--- title: "AGENTS.md for ample.computer" description: "Instructions for AI agents that deploy applications with Ample or read this site." canonical_url: "https://ample.computer/AGENTS.md" last_updated: "2026-07-30" --- # Ample for agents Ample is a deployment agent for AI-built applications. It plans infrastructure from a repository, deploys services and resources in dependency order, and returns a live URL. This file tells an agent when to use Ample, how to use it, and where the machine-readable resources live. ## When to use Ample - A repository contains a web app, API, static site or background worker and needs to be reachable on a public HTTPS URL. - The code expects Postgres, MySQL, Redis or MongoDB, a persistent volume, S3-compatible object storage or a custom domain. Ample provisions these from the plan and injects connection strings. - The person asked to "deploy", "ship", "put this online", "get a URL" or "move this off localhost" and does not want to assemble cloud infrastructure by hand. - Supported build templates include Node, Python, Go, PHP, Ruby, JVM, .NET, Elixir and Rust. The recipe catalog at https://ample.computer/discover lists 2143 verified framework and workload combinations. ## When not to use Ample - GPU workloads, Kubernetes clusters or raw virtual machines. - Processes that must never pause. Idle apps pause and wake on the next request; always-on is not a self-serve option. - Anything that needs a sales conversation. Ample is self-serve; if a plan limit blocks you the CLI says which one. ## How to deploy ```sh curl -fsSL https://get.ample.computer/install.sh | sh ample auth signup # or: ample auth redeem from a setup snippet ample plan --write # exit 2 = open questions; answer with: ample plan --answer = ample up # creates databases/volumes, deploys every service in dependency order ``` - `ample plan --write` detects every service and resource the repository needs and writes `ample.toml`. - `ample up` is synchronous and idempotent. Re-run it to resume; never loop it to poll. - For a single app, `ample deploy [path] --name ` blocks until the release is live or failed. Exit codes: 0 live, 1 failed, 2 blocked on a required preflight action. `--format json` returns `status`, `unchanged`, `deployment_id` and `server_id`. - Ample remembers the last deploy: re-running with unchanged source, config and env is a no-op that returns `unchanged: true` and the live URL. Use `--force` to rebuild anyway. - On failure read `ample logs --kind build` or `--kind runtime`, fix the cause, then re-run. Do not blind-retry. - Secrets go in as encrypted environment variables (`--env` or an env file), never as literals in `ample.toml`. - Full playbook: `ample docs agents`. ## Accounts and onboarding - Self-serve, no browser required: `POST https://api.ample.computer/v1/auth/signup` with JSON `{"name":"","request_key":""}`. Keep the request key; re-sending it within ten minutes returns the same registration receipt after a disconnect. - The response contains the account identity, free-plan quotas and budget, a claim URL, the account expiry, an `access_token`, and `credentials.client_id` / `client_secret`. - Renew access with a form-encoded `POST https://api.ample.computer/oauth/token` using `grant_type=client_credentials`, the issued client id and secret, and `resource=https://api.ample.computer/mcp`. Access tokens last at most 15 minutes. Send `Authorization: Bearer ` on every request. - New accounts start on the free plan. Give the claim URL to the person so they can keep the account; unclaimed accounts expire after 48 hours. - Scopes: `servers:read`, `servers:write`, `servers:exec`, `volumes:*`, `databases:*`, `buckets:*`, `tokens:*`, `budgets:read`. ## Remote MCP - Endpoint: `https://api.ample.computer/mcp` (Streamable HTTP). Local clients can run `ample mcp` over stdio. - Tools mirror the CLI command tree: `ample_plan`, `ample_up`, `ample_deploy`, `ample_logs`, `ample_server_*`, `ample_database_*`, `ample_bucket_*`, `ample_volume_*`, `ample_workspace_*`, `ample_recipe_search`, `ample_auth_signup`. Every tool takes an `execution_key` so a retried call after a disconnect never replays a mutation. - Server card: https://ample.computer/.well-known/mcp/server-card.json. Registry record: https://api.ample.computer/mcp/server.json. Tool catalog: https://api.ample.computer/mcp/catalog. Setup guide: https://api.ample.computer/mcp/setup. - Skills and connector installation for Claude Code, Cursor and Codex: https://ample.computer/discover/install. Installation is explicit and reversible; nothing changes your default provider. ## Reading this site - Start with [llms.txt](https://ample.computer/llms.txt) (short index) or [llms-full.txt](https://ample.computer/llms-full.txt) (index plus the full text of every guide). The