Persistent live state
Retain files, memory, and processes.
Keep the sandbox ready while the agent waits for a person, API, schedule, or next task.
- Filesystem state retained
- Memory state retained
- Running processes restored
Persistent agent sandbox · AI code execution
Give your agent a sandbox it can return to.
Keep the repo, dependencies, running processes, and in-memory state between runs. Resume in about 25ms instead of rebuilding.
Up to $200 in credits · No credit card required
Teams building with Blaxel
Persistent vs. ephemeral agent sandboxes
Resume the workspace, not just the files.
A coding agent needs its repo, dependencies, services, and test process ready when it returns.
Workload-fit comparison
Agent lifecycle requirementBlaxel persistent sandboxTypical ephemeral sandbox
What survives the pause
Blaxel persistent sandbox
Files, memory, and processes
Snapshot the filesystem, memory, and running processes.
Typical ephemeral sandbox
Files or an image may remain
Restore files or an image, then recreate the live runtime.
When the agent returns
Blaxel persistent sandbox
Resume the same environment
Reconnect in about 25ms with dependencies and services ready.
Typical ephemeral sandbox
Build a new runtime session
Provision compute, reinstall what is missing, and restart services.
While the agent waits
Blaxel persistent sandbox
$0 active compute on standby
Active compute billing stops. Snapshot and volume storage continue.
Typical ephemeral sandbox
Keep compute active or rebuild later
Keep the runtime running or reconstruct it for the next action.
Execution boundary
Blaxel persistent sandbox
One microVM per sandbox
Give each agent, application, or job its own microVM.
Typical ephemeral sandbox
Depends on the runtime
Isolation varies from shared processes to dedicated virtual machines.
Human and operator access
Blaxel persistent sandbox
Inspect the live work
Stream output, watch events, and open a private preview for review.
Typical ephemeral sandbox
Often session-specific
Re-establish logs, previews, and inspection after the session ends.
Let the agent continue in the same workspace.
Up to $200 in credits · No credit card required
The full working state
Keep the environment your agent already prepared.
Repos, dependencies, test watchers, and previews stay ready between actions.
About 25ms to resume
Return to the same environment.
Resume in about 25ms with dependencies loaded and services running.
- Same sandbox environment
- No clone or reinstall cycle
- No service restart sequence
Preview network controls
Control outbound access.
Use public-preview domain filters and proxy secret injection. Private-preview egress gateways support IP allowlisting.
- Public-preview domain filters
- Public-preview proxy secret injection
- Private-preview dedicated egress
Runtime visibility
See what the agent is doing.
Stream output, watch filesystem and process events, and open a private preview.
- Real-time output streaming
- Process and filesystem events
- Controlled preview URLs
Keep the agent framework
Use the agent stack you already have.
Connect through an SDK, REST API, or MCP. Blaxel handles the compute, storage, and networking.
First-class sandbox provider
OpenAI Agents SDK
Run OpenAI agent logic with persistent code execution, file transfer, streaming, and previews.
MCP and hosted execution
Claude ecosystem
Run Claude Code or Claude Agent SDK in a workspace that persists between interactions.
Connect your agent to a persistent workspace.
Up to $200 in credits · No credit card required
Persistent agent sandboxes
What builders ask before keeping agent state.
What is a persistent agent sandbox?
It is an isolated environment that keeps the filesystem, memory, and running processes between runs. The agent returns to the same live workspace instead of rebuilding.
What should persist between AI agent actions?
Keep the working tree, installed dependencies, in-memory state, and long-running processes such as development servers or test watchers.
How is a persistent sandbox different from persistent storage?
Persistent storage keeps data after compute ends. A persistent sandbox keeps the live runtime ready. Use a Volume when data must survive sandbox deletion.
Does a Blaxel sandbox use active compute while the agent waits?
Active compute billing stops while the sandbox is in standby. Snapshot storage that preserves the working state and any attached volume storage remain billable.
How quickly does a persistent Blaxel Sandbox resume?
About 25ms from standby, with the filesystem, memory, and running processes restored.
Which AI agents can use a persistent Blaxel Sandbox?
Any agent that can call the Blaxel SDK, REST API, or MCP tools. Common fits include coding, app-generation, code-review, QA, DevOps, and data or code execution agents.
Keep the work ready
Start with the workload your agent already runs.
Clone the repo, start the services, move the sandbox to standby, and reconnect. Verify that the agent returns to the same workspace.