Agentic

Agentic exposes tools through SDK-specific adapters (@agentic/ai-sdk for Vercel AI SDK, @agentic/platform-tool-client for direct consumption) that normalize tool schemas across different LLM frameworks. Each adapter translates Agentic's tool definitions into the native tool-calling format expected by the target SDK (OpenAI function calling, Vercel AI tool format, etc.), enabling developers to use the same Agentic tools across Vercel AI SDK, OpenAI, LangChain, LlamaIndex, Mastra, and Firebase GenKit without rewriting tool integration code.

Agentic operates a curated marketplace of LLM tools (e.g., @agentic/search for web search) hosted on Agentic's cloud infrastructure (Cloudflare Workers for MCP gateway, Node.js backend on Vercel). Tools are consumed via HTTP APIs or MCP protocol, with usage tracked and billed via Stripe on a per-tool, pay-as-you-go basis. Developers load tools by identifier (e.g., 'AgenticToolClient.fromIdentifier(@agentic/search)') and invoke them through their LLM SDK's tool-calling mechanism; Agentic handles execution, caching, rate-limiting, and billing transparently.

Agentic enforces tool schema validation using JSON Schema or OpenAPI specifications, ensuring that tool parameters and responses conform to defined types. SDK adapters (e.g., @agentic/ai-sdk) provide TypeScript type definitions generated from tool schemas, enabling compile-time type checking and IDE autocomplete. When tools are invoked, Agentic validates parameters against the schema and returns type-safe results, reducing runtime errors and improving developer experience.

Agentic tools are designed to compose seamlessly within LLM SDK tool-calling workflows, enabling developers to chain multiple tools together in a single agent loop. The LLM SDK (Vercel AI, OpenAI, etc.) orchestrates tool calls based on the model's reasoning, and Agentic tools integrate transparently into this workflow. Developers can combine Agentic tools with custom tools and SDK-native tools without special composition logic — the LLM SDK handles orchestration.

Agentic exposes all marketplace tools as MCP servers accessible through a Cloudflare Workers-based gateway, enabling any MCP-compatible client (Claude Desktop, custom MCP consumers) to invoke Agentic tools without SDK integration. The MCP gateway runs on Cloudflare's global edge network, providing low-latency access to tools and handling protocol translation, authentication, and request routing. Developers can consume Agentic tools via standard MCP client libraries by connecting to the Agentic MCP gateway endpoint.

All Agentic tools are accessible via HTTP POST requests to Agentic's REST API, enabling developers to invoke tools directly without SDK integration or MCP protocol overhead. Each tool exposes a documented HTTP endpoint accepting JSON parameters and returning JSON results. This fallback mechanism allows developers to use Agentic tools from any programming language or environment (Python, Go, Rust, etc.) by making standard HTTP requests, bypassing the need for TypeScript SDK adapters.

Agentic provides a built-in web search tool (@agentic/search) that integrates with major search APIs and implements production-grade caching (likely Redis-based) and customizable rate-limiting to optimize cost and performance. The tool accepts search queries as input and returns structured search results (title, URL, snippet, etc.). Caching reduces redundant API calls for identical queries, while rate-limiting prevents abuse and controls costs. Developers invoke the search tool through their LLM SDK's tool-calling mechanism, and Agentic handles the underlying search API orchestration transparently.

Agentic enables developers to publish custom tools (as MCP servers or OpenAPI services) to the Agentic marketplace and monetize them through usage-based pricing. Publishers define tool schemas, set pricing per invocation, and Agentic handles billing, payment processing (Stripe), and distribution. The platform manages tool versioning, SLAs, and monitoring. Developers can publish tools written in any language (as long as they expose MCP or OpenAPI interfaces) and earn revenue based on tool usage by other developers.

Agentic's marketplace provides semantic search and discovery capabilities enabling developers to find tools by natural language description or capability (e.g., 'I need a tool that extracts data from PDFs'). The discovery system likely uses embeddings-based search to match developer queries against tool descriptions and capabilities, surfacing relevant tools without requiring exact tool name knowledge. This reduces friction in tool selection and helps developers discover tools they didn't know existed.

Agentic supports semantic versioning for published tools, enabling publishers to release updates while maintaining backward compatibility with existing consumers. The platform manages tool version resolution, allowing developers to pin specific tool versions or use version ranges (e.g., '@agentic/search@1.x'). Breaking changes are handled through explicit version bumps, and Agentic provides tooling for publishers to deprecate old versions and migrate users to new ones.

Agentic claims to provide production-grade tool execution with SLAs (Service Level Agreements) and monitoring, implying uptime guarantees, latency targets, and observability. Tools are executed on Agentic's managed infrastructure (Vercel, Cloudflare Workers) with built-in error handling, retry logic, and monitoring. Developers can rely on Agentic tools for mission-critical applications without building their own reliability infrastructure.

Agentic is fully open-source on GitHub, with a TypeScript codebase that developers can inspect, fork, and contribute to. The SDK includes core client libraries (@agentic/platform-tool-client), SDK adapters (@agentic/ai-sdk), and example tools (@agentic/search). The open-source model enables community contributions, security audits, and transparency into Agentic's implementation. Developers can self-host Agentic components or contribute improvements back to the project.