Shizhi AI

What is StartSmart AI WiseModel?

WiseModel, developed by Beijing Shizhi Technology Co., Ltd., is an open AI model and agent ecosystem platform. It brings together models, datasets, code, agent communities, MaaS interfaces, GPU computing power, images, and application spaces within a single development framework.

The platform is intended for developers, researchers, and teams who need to find resources, invoke models online, rent computing power for training, or publish AI projects. It is not a single chatbot; users usually have to first select the appropriate resources and read the relevant licenses before deciding whether to download, invoke, or deploy them.

Main functions

Models, datasets, and code resources

• The open-source resources module is used to browse, search for, and filter AI models, datasets, and code; project details may include documentation, a list of files, and version information.
• Registered users can create projects and publish their own resources, which is ideal for teams to showcase model results, maintain datasets, or share reproducible code.
• The inclusion of open resources on the platform does not mean that all entries use the same license; it is necessary to check separately the licensing terms for the models, weights, data, and code before downloading them.
• During this verification, some resource lists showed 0 entries; the number of resources recorded in the past cannot be considered as the amount of resources available for download at present.

MaaS model API

• MaaS allows users to call APIs online without having to deploy models locally, and it enables filtering of models based on task type, parameter size, and context length.
• The usage process includes supplier authorization, purchase of resource packages, and API Key management; it is suitable for integrating text generation or other model capabilities into applications.
• The official documentation describes the billing method as pay-as-you-go, but different models should have their own prices, resource packages, and licensing conditions.
• During this verification, the MaaS market showed that no models were available; therefore, the capabilities mentioned in the document cannot be interpreted as indicating the existence of models that can be purchased immediately.

GPU computing power and training environment

• The computing power platform displays GPU configurations such as 4090, H20, H800, H100, A100, and A800; filtering is possible by the number of cards, CPU, memory, system drive, and duration in hours.
• Users can create training or inference environments by leveraging platform models and images, thereby reducing the effort required to prepare drivers, dependencies, and servers on their own.
• The price increases depending on the configuration and the number of cards, and the availability of resources changes in real time; at the time of this verification, most of the configurations listed were out of stock.
• Before renting, it is necessary to verify the rules regarding video memory, storage, network, location, images, data migration, billing for shutdowns, and instance termination.

Image center and application space

• The image center is used to manage Docker development environments, enabling teams to reuse frameworks, drivers, and inference dependencies.
• The application space is used to create and manage AI application services, and it can serve as a platform for demonstrating models, providing online experiences, or delivering projects.
• During this verification, the list of available applications showed none, and a message indicating that it was not possible to retrieve application details appeared; the concept of creation capabilities and the inventory available in the public market should be understood separately.

AgentVerse Agent Community

• AgentVerse is used to explore, create, and manage agents, and it offers features for community posting, commenting, interaction, and networking.
• The user agreement specifies that Agents are created by human users and connected to the platform through its SDK, and they do not possess independent legal status.
• The creator is responsible for managing the Agent’s configuration, authentication credentials, content distribution, and interaction behaviors, as well as bearing accountability for the actions taken by the Agent.
• The platform offers functions such as prompt injection prevention, protection against social engineering attacks, and filtering of input and output data containing sensitive information, but it does not guarantee complete protection.

Comparison of core modules

Registration and complete workflow

1. Register for a StartSmart AI account, complete the email verification, and read the user agreement and privacy policy.
2. Search for models, datasets, or code in open-source resources, and verify the project maintainer, version, files, and license.
3. If it is to be used only locally, download the resources according to the project instructions, and note the licensing requirements for the model weights, the code, and the data.
4. To make an online call, go to MaaS to view the available models, complete the supplier authorization and purchase of resource packages, and then create an API Key.
5. If you need training or dedicated inference, check the GPU availability and select an instance based on VRAM, CPU, memory, system drive, and the number of cards.
6. Select or create an image, upload the code and data, and verify dependencies, GPU memory usage, and inference results in an isolated environment.
7. An application space needs to be created when an online demonstration is required; agents need to be configured and their authentication credentials protected when interaction with them is necessary.
8. Once the project is completed, stop its operation and release the computing resources, revoke any keys that are no longer in use, back up the necessary outputs, and delete sensitive data.

Price and current availability

The GPU prices listed above are the representative starting prices shown on the page at the time of verification; in most cases, the corresponding resources are listed as out of stock, and therefore cannot be considered as an assurance that they can be ordered at those prices. The actual prices, inventory levels, promotions, resource packages, and settlement amounts should be based on the information displayed on the real-time page within the account.

The public agreement does not specify uniform rules for refunds, nor does it outline how refunds should be handled in cases of service shutdown, waiting times, instance failures, or unused resource packages. It is necessary to review the rules applicable to each individual order before making a payment or renting services, and to test the workflow using instances with small capacities and for short periods of time.

Precautions for API integration

• The MaaS documentation states that online model APIs and API key management are available, but the interface addresses, parameters, and authentication methods shall be as specified on the details page of each particular model.
• Different supplier models may require separate authorization, and resource packages cannot be used across models, suppliers, or accounts by default.
• The MaaS market is currently empty; it is therefore not possible to determine the list of available models, the pricing per invocation, the concurrency limits, the rate limits, or the maximum context size.
• For production environments, the minimum level of permissions should be assigned to the keys, which should be stored in the server’s security settings; writing these keys to the frontend, public repositories, logs, or Agent responses should be avoided.
• The caller should implement timeout, retry, rate limiting, error handling, balance alerts, and result verification, while also keeping track of the model version and key parameters.

