TinyAGI

TinyAGI is a multi-agent, multi-team collaborative personal AI assistant system. It organizes multiple AI assistants with different roles into an "office": you can create Agents dedicated to writing code, retrieving data, organizing documents, and managing projects, or you can organize multiple Agents into teams and let them work through tasks, messages and queues collaborate on processing work. In Lazy Cat Micro Services, TinyAGI has integrated the TinyOffice Web console. After installation, open the application from the browser and you can manage agents, teams, tasks, logs, model keys and message channels through the graphical interface. ##What is suitable for doing - Build your own multi-agent workbench. - Let different Agents assume different responsibilities, such as R & D, review, data collation, operations, customer service, etc. - Route messages to the specified Agent or team through `@agent` or `@team`. - Use project kanban to manage tasks and assign tasks to agents for execution. - Receive messages through Telegram, Discord, WhatsApp and other channels and hand them over to Agent for processing. - Run in the background for long periods of time, retaining queue, session, settings, and workspace data. ##Step 1: Confirm that the Control page is connected normally After opening Control, if you can see the three tabs 'Overview','Services', and 'Logs' at the top of the page, it means that the front end has been connected to the TinyAGI backend. ##Step 2: Configure the model key For an Agent to actually perform a task, it must first configure available model vendors. Recommended path: 1. Open Control on the left. 2. Go to the Services tab. 3. Select a supplier in Built-in Providers: - `Anthropic`: Enter the Anthropic API Key, or OAuth Token. - `OpenAI`: Fill in the OpenAI API Key. 4. Click Save to save. If you use a third-party model service that is compatible with OpenAI or Anthropic protocols: 1. Open Control-> Services. 2. Click the plus sign in Custom Providers. 3. Fill in the Provider ID, name, Base URL, API Key, and model name. 4. After saving, set the Provider to the corresponding ID when creating the Agent. You can refer to the attached figure to use DeepSeek configuration. When the model key is not configured, the interface can open normally, but the Agent will fail to perform tasks or login/API Key-related errors will appear in the log. ##Step 3: Create an Agent Agent is the basic unit of work of TinyAGI. 1. Click the Manage button next to the `Agents` area on the left, or open the `/agents` page. 2. Click `Add Agent`. 3. Fill in: - `Agent ID`: The unique ID of the Agent, such as `coder`,`writer`, and `researcher`. Do not include spaces. - `Display Name`: Display name, such as `Coder`. - `Provider`: Select `anthropic`,`openai`,`opencode` or a custom provider. - `Model`: Fill in the model name, such as `sonnet`,`opus`,`gpt-4.1`,`gpt-5-codex`. - `System Prompt`: Optional, used to define the roles, responsibilities, and constraints of this Agent. 4. Click `Create Agent`. After creation, the Agent will appear in the `Agents` list on the left. Click on an Agent to enter the details page and view information such as conversations, skills, memories, heartbeats, and system prompts. ##Send a message The most commonly used entry is the message entry box at the bottom right of the Office page. Send to the specified Agent: ```text @coder helped me check the launch process of this project and list potential problems. ``` ##Using Task Board The Projects page is used to manage tasks. It is suitable for breaking things to be done into clear items and then handing them over to an Agent or team for execution.