Give an AI agent its own container for up to 24 hours, where it can write files and install packages while Runtime streams the output to your app. Through the Conjoin SDK in that container, the agent can call your databases, storage, and messaging with your project credentials.
Runtime keeps the container open until the time limit that you set, so an agent can clone a repository, install its dependencies, and build it across as many commands as the agent needs. Because Runtime sets your project credentials in the container, the agent can call the same Database and Storage resources as your production backend.
A session lasts between 5 minutes and 24 hours, depending on the limit that you set. Every command in the session operates on the files and packages that earlier commands produce.
Runtime provides a writable workspace directory in each session, where the agent can create files, clone repositories, and execute build tools. Runtime preserves the contents of that directory until the session ends.
Runtime streams stdout and stderr over a WebSocket while each command executes, so your interface can show the agent's terminal output line by line.
Every session starts with the Conjoin SDK configured with your project credentials. An agent imports the SDK like any other package and can call Database, Storage, and the other Conjoin services from its first command.
When a session reaches its time limit, Runtime stops the container and removes it along with its workspace. If the agent finishes sooner, end the session through the API.
Runtime places each session in its own container and blocks network traffic between that container and every other session or workload. Runtime also caps the container's CPU and memory, so a runaway process slows only its own session.
Runtime places each session in its own container and blocks network traffic between that container and every other session or workload. Runtime also caps the container's CPU and memory, so a runaway process slows only its own session.
Configurable from 5 minutes to 24 hours
From the API call to a container that accepts commands
stdout and stderr over a WebSocket as each command executes
Runtime preserves the session's filesystem and installed packages for the whole task. The agent can therefore execute a series of commands in which each one builds on the output of the last.
The agent can edit and retest its code in one container, so your application has no state to save and restore between steps.
In a session with the filesystem enabled, the assistant can clone a repository, install its dependencies, and execute the code against your Database and Storage through the pre-installed SDK.
The assistant tests its changes against the data in your Conjoin project with credentials that Runtime sets in the session, so you have no secrets to configure for the assistant.
Start a Python session, install pandas and matplotlib, execute the analysis scripts, and download the charts that they produce. Because every step executes in the same container, each script can open the files that the previous script writes.
Runtime keeps the packages and intermediate files from the first step to the last, so you can skip uploading intermediate files from one execution to the next.
Each session's container includes a writable filesystem, package managers, and build tools. Execute npm install, pip install, or cargo build, and then the test suite, in that same container.
A full build and test cycle executes in one isolated container, so you have no CI runner of your own to maintain for the job.
Conjoin will open with a free Build tier, while the pricing page already lists the launch prices for Pro and Business. If you tell us what you're building, we'll talk you through early access.