Open-source status and license

• Shizhi AI is a community that offers open AI resources; however, it was not found that the entire code of its website is governed by a unified open-source license, so the platform itself cannot be considered open-source software.
• The model page can host weights, descriptions, and code, but the model weights, inference code, training data, documentation, and trademarks may be subject to different licenses.
• Users must possess the necessary rights before uploading resources, and must provide explanations, licenses, or contributor agreements as required by the project.
• The presence of third-party models on the platform does not imply that StartSmart AI grants any additional commercial rights, nor does it mean that the platform guarantees the copyright of the training data or the legitimacy of the outputs.
• The AgentVerse repository with the same name that is found through searches may belong to another team; it can only be considered a component of OriginAI if the project page or platform documentation explicitly indicates a connection to it.

Which users are it suitable for

• Model researchers: Look for weights, datasets, code for reproducing papers, and model descriptions.
• Application developers: Use MaaS to integrate model inference capabilities into websites, services, or internal tools.
• Algorithm engineer: Renting GPUs for fine-tuning, evaluation, quantization, inference, and image verification.
• Open-source team: Publishes model, data, or code projects, maintains versions, and collaborates with the community.
• Agent developers: Create agents, integrate SDKs, and participate in community interactions.
• Corporate R&D team: Evaluates options for domestic resource hosting, computing power, and application deployment, while also needing to confirm inventory levels and service levels.

Privacy, security, and content rights

• The platform may record information such as the browser used, operating system, IP address, pages visited, time of requests, and user behavior logs; for registration and access to certain features, information like phone number, email address, username, and profile picture may be required.
• AgentVerse handles the agent’s name, capability descriptions, avatar, past published content, and interaction data, and links them to the creator’s account.
• The privacy policy states that information is retained only for as long as it is necessary to achieve the intended purpose, and it is deleted or anonymized after that period; users can log in to manage their basic information and request its deletion under certain circumstances.
• Training data, prompts, keys, and model outputs may contain personal information, trade secrets, or copyrighted content; therefore, masking and authorization checks must be carried out before uploading them.
• The user agreement sets fairly strict rules regarding the platform’s right to generate content; before using such content for commercial purposes, redistributing it, or delivering it to customers, it is necessary to verify the specific terms related to the models and individual services.
• Agent security filtering cannot replace permission isolation, input validation, output review, and key rotation; the creator remains responsible for the behavior of the Agent.

Advantages and limitations

Main advantages

• Resource retrieval, APIs, computing power, images, and application deployment are all gathered in one platform, facilitating the creation of a complete development workflow.
• MaaS and GPU leasing offer both online and self-deployment options, allowing teams to choose based on cost, control, and performance.
• AgentVerse incorporates agent deployment and community interaction into its platform, making it suitable for exploring the Agent ecosystem.

Current restrictions

• During this verification, several market lists were empty; a large portion of the computing capacity was sold out, resulting in an actual available capacity that is lower than what is shown in the feature list.
• MaaS lacks currently available models, which prevents unified comparison of unit prices, concurrency levels, context factors, and supplier service levels on public pages.
• The platform resources come from various publishers, and their quality, frequency of maintenance, licenses, and security levels vary.
• The rules regarding unified refunds, compensation for resource failures, and the handling of unused balances have not been made public yet; they need to be confirmed separately before making a payment.

Frequently Asked Questions

Is Shizhi AI a chatbot?

It is not a single chatbot; rather, it is an AI development platform that consists of models, data, code, APIs, GPU computing power, images, applications, and a community of agents.

Can all the models on the platform be used for commercial purposes for free?

No. Each model, dataset, and code project has its own licensing terms; downloading them publicly does not mean they can be used for commercial purposes free of charge, nor does it imply that the weights, data, and code are covered by the same license.

Can the MaaS API be called directly now?

The platform documentation outlines the procedures for obtaining authorization, accessing resource packages, and using API keys; however, during this verification the model market showed no available models. It is necessary to log in to determine the actual list of available models before planning the integration process.

How much does GPU computing power cost?

The page shows the hourly price based on the GPU model, the number of cards, and the resource configuration; the starting price for a single card is 1.88 yuan per hour. Inventory and prices change in real time, and most configurations are currently listed as out of stock.

Is the Shizhi AI platform open-source?

At present, no unified open-source license for the overall code of the platform’s website has been identified. The fact that the platform makes its resources available publicly does not mean that the code of its products is itself open source.

How to reduce the risks associated with computing costs?

First, use a minimal configuration and short-duration tasks to verify the requirements regarding images, data, and video memory; set budgets and monitoring mechanisms, and stop the instance and release its resources as soon as the task is completed.

Who is responsible for the Agent’s behavior?

The user agreement specifies that responsible human users are accountable for these matters. Creators must protect credentials, review the content that is published, set up security filters, and address any abnormalities promptly.

Summary

WiseModel from Shizhi AI is suitable for users who wish to integrate open resources, MaaS, GPU-based training, as well as image and Agent development into a single workflow. Before using it, it is essential to verify the actual inventory of available models, the pricing rules, and the licensing options for each individual service; one cannot assume that a service is available just based on its functional categories.