistant",
+      "content": "I've navigated to bloomberg.com. I can see the homepage."
+    },
+    {
+      "role": "user",
+      "content": "Summarize todays news"
+    }
+  ],
+  "url": "https://bloomberg.com",
+  "model": {
+    "main_model": {
+      "provider": "litellm",
+      "model": "gemma3:12b",
+      "endpoint": "http://localhost:4000"
+    }
+  }
+}
+```
+
+**Message Roles:**
+- `system` - System prompt/instructions (extracted and used as system prompt)
+- `user` - User messages
+- `assistant` - Previous assistant responses (for conversation history)
+
+**Requirements:**
+- At least one `user` message must be present
+- Each message must have `role` and `content` fields
+- Maximum 100 messages per conversation
+- Maximum 10,000 characters per message
+
 **Response Format:**
 ```json
 [