Create agent
Says what to do on a sandbox you hold, and a model does it there while you watch: a shell on any sandbox, and on a desktop or a browser the screen too — it takes a screenshot, clicks where it sees, types, presses keys, opens addresses and runs commands.
POST /v1/sandbox/{id}/agent
| Address | https://api.hanzo.ai/v1/sandbox/{id}/agent |
| Method | POST |
| Operation | post_sandbox_by_id_agent |
| Auth | Authorization: Bearer $HANZO_API_KEY |
Says what to do on a sandbox you hold, and a model does it there while you watch: a shell on any sandbox, and on a desktop or a browser the screen too — it takes a screenshot, clicks where it sees, types, presses keys, opens addresses and runs commands. The answer is a text/event-stream, one event a step, ending with end.
The body is JSON: message (required, what to do, in words), history (the conversation so far, oldest first, each {role: user|assistant, text}) and model (a model id from GET /v1/models; empty takes the deployment's).
Events (event: name, data: JSON): say {text} the model's words; act {tool, args} a tool it reached for; result {tool, ok, text} what the tool answered; error {text} the turn could not go on; end {rounds} the turn is over. A : ping comment every 15 seconds keeps intermediaries from closing the stream.
The server keeps nothing between turns: send the conversation so far as history ({role: user|assistant, text}). The model is billed to the caller's account at the gateway's rates, and the sandbox hour is the one the lease already bills. Open the sandbox's screen (POST /v1/sandbox/{id}/screen/ticket) beside it to watch.
409 when the sandbox is not running; 404 for a sandbox you do not hold; 503 where no model gateway is mounted.
Request
1 field.
| Field | In | Type | Required | Description |
|---|---|---|---|---|
id | path | string | yes |
Response
The document declares no response body for this operation. It answers 200 on success and the platform error shape on failure — see Errors.
Examples
hanzo sandbox agent <id>import { Configuration, SandboxApi } from 'hanzoai';
const api = new SandboxApi(new Configuration({ accessToken: process.env.HANZO_API_KEY }));
const { data } = await api.postSandboxByIdAgent({ id: 'id' });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).post_sandbox_by_id_agent(id='id')cfg := hanzoai.NewConfiguration()
cfg.AddDefaultHeader("Authorization", "Bearer "+os.Getenv("HANZO_API_KEY"))
client := hanzoai.NewAPIClient(cfg)
resp, _, err := client.SandboxAPI.PostSandboxByIdAgent(context.Background()).Execute()
if err != nil {
return err
}use hanzo_client::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::post_sandbox_by_id_agent(&cfg, Default::default()).await?;import ai.hanzo.cloud.ApiClient;
import ai.hanzo.cloud.api.SandboxApi;
ApiClient client = new ApiClient();
client.setBearerToken(System.getenv("HANZO_API_KEY"));
var result = new SandboxApi(client).postSandboxByIdAgent();The method above is the one at the current release of the document. [email protected] (npm) and [email protected] (PyPI) were generated from an earlier release, where this operation carried a different id, so it spells the method differently — regenerating the clients is what makes the two agree. SDKs →
curl -X POST https://api.hanzo.ai/v1/sandbox/<id>/agent \
-H "Authorization: Bearer $HANZO_API_KEY"MCP reaches sandbox through the sandbox tool, which names its 18 operations with its own verbs — this one among them, under a name only MCP declares. describe explains any of them:
curl -X POST https://api.hanzo.ai/v1/mcp \
-H "Content-Type: application/json" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "describe",
"arguments": {
"op": "list_sandboxes"
}
}
}'