> ## Documentation Index > Fetch the complete documentation index at: https://docs.infery.ai/llms.txt > Use this file to discover all available pages before exploring further. # Code tools > Code tools available as workflow steps and through /v1/capabilities. **2 tools.** Each one is deterministic: the same input gives the same output, and nothing is inferred. Call one directly with [`POST /v1/capabilities/{id}/run`](/api-reference/capabilities/run), or use it as a step in a [workflow](/workflows/step-types). Prices are on the [live catalogue](https://infery.ai/models) — they change, and a stale price here would be worse than none. | Tool | Does | In → out | | ------------------------------------ | --------------------------------------- | ---------- | | [`code.run_node`](#coderun_node) | Execute Node.js in a sandboxed runtime. | any → text | | [`code.run_python`](#coderun_python) | Execute Python in a sandboxed runtime. | any → text | ## `code.run_node` **any → text** · Billed per second Execute a Node.js snippet in an isolated sandbox and return the value of `result` (or `output`). ### When to use * **JSON manipulation** — parse, transform, or validate JSON payloads with native JS tooling. * **String processing** — use regex, template literals, or Buffer operations. * **Lightweight scripting** — run any logic that benefits from the Node.js standard library. ### Inputs & outputs | | | | --------------------- | --------------------------------------------------- | | **Input modality** | Any (named variable bindings via `vars`) | | **Output modality** | Text / JSON (the value of `result` after execution) | | **Max source length** | 50 000 characters | | **Max timeout** | 30 seconds | ### Input | Field | Description | | ------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `vars` | Named bindings. Each key that is a valid identifier — and is not `input`, `result` or `output` — is injected as a bare variable. Every key, including ones that are not valid identifiers, is also reachable through the `input` object. | ### Parameters | Param | Required | Description | | ----------------- | ----------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | `source` | one of `source`/`files`/`archive_file_id` | Node.js source code (1–50 000 chars). Assign your answer to `result` (or to `output`, which is accepted for compatibility; `result` wins if you set both). Assigning neither returns `null`. **Unlike Python**, an explicit `result = undefined` does *not* win over `output`: the wrapper runs your code with `new Function`, which returns `undefined` for a `var` that was declared but never assigned — so "assigned `undefined`" and "never assigned" look identical, and the value falls through to `output` (or `null` if that is also unset). If you need to return a genuine "no value", assign `result = null` explicitly. | | `files` | one of `source`/`files`/`archive_file_id` | A project: path → content. See [Multi-file projects](#multi-file-projects). | | `archive_file_id` | one of `source`/`files`/`archive_file_id` | A stored file id whose unpacked contents become the working directory. See [Vendored dependencies](#vendored-dependencies). | | `entrypoint` | with `files` or `archive_file_id` | Which file runs: a key of `files`, or a path inside the unpacked archive. | | `timeout_seconds` | no | Execution time limit in seconds (1–30). Default: `5` | ### Output | Field | Description | | ------------------ | ---------------------------------------------------------------------------- | | `result` | The value you assigned. Must be JSON-serialisable. | | `stdout` | Everything the program printed, up to 64 KiB. | | `stderr` | Anything written to standard error, up to 64 KiB. | | `stdout_truncated` | `true` when `stdout` hit the 64 KiB cap and was cut short. Absent otherwise. | | `stderr_truncated` | `true` when `stderr` hit the 64 KiB cap and was cut short. Absent otherwise. | ### Runtime * **Standard library only.** The sandbox image has no package manager; `numpy`, `pandas` and `requests` are not available. * **No network.** Outbound connections, including DNS, are blocked. * **No persistent filesystem.** Anything written is discarded when the run ends. * Memory is capped at 512 MB (Python) / 1 GB (Node) and the run is killed at `timeout_seconds`. ### Examples #### Simple arithmetic ```json theme={null} { "type": "code.run_node", "input": { "vars": {} }, "params": { "source": "result = 1 + 1", "timeout_seconds": 5 } } ``` #### Transform an array with input variables ```json theme={null} { "type": "code.run_node", "input": { "vars": { "items": ["foo", "bar", "baz"] } }, "params": { "source": "result = items.map(s => s.toUpperCase()).join(', ')", "timeout_seconds": 5 } } ``` #### Print progress while returning a value ```json theme={null} { "type": "code.run_node", "input": { "vars": { "data": [1, 2, 3] } }, "params": { "source": "console.log('summing')\nresult = data.reduce((a, b) => a + b, 0)", "timeout_seconds": 5 } } ``` ### Multi-file projects Instead of `source`, a step can carry `files` — a map of path to content — with `entrypoint` naming which of them runs. `source` and `files` are mutually exclusive: set exactly one. The files are written into a fresh working directory for the run, and the entry point executes from there, so one file can import another (`require('./lib/util')` in Node). The directory is deleted when the run ends; nothing is shared between runs. In a multi-file run the entry point does not execute as a plain script: it runs inside a wrapper with a real `module`/`exports` pair bound, the way an ordinary CommonJS module does. The run's answer is chosen in this order: `result`, then `output`, then `module.exports` (counted as an answer when it was reassigned — `module.exports = {...