【Platform Guide】Lancun Open Platform (Hardware AI Runtime) Comprehensive Introduction: Let AI Run in the Hardware World

Key Takeaway: The Lancun Open Platform (Hardware AI Runtime) is a complete building and operation system for AI toys, robots, IoT terminals, small appliances, educational devices, and enterprise hardware products, providing everything from agent creation to device operation. It integrates speech engines, model invocation, Skills, knowledge bases, device protocols, memory, logs, OTA, and hardware debugging into one pipeline, allowing hardware teams to complete a POC first and then gradually move into productization and batch operations.

1. The Platform Solves Not "Chat" but Hardware Operation

When AI enters hardware, a single voice command typically goes through multiple stages:

Microphone capture → Speech recognition → Agent understanding → Model and tool routing → Device command dispatch → Action/screen/sound feedback → Memory and log saving

If a company only integrates a large model API, it still needs to handle device protocols, tool invocation, execution results, error retries, user binding, usage statistics, and firmware upgrades. The value of the Lancun Open Platform is to encapsulate this complex pipeline into a runtime that hardware products can use.

2. Three-Layer Object Relationship of the Platform

The platform organizes hardware projects around three objects:

1. Enterprise or Merchant: Represents the brand, customer, or development team, managing applications, devices, usage, and keys.
2. Application and Agent: The application defines the product's usage boundaries and business scenarios; the agent carries the persona, model, Skills, knowledge base, and memory.
3. Device: Devices connect to the agent via standard protocols or the Lancard LC.AI module, handling voice, sensing, and action execution.

These three layers can form isolation relationships to prevent memory confusion or data cross-talk between different enterprises, applications, and devices.

3. Agent Runtime

The platform provides an agent runtime that takes devices from "receiving a sentence" to a complete loop of "understanding, deciding, executing, and feedback."

3.1 Intent Understanding

What users say is not standard API commands but natural language. The platform needs to determine what the user wants to do, which device is involved, what parameters are needed, and whether external capabilities need to be invoked.

For example:

• "I'm a bit cold, but not too noisy" requires understanding both temperature and noise preferences;
• "Check the battery level" requires querying device status;
• "I'm stuck" may trigger robot action strategies;
• "Tell a story about this mountain" requires invoking knowledge base and content generation capabilities.

3.2 Tool and Skill Invocation

The platform encapsulates external capabilities as Skills or tools, allowing agents to call weather, device status, sensors, lights, air conditioning, actions, battery level, knowledge bases, and third-party APIs.

A Skill should at least specify:

• Applicable scenarios;
• Trigger conditions;
• Input parameters;
• Executable actions;
• Return results;
• How to respond on failure.

This way, hardware teams don't need to write all capabilities into a complex prompt; instead, they can break them into maintainable, reusable skill units.

4. Full-Chain Speech Engine

The platform covers the main stages from speech input to speech output:

• Microphone capture;
• ASR speech recognition;
• LLM or other model understanding;
• TTS speech synthesis;
• Synchronization of sound, actions, and device status;
• Notifications for network disconnection, low battery, and abnormal states.

The platform can select different models and routing strategies based on task complexity: simple commands prioritize lighter paths, while complex understanding, knowledge retrieval, and multi-device collaboration enter stronger models and toolchains.

5. Standardized Device Protocols

Hardware teams don't need to design every device message from scratch. The platform supports hardware integration through standardized device protocols and handles device status, sensor data, and execution commands.

Connection methods mentioned in the current platform introduction include WebSocket, MQTT, UDP, etc. Actual projects can choose appropriate protocols based on network conditions, device types, real-time requirements, and security requirements.

After device integration, the platform needs to know:

• Who the device is;
• What the device's current status is;
• What commands the device supports;
• What sensors and peripherals the device has;
• Whether commands actually executed successfully;
• How to return results on execution failure.

This is also the key to the collaboration between the Lancard LC.AI standard module and the platform: the module connects physical devices, while the platform manages agents and device capabilities.

6. Knowledge Base and Long-Term Memory

The platform's knowledge base addresses "what the product should know," while long-term memory addresses "who the product should remember."

Knowledge Base

Suitable for importing:

• Brand materials;
• Product descriptions;
• Educational content;
• Story materials;
• Venue, scenic area, or exhibition hall materials;
• Industry knowledge and after-sales Q&A.

Long-Term Memory

Suitable for saving:

• User names and role relationships;
• User preferences;
• Important events;
• Character settings;
• Historical task and interaction summaries;
• Device status and usage habits.

