Your coding agent owns the framework, prompts, tools, and application logic. The Anyhost skill gives it a safe way to deploy that existing HTTP service with runtime, model budget, data, secrets, domains, logs, and health in one project.
For existing HTTP/API agent services · GitHub repository required · no new Anyhost agent involved
The Anyhost skill does not replace the agent framework or inspect prompts in the control plane. It gives the coding agent working in your repository a scoped interface for deployment and managed services.
Your coding agent requests the services; Anyhost provisions, scopes and injects them into the selected environment.
Run the existing agent API as a container with a stable public URL and health-aware deployment state.
An OpenAI-compatible runtime key, environment injection and an independent optional budget without committing a provider key.
Provision state and cache per environment. Generated URLs appear in new deployments after resources are ready.
S3-compatible storage scoped to the project environment, with bucket, prefix and region generated for the runtime.
Dev and prod keep separate values and resources. Secret contents stay masked; generated names cannot be overwritten casually.
Give the coding agent current operational evidence without giving it raw AWS, database or DNS-provider power.
A successful build is not enough. The result should identify the commit, runtime, attached resources, public URL and health evidence so the next agent conversation starts from state instead of guesses.
The agent fetches the official setup prompt, installs the CLI and loads the project-local cloud skill.
It inspects the start command, health path, environment variables, model access and data dependencies.
Anyhost provisions requested resources and deploys the pushed commit; the agent verifies one real API request.
The agent shows the dev evidence and asks before any production resource or deployment mutation.
The prompt keeps application decisions with the agent and cloud mutations with Anyhost.
Fetch https://anyhost.app/agent-setup/prompt.md and follow it to set up Anyhost in this repository. Then deploy this existing AI agent service to the Anyhost dev environment. Inspect the repository first: identify its real start command, supported health or smoke path, required environment variables, model access, and any Postgres, Redis, storage, or email dependency. Keep all application and agent-logic changes in this repository. Use Anyhost only for scoped cloud resources, runtime configuration, deployment, status, logs, health, and domains. Deploy the pushed commit to dev, verify one real API request and the required resource-backed behavior, and show me the deployment receipt. Ask for my explicit confirmation before creating or changing anything in production.
No. The framework, prompts, tools, model behavior and application code stay in your repository. Your coding agent may change them when you ask; Anyhost provides cloud operations and managed resources.
This page is scoped to an existing agent application that runs as an HTTP or API service from a GitHub repository. Validate background-worker, scheduled-job, GPU or multi-service requirements before relying on them.
No. An optional Anyhost Model API resource injects an OpenAI-compatible runtime key and base URL into the deployment. The application can also use its own secret variables when that is the intended architecture.
The Model API design does not persist prompt content. Your own application may store conversations or traces, so review the repository's behavior separately.
No. Production deployments, production resources, destructive operations and sensitive reveals require explicit confirmation immediately before the operation.
Start in dev, verify one real request, and keep production behind an explicit confirmation boundary.