Manifest

**Manifest** is a self-hosted AI agent proxy gateway that can uniformly access cloud APIs, subscription accounts and local models, and provides agents with model routing, automatic fallback, request auditing, usage statistics and other capabilities. --- ##Application scenarios Manifest is suitable for the following users: * Run an AI Agent or automated tool * Developing applications that require calling multiple large models * Deploy AI services on home servers or private environments * Want to manage cloud models and local models in a unified manner * Need to view model call records, costs, usage and performance It can help solve: * Different model platform configurations are scattered * API keys and model accounts are difficult to manage uniformly * Complicated model switching and alternate model configuration * Manual switching is required after the main model request fails * It is difficult to uniformly view call records, Token usage and fees --- ##Application Overview Manifest is located between the AI agent and the model provider ** and is responsible for managing the model channels, routing policies and request records used by the agent. Currently available connections include: * OpenAI * Anthropic * Google * DeepSeek * Qwen * Ollama * LM Studio * llama.cpp * Customize OpenAI compatible interfaces * Customize Anthropic compatible interfaces You can add different API channels or local models in Manifest, and configure the preferred model and multiple alternate models separately for each agent. > **Manifest is not a traditional model transit service. ** > It is more inclined to the model agent layer of the agent, which is responsible for functions such as model routing, request audit, usage statistics, and automatic fault fallback. It does not simply forward all models directly to the agent for use. --- #Main characteristics ##Unified management of multiple model channels Manifest can uniformly manage multiple model sources, including: * API Key * Model subscription account * Local deployment model * OpenAI compatible interface * Anthropic compatible interface Models from different sources can be centrally configured and then allocated to different agents for use. --- ##Model routing and automatic fallback Each agent can configure model routing individually. You can set: * ** Preferred model **: The preferred model under normal circumstances * ** Alternate models **: Multiple models can be configured. When an exception occurs in the preferred model, automatic fallback will be performed according to the routing configuration When a request failure, timeout, or other exception occurs in the preferred model, an alternate model can be automatically tried to improve the stability of the agent's runtime. --- ##Requestaudit, cost and usage analysis Manifest can record the model calls of the agent, including: * requested content * model response * request status * Time to respond * Token usage * call cost It is convenient to view the actual use of different agents and models, and to troubleshoot problems such as request failures. --- ##Supports local models and custom services In addition to the cloud model, you can also connect: * Ollama * LM Studio * llama.cpp * Other OpenAI API compatible model services On-premises and cloud models can be managed together and selected based on cost, speed, privacy, and model capabilities. --- #Get started quickly When using Manifest, you can follow the following process to complete the basic configuration. ### 1. Create an agent After entering Manifest, first create an agent that you need to use. After the agent is created, it is configured with the model that actually needs to be called. ### 2. Add API channels Add the model API channels you need to use, and fill in the corresponding API Key, service address and other information. You can add one or more different model channels as needed. ### 3. Configure model routing Enter the agent you just created and configure the model in **"Routing"**. At least settings are required: * ** Preferred model ** * ** One or more alternate models ** The preferred model is used for normal requests; when an exception occurs in the preferred model, the system can continue to try the configured alternate model. > After creating an agent and adding API channels, routing configuration needs to be completed, otherwise the agent may not be able to call the model normally. ### 4. Get call information Go to the **"Settings"** tab on the agent's home page and swipe to the bottom of the page to view the tutorial. It provides information needed to invoke the agent, such as: * API address * API Key * access mode Configure it into the AI application or Agent you need to use according to the information provided on the page. --- #Lazy Cat Version Instructions ##Administrator account and login The lazy cat version will generate and save an administrator account and password for the ** current Manifest application instance **, and does not provide a fixed default password common to all instances. When you enter the settings or login page for the first time, the system will automatically fill in the account password corresponding to the current instance, and then manually complete the login or submission after confirming the information. After opening the application for the first time, you need to enter: `/setup` Complete the administrator account creation process. After setting up, this account is used to log in to the Manifest web console. > The Manifest console uses the administrator account to log in; external applications or AI Agents use the API information corresponding to the agent to make calls. --- ###Automatic filling by lazy cat The app has been adapted to the lazy cat auto-fill function and can be used to: * login page * First set up page * Change password page Automatic fill is only responsible for filling in the form content and will not automatically click the login, save or submit button. After completing the filling, please confirm the information and manually click the corresponding button. --- ##API address and service port The API entry for the lazy cat version is located under the application domain name: `https://Application Domain Name/v1` The application internal service port is: `2099` For specific information such as **API address and API Key required to call the agent, please enter the **"Settings"** page of the corresponding agent and check it in the tutorial at the bottom. --- ##Data persistence Manifest's main persistence data catalogs are as follows: | data type | path | | ------------- | -------------------------------- | | PostgreSQL data| `/lzcapp/var/postgresql` | | request record | `/lzcapp/var/request-recordings` | The authentication key, database password, and model credential encryption key are automatically generated and persisted by the lazy cat. ** Please clean up application data carefully. ** Cleanup may result in the loss of administrator accounts, agent configuration, model configuration, authentication information, and historical request records. --- ##Request records and privacy Request records are retained by default for **365 days **. Records may include: * Messages sent by users to models * Text returned by the model * Token usage * request status * Call time consuming * Other request-related information If Manifest is used to process personal information, work data, or other sensitive content, please note the preservation and access rights of the request records. --- > ** Note: ** When using foreign models, corresponding network access conditions are required. If you cannot solve the problem yourself, please contact the VIP group for help.