The knowledge base ensures content consistency, and memory ensures interaction continuity; both need to be managed separately with permissions and deletion mechanisms.

7. Enterprise-Level Isolation and Operations Backend

For enterprises and small B customers, the platform needs to move from "can run" to "can manage":

• Enterprise and merchant isolation;
• Layered management of applications, agents, and devices;
• Device lists and batch binding;
• Conversation logs and memory summaries;
• Operation logs and server-side logs;
• Usage, credits, and billing statistics;
• Subscription plans and resource management;
• Merchant-owned Skills and tools;
• Custom voices, quick replies, and knowledge bases;
• Open API key management.

For large brands, the platform should not replace their existing apps or backends but serve as their backend intelligent capability layer. Brands can retain their own user interfaces and use the Lancun platform as the foundation for agent operation, device integration, model invocation, and hardware execution.

8. OTA and Device Lifecycle

Hardware products continue to change after launch, and OTA is an important part of platform-based delivery. The platform needs to support:

1. Establishing version relationships for devices, products, and firmware;
2. Pushing firmware or resource updates;
3. Recording upgrade results;
4. Retrying or rolling back failed devices;
5. Allowing old and new batches of devices to run continuously on the same platform.

For AI toys, music, stories, skills, and character content can also be continuously updated through the platform; for robots and IoT devices, OTA also needs to coordinate with device protocols, action libraries, and security policies.

9. Complete Developer Integration Process

Step 1: Registration

Enter the platform, click "Get Started," and complete registration using a phone number and verification code.

Step 2: Create an Application

Create an application first, then create an agent. The application determines the product scenario, user scope, and operational boundaries.

Step 3: Configure the Agent

Fill in the name, persona, role background, prompt, ASR, LLM, TTS, and voice.

Step 4: Configure Knowledge Base and Skills

Import product materials or scenario content, select existing skills, or create custom Skills.

Step 5: Web Chat Testing

Without physical devices, you can first chat with the agent on the web to check persona, knowledge base, response speed, and task logic.

Step 6: Bind Devices

After obtaining the Lancard LC.AI test module, complete binding via the six-digit verification code broadcast by the device, MAC address, or QR code scanning.

Step 7: Hardware Debugging

Test wake-up, continuous conversation, button interruption, sensor triggers, action execution, low battery feedback, network disconnection prompts, and recovery state after restart.

Step 8: Enter Productization

Complete device protocols, logs, OTA, batch binding, usage management, exception handling, and mass production testing before moving to larger-scale device integration.

10. Which Teams Is the Platform Suitable For

Individual Developers

Suitable for quickly experiencing voice, Skills, knowledge bases, and standard modules, and completing a hardware-level POC first.

AI Product Teams

Suitable for validating persona, emotion, memory, knowledge bases, and multi-device collaboration without building a complete cloud service first.

Embedded Teams

Suitable for focusing on hardware structure, sensors, and actuators, while the platform handles agents and cloud capabilities.

Brands and Small B Customers

Suitable for connecting their own products, user interfaces, and business APIs to the platform, gaining manageable intelligent hardware capabilities.

Solution Teams

Suitable for reusing the same set of agents, Skills, and device protocols across multiple scenarios such as toys, home, education, retail, and industrial IoT.

11. Key Questions When Choosing a Platform

1. Does it provide a hardware runtime, not just a model API?
2. Are device protocols and execution results traceable?
3. Does it support Skills, knowledge bases, and long-term memory?
4. Does it support isolation between enterprises, agents, and devices?
5. Does it have logs, usage, OTA, batch binding, and Open API?
6. Can it scale from small POCs to multi-device and mass production operations?

FAQ

Can the platform work without the Lancard module?

The platform supports various AI hardware integrations, depending on device protocols and project adaptation; the Lancard LC.AI is the standard physical access method provided by Lancun.

What is the difference between the platform and a regular large model backend?

A regular large model backend mainly handles model invocation, while the Lancun Open Platform also handles agents, Skills, knowledge bases, device protocols, sensors, execution feedback, logs, OTA, and hardware lifecycle.

Do small teams need to deploy their own servers?

You can first use the platform to complete agent and hardware POCs, then plan deeper integration based on product scale, business systems, and deployment requirements.

Does it support your own app?

Yes, it supports connecting with brand-owned apps, backend systems, or mini-programs through platform capabilities on a project basis; specific interfaces and permissions are subject to the current version of the platform.

Apply for the Lancard LC-1.7 standard module test and enjoy free token testing.