AI-native development

Use Quake AI Developer as a first-class context source for the AI coding tools your team already runs, then run and operate what you build on Quake AI Compute you control. The always-on apps and agents you deploy are a control plane you own, from the solo builder running one agent to a team running many.

What this is for#

Engineering teams that build with AI coding assistants need documentation that those tools can fetch without copy-paste from rendered HTML. Quake AI Developer exposes plain markdown, standard llms.txt corpora, a hosted MCP server, and a knowledge graph so Cursor, Claude Desktop, VS Code Copilot, and custom MCP clients retrieve current answers. The outcome is less context-switching, tighter links from a question to a validated OpenTofu template or doc anchor, and IaC drafts you can run through tofu validate before apply.

A control plane you operate#

The same builder who drafts infrastructure with an AI assistant also runs the long-lived result on Quake AI: an API, a scheduled job, or an always-on agent on Compute you own. The durable runtime, your data, and your provider credentials stay on instances inside your project, and any heavy model inference is a call out to a model backend you choose.

OpenClaw, a one-click Console App, is a concrete example: a persistent AI agent on a Quake AI CPU VM that drives external model providers, or a local model through Ollama, with your keys and conversation data on the VM you own. For a model-routing tier that fans out to several backends, the inference gateway template builds the same pattern as a standalone service, and AI inference and RAG pipelines covers the full control-plane layout.

Reference architecture#

Download diagram: SVG, PNG, and PDF.

This pattern has two halves. The developer platform surfaces (llms.txt corpora, plain .md pages, the MCP server, and the knowledge graph) are the context the AI client retrieves; the deployable half is Quake AI, where the developer applies drafted IaC against the validated template library and a Development Environment bootstrap VM you extend with your toolchain.

1. AI coding client. Cursor, Claude Desktop, ChatGPT with a custom GPT, VS Code Copilot in agent mode, or a custom MCP client in CI connects to the developer platform. The client fetches https://docs.rumble.cloud/llms.txt, https://docs.rumble.cloud/llms-full.txt, and individual pages at https://docs.rumble.cloud/docs/{slug}.md when MCP is not available.

2. MCP tools. The hosted MCP endpoint at https://docs.quake.ai/api/mcp exposes callable tools for documentation search and OpenTofu and Heat template retrieval. The client invokes tools over standard MCP transport; no Quake AI account is required to install the server.

3. Knowledge graph and topical index. https://docs.rumble.cloud/knowledge-graph.json and https://docs.rumble.cloud/topical-index.json expose services, concepts, templates, and doc relationships. Clients that traverse typed edges retrieve related pages beyond a single keyword hit.

4. Infrastructure scaffolding. The developer applies generated or adapted OpenTofu, Heat, or console workflows on Quake AI. The OpenTofu template library is the deterministic target: sixteen validated templates cover compute, network, storage, Kubernetes, and full-stack patterns.

5. Optional inference workloads. Teams that also run models on Quake AI can follow AI inference and RAG pipelines for CPU-only self-hosted inference, Qdrant, and small LLMs via Ollama, separate from using the developer platform as a doc surface.

Services involved#

Get started#

• How to connect AI tools to Quake AI docs: MCP setup for Cursor, Claude Desktop, VS Code, Continue, and Zed, plus URLs for non-MCP tools.
• AI tools reference: MCP tool inputs, response shapes, and knowledge-graph.json schema.
• Self-hosted AI on cloud infrastructure: when to run models on your own VMs on Quake AI.
• Development Environment template: OpenTofu pattern for a bootstrap dev VM you can extend with your AI toolchain.
• Deploy the development environment template: step-by-step walkthrough for the bootstrap dev VM.
• Next.js app template and its deploy walkthrough: take an app you drafted with an AI coding tool from a local build to a TLS-served Next.js deployment on Quake AI.
• Inference gateway template and its deploy walkthrough: a model-routing endpoint for the agents and apps you run on Quake AI.
• Run a Cursor SDK agent in CI: invoke a Cursor SDK agent from a CI job on a self-hosted runner, for a PR-review or doc-generation task.
• OpenTofu template library: validated IaC starting points the MCP get_template tool returns.
• AI inference and RAG pipelines: deploy inference workloads on Quake AI infrastructure.

Estimate the cost#

Considerations and limits#

• Compliance posture. Quake AI holds SOC 2 Type I and Type II attestations. See Compliance and certifications for scope. This Solutions page describes developer tooling surfaces, not a regulated workload attestation.
• The developer platform surface evolves. Endpoints and MCP tools ship with the documentation set. AI clients should fetch the current llms.txt, MCP tool list, and per-page .md routes rather than caching a long-lived snapshot.
• MCP tool availability. Not every Quake AI API operation is exposed as an MCP tool. The MCP server is a curated read surface; mutating operations use the documented REST APIs and OpenStack-compatible clients.
• No platform-managed AI client. You bring Cursor, Claude, Copilot, or a custom MCP client. The platform supplies context and doc-backed tools; it does not ship an IDE or chat product.
• AI client output is not authoritative. Run tofu validate and compare generated IaC to canonical templates before apply. MCP tools return doc-backed artifacts where possible; synthesis on top of those artifacts is your responsibility.
• When this pattern is not the right fit.
  • Teams that standardize on manual docs and consoles only, with no AI-assisted authoring.
  • Teams whose security posture blocks outbound HTTPS from AI clients to docs.rumble.cloud.
  • Teams whose tooling cannot use MCP, llms.txt, or plain-markdown fetches and will not maintain a custom integration.