Mappedin Blaxel Agent

An AI-powered indoor navigation agent built on the Blaxel platform, demonstrating seamless integration of Blaxel's core features: Model API, MCP server hosting, and serverless agent deployment.

🚀 Blaxel Features Showcase

This project leverages three key Blaxel platform capabilities:

1. Blaxel Model API 🤖

The agent uses Blaxel's unified Model API gateway to access AI models through a single, standardized interface:

import { blModel } from "@blaxel/vercel";

const model = await blModel("gpt-5-mini");

Key Benefits:

Unified Gateway: Single endpoint to access hundreds of models from OpenAI, Anthropic, Cohere, and other providers
Intelligent Routing: Automatic routing across the Global Agentics Network based on deployment policies
Built-in Observability: Embedded LLM telemetry with OpenTelemetry for monitoring latency, tokens, and performance
Seamless Integration: Works with popular frameworks like Vercel AI SDK

2. Blaxel MCP Server Integration 🔧

The agent connects to custom MCP (Model Context Protocol) servers hosted on Blaxel to access specialized tools:

import { blTools } from "@blaxel/vercel";

const tools = await blTools(["mappedin-mcp"]);

Key Benefits:

Standardized Tool Access: MCP servers expose APIs and capabilities in a format AI agents can understand
Serverless MCP Hosting: Deploy MCP servers as auto-scalable HTTPS endpoints on Global Agentics Network
Streamable HTTP Transport: Modern, efficient communication protocol for agent-tool interactions
Global Endpoints: Each MCP server gets a unique global endpoint accessible from anywhere

In this project, the mappedin-mcp server provides indoor navigation tools like getMapInstructions for wayfinding.

3. Blaxel Agent Hosting 🌐

The entire agent is deployed as a serverless, auto-scalable API on Blaxel's infrastructure:

Configuration (blaxel.toml):

type = "agent"
functions = ["all"]        # Access to all MCP servers
models = ["sandbox-openai"] # Model access configuration

Key Benefits:

Zero-Configuration Deployment: Simply push code and Blaxel handles build, deployment, and scaling
Framework Agnostic: Works with any TypeScript/Python agent framework
Global Agentics Network: Ultra-fast sandboxed VMs with sub-20ms boot times
Automatic GitHub Sync: Continuous deployment from your repository
Built-in Telemetry: Comprehensive observability for agent runs, including end-to-end latency and time-to-first-token

Inference Endpoint:

POST https://run.blaxel.ai/{workspace}/agents/mappedin-blaxel-agent

🏗️ Architecture

User Request
    ↓
Blaxel Agent (TypeScript + Vercel AI SDK)
    ↓
Blaxel Model API → GPT-5-mini (via unified gateway)
    ↓
Blaxel MCP Server → mappedin-mcp (navigation tools)
    ↓
Response with navigation instructions

📦 Dependencies

The project uses Blaxel's official SDKs:

@blaxel/core - Core Blaxel platform utilities
@blaxel/vercel - Vercel AI SDK integration for Blaxel
@blaxel/telemetry - Built-in observability and monitoring
ai - Vercel AI SDK for agent orchestration
zod - Schema validation for tool inputs

🎯 Use Case: Indoor Navigation

This agent helps users navigate indoor spaces (malls, airports, hospitals) by:

Receiving natural language queries (e.g., "How do I get to the food court?")
Using the getMapInstructions tool from the Mappedin MCP server
Returning step-by-step navigation instructions

🔗 Learn More

📄 License

MIT