Click

Product Introduction

1. Definition: Click is a managed agents API platform, a technical category known as Agent-as-a-Service or AI orchestration infrastructure. It provides a fully managed runtime for autonomous AI agents, complete with a sandboxed execution environment, a headless browser, and web search capabilities, all accessible via a single REST API.
2. Core Value Proposition: Click exists to eliminate the infrastructure complexity of building and scaling production-grade AI agents. Its primary value proposition is offering a serverless agents API that allows developers to integrate sophisticated, tool-using AI workflows into their applications without managing servers, state, or execution environments. The core keywords are managed agents API, agent orchestration, and serverless AI.

Main Features

1. Managed Agent Sessions: Click provides durable, stateful sessions that persist conversation history, tool outputs, and execution context. Each session is provisioned with an isolated sandbox, a browser instance, and configured tools. This allows for long-running, multi-step agent tasks without developer overhead.
2. Deferred Tool Discovery & Code Mode: The platform's agent harness supports advanced capabilities like deferred tool discovery, enabling agents to dynamically learn about and use thousands of available tools. Its "Code Mode" allows agents to execute scripts (e.g., bash, Python) within the secure sandbox to manipulate data or chain operations, moving beyond simple API calls.
3. Streaming Event API & Webhooks: Results are delivered in real-time via Server-Sent Events (SSE), streaming back message.delta (token-by-token AI output), tool.started/completed events, and final status. For asynchronous workflows, Click can send webhooks (e.g., submission.updated, artifact.created) to notify your application when an agent's task is complete.
4. Bring-Your-Own-Model (BYOM) & Tools: Click is model-agnostic. Developers can use any LLM provider (OpenAI, Anthropic, Google, etc.) with their own API keys. It also supports bringing custom tools via Model Context Protocol (MCP) servers or standard HTTP endpoints, with granular session-level permissions and credential management.

Problems Solved

1. Pain Point: The immense engineering burden of building, securing, and maintaining the backend infrastructure for reliable AI agents—including sandboxing, browser automation, tool integration, state management, and scalable execution.
2. Target Audience: The primary personas are Software Engineers and DevOps teams building AI-powered features (e.g., support bots, research assistants, automated workflows); Product Managers needing to prototype and ship agentic features quickly; and Startups lacking the resources to build complex agent infrastructure from scratch.
3. Use Cases: Essential scenarios include: automating customer support ticket investigation and draft replies; conducting multi-source web research and compiling reports; performing automated QA testing by interacting with a web application; and executing data analysis scripts on uploaded files within a secure environment.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros: Drastically reduces time-to-market for agent features; abstracts away critical but non-differentiating complexity (sandboxing, browser mgmt.); offers excellent flexibility with BYO model and tools; real-time streaming is ideal for interactive applications.
   • Cons: As a managed service, it introduces vendor lock-in for the agent runtime layer. Deep, low-level customization of the agent's core reasoning loop or memory architecture may be limited compared to self-hosted frameworks. Cost is based on usage, which for high-volume tasks could become a significant operational expense.

2. Key Alternatives & Differentiation:

   • vs. Self-hosted Frameworks (LangChain, LlamaIndex): Click differentiates by providing the entire production-ready runtime. While LangChain is a library for building agents, you must still host, scale, and secure the execution environment. Click offers this as a managed API, trading customization for operational simplicity.
   • vs. Cloud AI Services (Azure AI Agents, Google Vertex AI): Click is provider-agnostic and tooling-focused. Cloud services often lock you into their model ecosystems and have less emphasis on BYO tools or seamless integration with custom MCP servers. Click acts as a neutral orchestration layer.
   • vs. Emerging Competitors (CrewAI, SmythOS): Click's differentiation lies in its specific technical implementation, like deferred tool discovery for massive toolkits and its integrated "Code Mode" sandbox, positioning it for more complex, programmatic agent tasks compared to simpler workflow orchestrators.

Frequently Asked Questions (FAQ)

1. What is a managed agents API? A managed agents API is a cloud service that provides a fully hosted runtime for autonomous AI agents, handling execution, state, tools, and security (like sandboxing) through simple API calls, removing the need to manage underlying infrastructure.
2. How does Click handle security and sandboxing for AI agents? Click runs each agent session in an isolated, ephemeral OS-level sandbox. This containerized environment securely contains all code execution, file system access, and browser activity, preventing agents from affecting the host system or other sessions.
3. Can I use my own OpenAI or Anthropic API key with Click? Yes, Click operates on a Bring-Your-Own-Keys (BYOK) model. You configure your session to use your own credentials with supported LLM providers like OpenAI, Anthropic, Google Gemini, and others, giving you full control over model costs and usage.
4. What is the difference between Click's streaming and webhook features? The streaming API (Server-Sent Events) is for real-time, interactive applications where you want to display agent thoughts, actions, and outputs as they happen. Webhooks are for asynchronous, background tasks where you only need a notification when the agent's work is fully complete.
5. What kinds of custom tools can I connect to Click agents? You can connect any tool accessible via the Model Context Protocol (MCP) or a standard HTTP/HTTPS endpoint. This includes internal APIs, database interfaces, software development tools, or custom business logic you've exposed as a service.