Run a command in a sandbox you hold and read its output

Runs one command inside the caller's sandbox and answers its exit code, stdout and stderr.

POST /v1/sandbox/run

Sandbox → · All API references → · Six flows, four surfaces →


Address	`https://api.hanzo.ai/v1/sandbox/run`
Method	`POST`
Operation	`run_in_sandbox`
Auth	`Authorization: Bearer $HANZO_API_KEY`

Runs one command inside the caller's sandbox and answers its exit code, stdout and stderr. A non-zero exit is a successful call carrying a failed program, so it comes back as data and not as an error.

Name a session and the command NARRATES INTO IT: its output is appended to that session's live log as the program produces it, so anything watching the session — GET /v1/agents/sessions/stream, scoped to one run with ?root= — watches the work happen rather than waiting for the verdict. Without it the call is what it always was: silent until it returns, which for an agentic run is twenty-five minutes of blank screen.

The session is named; the TENANT is not. It is the org the caller already proved, so a session belonging to somebody else is absent from the org this call acts for and the append is refused there.

Request

8 fields, body application/json (required).

Field	In	Type	Required	Description
`argv`	body	string[]	—	Argv is the program and its arguments, already split — the form no shell can misread.
`blind`	body	string[]	—	Blind is the set of secrets this command must never publish. It exists because output is redacted where it is PRODUCED or not at all.
`command`	body	string	—	Command is a shell line, run by `sh -c`.
`dir`	body	string	—	Dir runs the command somewhere other than the sandbox's working directory, which Leased.Workdir names.
`id`	body	string	—	ID is the sandbox to run in, from an earlier lease.
`session`	body	string	—	Session is the live agent session this command narrates into: its output is appended there AS IT IS PRODUCED, so every surface watching that session watches…
`stdin`	body	string	—	Stdin is fed to the program on standard input.
`timeoutSec`	body	integer	—	TimeoutSec bounds this ONE command, so a wedged program holds the caller for its own timeout rather than for the whole lease.

Response

Status	Body	Meaning
`200`	Ran	ok

200 body — 3 fields.

Field	In	Type	Always	Description
`exitCode`	body	integer	—	ExitCode is the PROGRAM's own status — 0 succeeded, anything else is what it returned, and a Command runs under `sh -c` so its shell's conventions apply.
`stderr`	body	string	—	Stderr is standard error, kept apart from Stdout so a caller reading a program's OUTPUT is not reading its diagnostics as data. Same 1 MiB cap, same redaction.
`stdout`	body	string	—	Stdout is what the program wrote to standard output, collected whole rather than streamed — to watch it arrive instead, name a RunIn.Session and read that…

Failure carries the platform error shape — see Errors.

Examples

hanzo sandboxes run

import { Configuration, SandboxApi } from 'hanzoai';

const api = new SandboxApi(new Configuration({ accessToken: process.env.HANZO_API_KEY }));
const { data } = await api.runInSandbox({ argv: ["<argv>"], blind: ["<blind>"] });

from hanzoai.cloud import ApiClient, Configuration
from hanzoai.cloud.api import SandboxApi

client = ApiClient(Configuration(access_token=os.environ["HANZO_API_KEY"]))
result = SandboxApi(client).run_in_sandbox(argv=["<argv>"], blind=["<blind>"])

cfg := cloud.NewConfiguration()
cfg.AddDefaultHeader("Authorization", "Bearer "+os.Getenv("HANZO_API_KEY"))
client := cloud.NewAPIClient(cfg)

resp, _, err := client.SandboxAPI.RunInSandbox(context.Background()).Execute()
if err != nil {
	return err
}

use hanzo_cloud::apis::{configuration::Configuration, sandbox_api};

let mut cfg = Configuration::new();
cfg.bearer_access_token = std::env::var("HANZO_API_KEY").ok();

let result = sandbox_api::run_in_sandbox(&cfg, Default::default()).await?;

import ai.hanzo.cloud.ApiClient;
import ai.hanzo.cloud.api.SandboxApi;

ApiClient client = new ApiClient();
client.setRequestInterceptor(b -> b.header("Authorization", "Bearer " + System.getenv("HANZO_API_KEY")));

var result = new SandboxApi(client).runInSandbox();

curl -X POST https://api.hanzo.ai/v1/sandbox/run \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
       "argv": [
         "<argv>"
       ],
       "blind": [
         "<blind>"
       ]
     }'

Tool sandboxes, op run_in_sandbox — POST the JSON-RPC envelope to https://api.hanzo.ai/v1/mcp.

curl -X POST https://api.hanzo.ai/v1/mcp \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
       "jsonrpc": "2.0",
       "id": 1,
       "method": "tools/call",
       "params": {
         "name": "sandboxes",
         "arguments": {
           "op": "run_in_sandbox",
           "input": {
             "argv": [
               "<argv>"
             ],
             "blind": [
               "<blind>"
             ]
           }
         }
       }
     }'

Sandbox API · All Hanzo APIs · Interactive reference

Run a command in a sandbox you hold and read its output

Request

Response

Examples

On this page