Integrations · Developer tools
E2B code execution for VegaDūta agents
VegaDūta integrates E2B to let agents execute code safely: when an agent needs to run code, it runs inside a sandboxed E2B cloud environment, completely outside the VegaDūta platform itself. The capability is switched on per agent, and the E2B key is platform-managed — you do not need your own E2B account to use it. The result is an agent that can compute, transform data, and verify its own work without any code ever touching platform infrastructure.
Why sandboxed execution matters
Letting a language model run code is powerful and dangerous in equal measure. E2B's cloud sandboxes give each execution an isolated environment away from the platform and away from your data stores — if generated code misbehaves, it misbehaves inside a disposable sandbox.
- Execution happens in E2B's cloud sandboxes, outside the platform.
- Per-agent enable toggle — only agents you choose can run code.
- Platform-managed E2B key — no separate account or key handling.
What agents use it for
Anything where reasoning alone is not enough: calculations that must be exact, data transformations, generating and checking output programmatically. The agent writes the code, the sandbox runs it, and the result comes back into the conversation or workflow.
Control and governance
Code execution is off by default and enabled per agent, so the capability exists only where you have deliberately granted it. Like every tool call, executions run under the platform's guardrails and can sit behind human approval gates.
Frequently asked questions
Can a VegaDūta agent run code?
Yes, if you enable it for that agent. Code executes inside a sandboxed E2B cloud environment, isolated from the VegaDūta platform, and the result returns to the agent's conversation or workflow.
Do I need my own E2B account or API key?
No. The E2B key is platform-managed — you switch the capability on per agent in the console and the platform handles the E2B side.
Is agent code execution safe?
Execution is sandboxed in E2B's cloud, entirely outside the platform, so generated code never runs on VegaDūta infrastructure or near your tenant data. The capability is per-agent opt-in and subject to the platform's guardrails.
Which agents can execute code?
Only the ones you have toggled it on for. E2B execution is a per-agent setting, so you can give a data-analysis agent the capability while your customer-facing WhatsApp agent goes without.
See it working in two minutes
The sandbox provisions a real tenant — describe an agent in one sentence and test it, no account, no card. Or browse ~90 industry workflow recipes to see what teams build.