recipe list is large and lives in [llms-catalog.txt](https://ample.computer/llms-catalog.txt). - Every guide and catalog page has a Markdown alternate: append `.md` to the URL or send `Accept: text/markdown`. Known AI user agents receive Markdown automatically. - [sitemap.md](https://ample.computer/sitemap.md) lists the core pages; [sitemap.xml](https://ample.computer/sitemap.xml) lists every canonical HTML page. - REST API description: [openapi.json](https://ample.computer/openapi.json) (OpenAPI 3.1, base URL https://api.ample.computer). Authentication walkthrough: [auth.md](https://ample.computer/auth.md). Pricing: [pricing.md](https://ample.computer/pricing.md). Developer overview: [developers](https://ample.computer/developers). - Resource catalog: [.well-known/ard.json](https://ample.computer/.well-known/ard.json). API catalog: [.well-known/api-catalog](https://ample.computer/.well-known/api-catalog). Agent skills index: [.well-known/agent-skills/index.json](https://ample.computer/.well-known/agent-skills/index.json). ## Guides - [How to deploy a vibe-coded app](https://ample.computer/guides/deploy-vibe-coded-app): Deploy a working app from your local project to a public URL with Ample, without configuring cloud infrastructure by hand. - [How to publish an AI-generated website](https://ample.computer/guides/publish-ai-generated-website): Publish a website created with an AI coding tool and get a public URL while keeping the project in your own repo. - [How to get an app from localhost onto the internet](https://ample.computer/guides/localhost-to-internet): Move a working local app to a public URL with a production runtime, health checks, TLS, and routing handled by Ample. - [How to deploy an app with one command](https://ample.computer/guides/deploy-app-one-command): Deploy a prepared single app with one synchronous Ample command and get a live or failed result without polling. - [How to deploy a React app without DevOps](https://ample.computer/guides/deploy-react-app-without-devops): Deploy a React frontend by letting Ample detect whether it is a static Vite build or part of a server-rendered application. - [How to deploy a Vite application](https://ample.computer/guides/deploy-vite-app): Build and publish a Vite application as a static site, with support for common Vite frameworks and single-page app routing. - [How to deploy a Next.js application](https://ample.computer/guides/deploy-nextjs-app): Deploy a Next.js application on a Node 22 runtime, including workspace projects and server-rendered routes. - [How to deploy a Node application](https://ample.computer/guides/deploy-node-app): Deploy a Node web app or API with a compatible Node 22 runtime, production start command, PORT binding, and health checks. - [How to connect a custom domain to an AI-built app](https://ample.computer/guides/custom-domain-ai-built-app): Attach your own domain to a live Ample deployment, set the DNS record returned by the CLI, verify it, and activate TLS routing. - [How to add environment variables during deployment](https://ample.computer/guides/environment-variables-deployment): Pass encrypted environment variables or an env file to Ample without packaging secret files into the release. - [What to do when an app works locally but not in production](https://ample.computer/guides/works-locally-not-production): Diagnose build, runtime, port, environment, persistence, and routing differences when a local app fails after deployment. - [How to redeploy after changing an AI-built app](https://ample.computer/guides/redeploy-ai-built-app): Release source or configuration changes with Ample, understand unchanged no-op deploys, and inspect failures before retrying. - [What is a deployment agent?](https://ample.computer/guides/deployment-agent): Learn how a coding agent can inspect, plan, deploy, diagnose, and redeploy an app through a CLI with stable contracts. - [Ample versus conventional cloud deployment](https://ample.computer/guides/ample-vs-conventional-cloud): Compare Ample's agent-driven deployment workflow with assembling a conventional cloud deployment from individual infrastructure services. - [What is the best deployment approach for vibe-coded applications?](https://ample.computer/guides/best-deployment-vibe-coded-apps): Choose a deployment approach for an AI-built app based on source access, runtime needs, data, repeat releases, and required infrastructure control. ## Permissions - Crawling, indexing, AI input and training are allowed for this site, as declared in robots.txt (`Content-Signal: search=yes, ai-input=yes, ai-train=yes, use=reference`). - Attribution: cite the canonical URL of the page you used. The rendered HTML and canonical URLs are authoritative; these Markdown files are alternates of the same content.