Cheap AI agent hosting
Cheap AI agent hosting can run your agent application in a small container while a separate provider serves the language model. JustRunMy.App bills allocated CPU, RAM, and disk per minute. Model API usage, external databases, and other services are separate costs.
A good fit for: API-based assistants, Telegram or Discord AI bots, tool-using workers, and lightweight Python or Node.js automations.
Agent hosting and model inference have different costs
An agent typically combines application code, a language model, and one or more tools. If your code calls a remote model API, its container runs the orchestration and handles responses without loading the model weights locally. The model provider bills that API usage separately.
The estimates on this page cover CPU, RAM, and disk for the application. They do not establish GPU availability or capacity for running large language models locally. Evaluate model serving as a separate workload with its own hardware requirements.
Control the work your agent can create
Set a maximum number of tool calls, cap concurrent jobs, and put timeouts around external requests. A retry loop can increase token spend even when the hosting allocation stays small. Give each job a clear completion condition and log enough information to diagnose failures.
Keep provider keys in environment variables. Use durable storage for checkpoints and task status if work must continue after a restart. Save only the conversation or document data your application needs, and keep growing caches within your disk allocation.
Run a worker, an API, or a chat integration
A background agent can consume work from your own queue or poll an external service. An API-based agent can expose an HTTP port with automatic HTTPS. A chat agent can connect the same logic to your Telegram or Discord bot project.
Deploy your project with Zip, Git, or Docker and set its startup command. Python and Node.js dependencies must fit the chosen runtime. Framework-based projects such as LangChain or LangGraph need the same checks: compatible dependencies, enough memory, and explicit handling of saved state.
Deployment checklist
Identify the agent process
Choose whether the container runs a background worker, a chat bot, or an HTTP application. Keep model inference requirements separate.
Deploy code and dependencies
Upload your project or deploy its container image. Start with resources informed by a local test.
Add keys and durable state
Configure model and tool credentials as environment variables. Attach storage or connect your database for checkpoints.
Run a bounded task
Verify a complete task in logs, check failure handling, and set concurrency and retry limits before increasing traffic.
Example resource allocations
Monthly USD rates: $1.90 per vCPU, $1.95 per GB RAM, and $0.25 per GB disk. All allocated resources are billed per minute across applications, including idle allocations.
Size for dependencies, concurrent tasks, and local state. Browser automation, document indexing, and loading model weights can need substantially more memory than a simple API client.
| Example | vCPU | RAM (GB) | Disk | Monthly estimate |
|---|---|---|---|---|
| Minimal allocation example | 0.15 | 0.15 | 0 GB | $0.58 |
| Lightweight worker example | 0.25 | 0.25 | 0.3 GB | $1.04 |
| More memory for dependencies | 0.25 | 0.5 | 1 GB | $1.70 |
| Larger worker example | 0.5 | 1 | 1 GB | $3.15 |