Create exec

Executes a program in a throwaway sandbox and answers with what it printed and what it left behind.

POST /v1/exec

Exec → · All API references → · Get started →

Address	`https://api.hanzo.ai/v1/exec`
Method	`POST`
Operation	`post_exec`
Auth	`Authorization: Bearer $HANZO_API_KEY`

Executes a program in a throwaway sandbox and answers with what it printed and what it left behind.

lang names one of the thirteen the sandbox image carries — py, js, ts, bash, r, php, go, rs, c, cpp, java, d, f90 — and code is the whole program, not a fragment: a compiled language is compiled and then run, an interpreted one is interpreted, and args becomes the program's own argv either way. Nothing is installed for you; the image is the environment.

A PROGRAM THAT FAILS IS A SUCCESSFUL CALL. A non-zero exit answers 200 with the diagnostics on stderr, because "the code threw" and "the interpreter is down" are different facts a caller renders differently. Only the second is an error status.

Runs are stateful through session_id. Omit it and the run gets a fresh sandbox whose id comes back on the answer; pass that id again and the next run sees the same filesystem, so a program can write a file one call and read it the next. files names bytes already uploaded to a session (POST /v1/exec/upload), copied in before the program starts. files on the ANSWER is what the program created or changed, by comparison against a marker taken at start — so it is the run's real output, not a listing of the directory — and each is fetched from GET /v1/exec/download/{session}/{name}.

The tenant is the caller's, never the body's, at every entry point. A typed op is also an MCP tool and an op-plane op; MCP's tools/call invokes it directly, with no route and therefore no middleware, so nothing there could have checked a credential. tenantOf refuses a context carrying no validated principal, so those entry points fail closed without a second gate to keep in step.

Request

11 fields, body application/json (required).

Field	In	Type	Required	Description
`args`	body	string[]	—	Args become the PROGRAM's argv, never the compiler's.
`code`	body	string	yes	Code is the WHOLE program, not a fragment: it is written to a single file and that file is what runs, so a compiled language needs its entry point and an interpreted one runs top to bottom.
`files`	body	exec.CodeFile[]	—	Files are inputs the host already put in some session.
`files[].id`	body	string	—	ID is the file's path RELATIVE to its session's artifact directory, which is also how it is fetched: GET /v1/exec/download/{session}/{id}.
`files[].name`	body	string	—	Name is the display name.
`files[].session_id`	body	string	—	SessionID is the other accepted spelling of the same fact on the way IN.
`files[].storage_session_id`	body	string	—	StorageSessionID names the session holding the bytes, and is the spelling the answer always uses.
`lang`	body	string	yes	Lang selects the toolchain, and with it the filename the code is written to and the line that runs it: py, js, ts, bash, r, php, go, rs, c, cpp, java, d, f90.
`runtime_session_hint`	body	string	—	RuntimeSessionHint is the stateful-session hint.
`session_id`	body	string	—	SessionID continues an EXISTING sandbox, which is what makes runs stateful: the same filesystem, so one run's output file is the next run's input.
`user_id`	body	string	—	UserID attributes the run inside the caller's org.

Response

Status	Body	Meaning
`200`	exec.CodeResult	ok
`default`	problem-details	refused

200 body — 8 fields.

Field	In	Type	Always	Description
`files`	body	exec.CodeFile[]	—	Files are what this run CREATED OR CHANGED, decided by mtime against a marker taken before the program started — so it is the run's output, not a listing of the directory.
`files[].id`	body	string	—	ID is the file's path RELATIVE to its session's artifact directory, which is also how it is fetched: GET /v1/exec/download/{session}/{id}.
`files[].name`	body	string	—	Name is the display name.
`files[].session_id`	body	string	—	SessionID is the other accepted spelling of the same fact on the way IN.
`files[].storage_session_id`	body	string	—	StorageSessionID names the session holding the bytes, and is the spelling the answer always uses.
`session_id`	body	string	—	SessionID is the sandbox this run used — the one that was passed in, or the fresh one that was leased.
`stderr`	body	string	—	Stderr is what the program wrote to standard error, INCLUDING a compiler's diagnostics and the trace of a program that exited non-zero.
`stdout`	body	string	—	Stdout is what the program wrote to standard output.

Failure carries the platform error shape — see Errors.

Examples

CLI

SDK

HTTP

MCP

hanzo exec create \
  --code <code> \
  --lang <lang>

.ts

.py

.go

.rs

.java

import { Configuration, ExecApi } from 'hanzoai';

const api = new ExecApi(new Configuration({ accessToken: process.env.HANZO_API_KEY }));
const { data } = await api.postExec({ lang: "<lang>", code: "<code>" });

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

client = ApiClient(Configuration(access_token=os.environ["HANZO_API_KEY"]))
result = ExecApi(client).post_exec(lang="<lang>", code="<code>")

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

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

use hanzo_client::apis::{configuration::Configuration, exec_api};

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

let result = exec_api::post_exec(&cfg, Default::default()).await?;

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

ApiClient client = new ApiClient();
client.setBearerToken(System.getenv("HANZO_API_KEY"));

var result = new ExecApi(client).postExec();

curl -X POST https://api.hanzo.ai/v1/exec \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
       "lang": "<lang>",
       "code": "<code>"
     }'

MCP reaches exec through the exec tool, which names its 2 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": "get_exec_file"
         }
       }
     }'

Exec API · All Hanzo APIs · Interactive reference

Was this page useful?

PreviousCreate upload NextList assign