# A2A Net A2A Net is an AI agent platform for B2B SaaS teams. It turns an existing OpenAPI specification or MCP server into a customer-facing agent that can perform multi-step workflows across a website, app, Slack, Teams/Copilot, and Gemini Enterprise. - Findings24: - 網站: - 價格: Freemium - 分類: AI 智能體建立工具, AI 智能體基礎設施 ## 什麼是 A2A Net？ A2A Net is an AI agent platform for B2B SaaS product and engineering teams. It creates customer-facing agents from an existing OpenAPI specification or MCP server, allowing customers to perform actions, analyze data, and complete complex multi-step workflows in the SaaS product. The workflow covers agent creation, conversational testing, rubric-based session evaluation, and automated optimization. Teams can test prompts and tool calls in a staging environment before publishing an agent to a website, app, Slack, Teams/Copilot, or Gemini Enterprise. The platform also supports variables and secrets for customer-specific authentication data, file exchange, and protocol compatibility with AG-UI and A2A. ## A2A Net 能做什麼？ ### OpenAPI and MCP-based agent creation Create an agent by providing an OpenAPI specification or MCP server URL, with support for custom headers. This connects the agent to the product capabilities already exposed by the SaaS team. ### Conversation testing and rubric scoring Use the playground to test the agent with real conversations, define evaluation rubrics, and rate sessions against criteria such as task completion, communication quality, and correct tool or parameter use. ### Automated prompt optimization Use session ratings and an LLM-based judge to optimize the agent’s system instruction and prompt behavior for conversational workflows. ### Staging and production environments Test prompts and tool calls in staging without affecting production data, then publish the configured agent when it is ready. ### Variables, secrets, and code mode Add variables and secrets to instructions and tools, send values as API request headers or set them during customer authentication, and use a customizable Python sandbox to write code, call tools, and create files. ### Multi-channel publishing and file support Publish agents to a website, app, Slack, Teams/Copilot, and Gemini Enterprise. Agents can exchange images, PDFs, and spreadsheets, and can generate channel-native elements such as buttons, images, and tables. ## 使用情境 ### Add an agent to an existing SaaS product A product team with an OpenAPI specification or MCP server can create an agent around its current APIs instead of designing a separate action layer for customer workflows. ### Validate workflows before production release An engineering or product team can hold conversations with the agent, identify failures, define rubrics, and test tool calls in staging before the agent reaches live product data. ### Let customers complete multi-step product tasks A SaaS provider can expose actions and data analysis through a conversational agent so customers can complete complex workflows in the product rather than handling each step separately. ### Extend product access into collaboration tools Once the agent is ready, a provider can publish it beyond its website or app to Slack, Teams/Copilot, or Gemini Enterprise, giving customers additional interaction channels. ### Handle files and structured responses in agent workflows Agents can receive or send images, PDFs, and spreadsheets, and can present supported results through interface elements such as buttons, images, and tables. ## 常見問題 ### What does A2A Net use as the starting point for an agent? You can start with an OpenAPI specification or an MCP server URL and add custom headers. The agent is then connected to the capabilities exposed by those interfaces. ### How are agents tested and improved? You can chat with an agent in the playground, define rubrics, rate sessions, and use those ratings with an LLM-based judge and optimization workflow to improve the agent. Staging is available for testing prompts and tool calls without affecting production data. ### Where can an agent be published? The site lists publishing to a website, app, Slack, Teams/Copilot, and Gemini Enterprise. It also states that agents are built to meet the requirements for publication to Slack, Microsoft, and Google marketplaces. ### How is A2A Net different from customer-support agent platforms? The site distinguishes A2A Net from platforms such as Intercom, Sierra, and Decagon by focusing on agents for B2B SaaS products. Its stated purpose is to let a product’s customers use the product’s APIs to take actions, analyze data, and complete multi-step workflows. ### Does the platform support files and customer-specific request data? Yes. Agents can send and receive images, PDFs, and spreadsheets. Variables and secrets can be used in instructions and tools, sent as API request headers, or set per customer during authentication. ## 快速資訊 - **Category**: AI agent platform for B2B SaaS - **Primary users**: B2B SaaS product and engineering teams - **Agent inputs**: OpenAPI specification or MCP server - **Initial testing surface**: A2A Net playground with conversational testing - **Publishing destinations**: Website, app, Slack, Teams/Copilot, and Gemini Enterprise - **Protocol compatibility**: AG-UI and A2A