# Noorle — The Agent Runtime > Noorle is a unified operating environment for AI agents. It connects AI agents to built-in capabilities, developer-built tools, and any external service through secure MCP gateways — so teams can build, deploy, and run agents in production without stitching together infrastructure. This is the expanded version of /llms.txt, with fuller descriptions of each area of the platform. Canonical documentation lives under /docs/ (hosted via Mintlify). All blog posts live under /blog/. ## What Noorle Is Noorle provides the runtime layer agents need to do real work: a place to run, a set of capabilities to act with, a way to connect to external systems, and the security and observability to do it safely. It is model-agnostic (works with Claude, GPT, and any MCP-compatible model), language-agnostic (tools compile to WebAssembly from many languages), and deployment-agnostic. The platform is organized around a few core ideas: - **MCP Gateways** — secure, unified endpoints implementing the Model Context Protocol that sit between agents and the capabilities they use. - **Capabilities** — the actions an agent can take, spanning built-in tools, developer-built plugins, and external connectors. - **Autonomous Agents** — the agent runtime itself: model routing, memory, generative UI, browser access, triggers, and multi-agent delegation. - **Workflows** — graph-based orchestration for multi-step, multi-agent processes with approval gates. - **Knowledge Engine** — hybrid search and retrieval over ingested documents (RAG). - **Unified Workspace** — session-scoped file storage with clear input/output scopes. ## Platform Pillars ### Autonomous Agents (/platform/autonomous-agents/) Production-ready agents with smart model routing, three-tier memory, Generative UI, browser access, automation triggers, and sub-agent delegation. The full agent runtime — no framework required. ### MCP Gateways (/platform/mcp-gateways/) Virtual MCP endpoints that unify built-in capabilities, plugins, and connectors behind a single secure interface, with authentication, authorization, and agent-to-agent (A2A) discovery. ### Capabilities (/platform/capabilities/) Three kinds of capabilities make up what an agent can do: - **Built-in capabilities** (/platform/builtin-capabilities/): Files, Web Search, HTTP Client, Knowledge Retrieval, Code Runner, Sandbox, Computer (agent-only), and Browser — available instantly. - **Plugins** (/platform/plugins/): custom tools built in any language that compiles to WebAssembly, described with WIT, deployed via the Noorle CLI, and auto-discovered. - **Connectors** (/platform/connectors/): three integration types — REST/OpenAPI import, the MCP Registry, and Custom MCP — with authentication and parameter mapping. ### Workflows (/platform/workflows/) Graph-based orchestration with agent nodes, branching, and human-in-the-loop approval gates for processes that need more than a single agent turn. ### Knowledge Engine (/platform/knowledge-engine/) Hybrid search combining vector and keyword retrieval over ingested documents, powering retrieval-augmented generation for agents. ### Unified Workspace (/platform/unified-workspace/) Session-scoped file storage with two scopes: /workspace/input/ (read-only) and /workspace/output/ (read-write), giving agents a clean filesystem for each task. ## Documentation - Overview: /docs/learn/introduction/ - Quick Start (5 minutes): /docs/use/five-minute-quickstart/ - Model Context Protocol: /docs/learn/concepts/model-context-protocol/ - Build Plugins: /docs/build/introduction/ - API Reference: /docs/reference/introduction/ - Use Cases: /docs/learn/use-cases/ - Security & Auth: /docs/learn/auth/overview/ - Team Management: /docs/use/platform/team-management/ ## Blog & Insights Authoritative, in-depth articles on AI agent infrastructure. ### Frameworks vs. Runtimes: The Next Shift in AI Agent Infrastructure (/blog/frameworks-vs-runtimes/) Frameworks taught us how to build agents. Runtimes are where agents actually run. The post argues that the next shift in agent infrastructure is from build-time frameworks to production runtimes, and that understanding this distinction is the key to shipping agents that work reliably in production. ### Compute Engines for AI Agents: Why One Size Doesn't Fit All (/blog/compute-engines-for-ai-agents/) When deploying AI agents that execute code, choosing the right compute engine becomes critical. The post examines the trade-offs between compute options and explains why successful agent platforms need multiple compute engines working in tandem rather than a single approach. ### The Perfect Fit: MCP + WebAssembly Components (/blog/the-perfect-fit-mcp-webassembly-components/) MCP wants modular, sandboxed, language-agnostic tools with clean contracts. WebAssembly Components and WASI Preview 2 deliver exactly that — turning fragmented toolchains into a unified, secure, polyglot platform for agent tools. ### What AI Agents Really Need: Fundamental Requirements for Effective Agent Systems (/blog/what-ai-agents-really-need/) The gap between AI agent promise and performance isn't about language model power — it's about understanding what individual agents actually need to work effectively: the right context, capabilities, memory, and environment. ## Pricing (/pricing/) Simple, transparent, usage-based pricing: - Gateway requests: $0.00005 each - Web search: $0.01 per query - Storage: $0.005 per GB/day - Many capabilities included at no extra cost (web fetch, HTTP requests) - Start free with $10 in credit, no credit card required ## Getting Started - Sign up free with $10 credit: /pricing/ - Platform tour: /platform/ - Contact for enterprise: /contact/ ## Key Benefits - **Zero Lock-in**: Use any AI model, any programming language, any deployment method. - **Instant Start**: No complicated setup — connect and start building. - **Enterprise Ready**: Security, compliance, and observability from day one. - **Future-Proof**: Built on open standards (MCP, WebAssembly/WASI) with a growing ecosystem. - **Scale Seamlessly**: From prototype to production without rewrites. Built for developers who want to ship AI applications that users love and enterprises trust.