AI Mate Space Documentation

AI Mate Space is a multi-tenant AI Agent management platform providing the following core capabilities:

• Integration and management of upstream model providers and models.
• Creation, configuration, and orchestration of AI Agents.
• Management of users, API Keys, and model whitelists.
• Enabling and customizing Skills.
• Accessing upstream MCP Servers and tools via an MCP Gateway.
• Configuration of scheduled tasks and notification channels.
• Workspace collaboration and session management.
• Monitoring of statistics, logs, audits, and security events.
• Billing, top-ups, invoicing, and ticket management.
• Service invocation via OpenAI-compatible APIs and MCP Gateway.

Orchestrator Agents are designed to coordinate multiple sub-agents to complete complex tasks. They are best suited for:

• Multi-step workflows (e.g., writing code then writing documentation).
• Scenario-based allocation (e.g., routing from pre-sales consulting to after-sales support).
• Collaborative tasks requiring multiple specialized agents.

Key Constraints:

• Current versions only support sequential execution; intelligent delegation and parallel execution are disabled in the UI.
• An Orchestrator Agent cannot be its own sub-agent (recursion is prevented).
• Recommended Temperature: 0.1 to 0.3 for precise decision-making.

A Workspace is a project-level collaboration unit used to create independent environments for different projects or teams. Each workspace encapsulates the following components:

• Task Board: Visual task management with drag-and-drop status tracking.
• File Management: Uploads and output files organized by directory.
• Agent Conversations: Chat history bound specifically to the workspace.
• Workflows: Reusable sequences of tasks.
• Memory Management: Structured memory files (SPEC, PLAN, LOG, KNOWLEDGE) with versioning support.

Model Groups allow you to combine multiple models and select between them based on a specific routing strategy. This is typically used for:

• Failover: Automatically switching to a backup model if a high-priority model fails.
• Load Balancing: Distributing traffic via round-robin or weighted distribution.
• Cost Control: Prioritizing lower-cost models and falling back to more expensive ones upon failure.

Key Constraint: The system supports a maximum of 3 candidate models for failover attempts before giving up.

The platform operates in two modes:

1. SaaS Mode: Hosted and operated by the platform provider. The Terms of Service apply directly between the user and the platform operator.
2. Private Mode: Self-deployed by the user. In this mode, the rights and obligations normally borne by the platform operator are transferred to the party deploying the software; the original platform operator is no longer the counterparty to the terms.

Quickly add funds to your account using WeChat Pay or Alipay. Note that payment orders are valid for 30 minutes.

Steps:

1. Click the Recharge (充值) button in the top right of the Billing Overview page.
2. Select the Online Recharge (线上充值) tab.
3. Choose a preset amount (50, 100, 200, 500 CNY) or enter a custom amount.
4. Select the payment method (WeChat Pay or Alipay).
5. Verify the total amount and click Pay (去支付).
6. Complete the payment via the provided QR code or redirect page.

Set up notifications to be alerted when your account balance drops below a specific threshold. This requires a configured notification channel (e.g., Email, SMS, DingTalk, Lark, or WeChat Work).

Steps:

1. Go to the Billing Overview page.
2. In the "Balance Alert Settings" (余额预警设置) area, toggle the Enable Alert (启用预警) switch.
3. Enter a value greater than 0 in the Alert Threshold (预警阈值) field.
4. Click Save Settings (保存设置).

Plugins are loaded by the platform at startup. Tenant administrators cannot upload, install, or uninstall plugin code via the console; they can only manage the enablement, disabling, and configuration of plugins for their specific tenant.

Lifecycle Stages:

• Global Startup: Occurs when the service starts. A failure in one plugin does not block others.
• Global Shutdown: Occurs when the service closes.
• Tenant Enablement: Triggered when a plugin status changes from disabled to enabled. Exceptions will bubble up.
• Tenant Disablement: Triggered when a plugin status changes from enabled to disabled. Exceptions will bubble up. Historical data is preserved.
• Configuration Change: Triggered when an enabled plugin's configuration is modified. Exceptions are silently ignored and recorded as warnings.

The Hybrid Router automatically routes user messages to the most appropriate agent using a three-layer matching mechanism. The process stops as soon as a match is found:

1. Keyword Matching: Checks for exact keywords or regular expressions. This is the fastest and does not consume LLM tokens.
2. Semantic Matching: Uses vector similarity between the user message and agent descriptions.
3. LLM Decision: The LLM selects the best candidate agent from a list for complex scenarios.