}` — or when the initially-empty object was populated in place — `exports.foo = ...`), then `null`. Files other than the entry point that you load with `require('./lib/util')` are ordinary CJS modules with no such caveat — their own `module.exports` behaves exactly as Node normally behaves, independent of this priority order. Paths must be relative, use `/` as the separator, and stay inside the working directory — no leading `/`, no `..`. Limits: at most 64 files, 64 KiB per file, and 96 KiB for the whole request (files, vars and all). Exceeding one is a validation error, not a failed run. Dependency packages are not available: only the standard library and whatever the runtime image already ships. If you need a third-party package, see [Vendored dependencies](#vendored-dependencies) below. ### Vendored dependencies **Archive support is off by default on this deployment** — an operator must explicitly enable it, or a run using `archive_file_id` refuses with `archive_mode_disabled` regardless of anything an author sets. Instead of `source` or `files`, a step can carry `archive_file_id` — the id of a previously-uploaded archive whose unpacked contents become the working directory. `entrypoint` names which unpacked file runs (a relative path, e.g. `src/main.js`). **Upload a `.zip` or a `.tar.gz`/`.tgz`** to get the id — those are the two formats the file upload endpoint accepts for this purpose. The sandbox can also unpack a `.rar` (and one can still reach it through a worker-registered file), but the upload endpoint does not admit one directly, so an author cannot get a `.rar` into `archive_file_id` by uploading it. Re-pack as `.zip` or `.tar.gz` if that's what you have. This is how you bring a third-party package into the sandbox: pack your project **together with its dependencies already installed** (e.g. a `node_modules` directory checked in ahead of time) into the archive. **The dependencies are vendored inside the archive — nothing is installed, and no package registry is ever contacted.** This is not a shortcut: **the sandbox has no network at all**, so `npm install`, `pip install`, or any other install step run *inside* the sandbox would simply fail. All of a run's dependencies must already be sitting on disk, inside the archive, before the run starts. `source`, `files` and `archive_file_id` are mutually exclusive: set exactly one. Unlike `files`, the archive's contents are not known until it is unpacked inside the sandbox at run time, so only `entrypoint`'s presence is checked when the step is saved — a wrong path is discovered when the run actually executes, not before. **Archive limits:** | Limit | Value | | ----------------------------- | ------------------------------------ | | Uploaded archive size | 21 MiB, before decompression | | Unpacked (decompressed) size | 200 MiB total | | Entries (files + directories) | 20 000 | | Size of any single file | 50 MiB | | Compression ratio | 100:1 per entry (rejects a zip bomb) | Exceeding any of these refuses the upload or the run with a named error — an oversized archive is refused before it is unpacked, before an unpack slot or an execution pod is spent on it. A real `node_modules` routinely has more than 20 000 entries; prune `devDependencies`, `.bin`, docs and test fixtures out of the vendored tree if you hit the entry-count limit. ### Pricing Billed per second of sandbox execution time, measured as an exact fraction — a run of 0.42 s is billed for 0.42 s, with no rounding up to a whole second. See pricing dashboard for current rates. ### Related capabilities * `code.run_python` — execute Python code in a sandbox * `web.search` — retrieve live data to process with this capability ### Examples **Add 1 to x** ```json theme={null} { "type": "code.run_node", "params": { "source": "result = x + 1", "timeout_seconds": 5 } } ``` **A two-file project** ```json theme={null} { "type": "code.run_node", "params": { "files": { "main.js": "const lib = require('./lib');\nresult = lib.double(x);\n", "lib.js": "module.exports = { double: (n) => n * 2 };\n" }, "entrypoint": "main.js", "timeout_seconds": 5 } } ``` ## `code.run_python` **any → text** · Billed per second Execute a Python snippet in an isolated sandbox and return the value of `result` (or `output`). ### When to use * **Data transformation** — reshape, filter, or aggregate structured data before passing it downstream. * **Calculations** — run numerical or statistical computations without a separate service. * **Scripted automation** — generate or manipulate text, JSON, or binary payloads inline. ### Inputs & outputs | | | | --------------------- | --------------------------------------------------- | | **Input modality** | Any (named variable bindings via `vars`) | | **Output modality** | Text / JSON (the value of `result` after execution) | | **Max source length** | 50 000 characters | | **Max timeout** | 30 seconds | ### Input | Field | Description | | ------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `vars` | Named bindings. Each key that is a valid identifier — and is not `input`, `result` or `output` — is injected as a bare variable. Every key, including ones that are not valid identifiers, is also reachable through the `input` dictionary. | ### Parameters | Param | Required | Description | | ----------------- | ----------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `source` | one of `source`/`files`/`archive_file_id` | Python source code (1–50 000 chars). Assign your answer to `result` (or to `output`, which is accepted for compatibility; `result` wins if you set both). Assigning neither returns `null`. | | `files` | one of `source`/`files`/`archive_file_id` | A project: path → content. See [Multi-file projects](#multi-file-projects). | | `archive_file_id` | one of `source`/`files`/`archive_file_id` | A stored file id whose unpacked contents become the working directory. See [Vendored dependencies](#vendored-dependencies). | | `entrypoint` | with `files` or `archive_file_id` | Which file runs: a key of `files`, or a path inside the unpacked archive. | | `timeout_seconds` | no | Execution time limit in seconds (1–30). Default: `5` | ### Output | Field | Description | | ------------------ | ---------------------------------------------------------------------------- | | `result` | The value you assigned. Must be JSON-serialisable. | | `stdout` | Everything the program printed, up to 64 KiB. | | `stderr` | Anything written to standard error, up to 64 KiB. | | `stdout_truncated` | `true` when `stdout` hit the 64 KiB cap and was cut short. Absent otherwise. | | `stderr_truncated` | `true` when `stderr` hit the 64 KiB cap and was cut short. Absent otherwise. | ### Runtime * **Standard library only.** The sandbox image has no package manager; `numpy`, `pandas` and `requests` are not available. * **No network.** Outbound connections, including DNS, are blocked. * **No persistent filesystem.** Anything written is discarded when the run ends. * Memory is capped at 512 MB (Python) / 1 GB (Node) and the run is killed at `timeout_seconds`. ### Examples #### Add 1 to an input variable ```json theme={null} { "type": "code.run_python", "input": { "vars": { "x": 5 } }, "params": { "source": "result = x + 1", "timeout_seconds": 5 } } ``` #### Parse and summarise JSON ```json theme={null} { "type": "code.run_python", "input": { "vars": { "data": [1, 2, 3, 4, 5] } }, "params": { "source": "result = {'sum': sum(data), 'count': len(data)}", "timeout_seconds": 5 } } ``` #### Print progress while returning a value ```json theme={null} { "type": "code.run_python", "input": { "vars": { "data": [1, 2, 3] } }, "params": { "source": "print('summing')\nresult = sum(data)", "timeout_seconds": 5 } } ``` ### Multi-file projects Instead of `source`, a step can carry `files` — a map of path to content — with `entrypoint` naming which of them runs. `source` and `files` are mutually exclusive: set exactly one. The files are written into a fresh working directory for the run, and the entry point executes from there, so one file can import another (`import lib.util` in Python, `require('./lib/util')` in Node). The directory is deleted when the run ends; nothing is shared between runs. Paths must be relative, use `/` as the separator, and stay inside the working directory — no leading `/`, no `..`. Limits: at most 64 files, 64 KiB per file, and 96 KiB for the whole request (files, vars and all). Exceeding one is a validation error, not a failed run. Dependency packages are not available: only the standard library and whatever the runtime image already ships. If you need a third-party package, see [Vendored dependencies](#vendored-dependencies) below. ### Vendored dependencies **Archive support is off by default on this deployment** — an operator must explicitly enable it, or a run using `archive_file_id` refuses with `archive_mode_disabled` regardless of anything an author sets. Instead of `source` or `files`, a step can carry `archive_file_id` — the id of a previously-uploaded archive whose unpacked contents become the working directory. `entrypoint` names which unpacked file runs (a relative path, e.g. `src/main.py`). **Upload a `.zip` or a `.tar.gz`/`.tgz`** to get the id — those are the two formats the file upload endpoint accepts for this purpose. The sandbox can also unpack a `.rar` (and one can still reach it through a worker-registered file), but the upload endpoint does not admit one directly, so an author cannot get a `.rar` into `archive_file_id` by uploading it. Re-pack as `.zip` or `.tar.gz` if that's what you have. This is how you bring a third-party package into the sandbox: pack your project **together with its dependencies already installed** (e.g. a `site-packages`-style vendor directory your code adds to `sys.path`, or anything else that works without a package manager) into the archive. **The dependencies are vendored inside the archive — nothing is installed, and no package registry is ever contacted.** This is not a shortcut: **the sandbox has no network at all**, so `pip install`, `npm install`, or any other install step run *inside* the sandbox would simply fail. All of a run's dependencies must already be sitting on disk, inside the archive, before the run starts. `source`, `files` and `archive_file_id` are mutually exclusive: set exactly one. Unlike `files`, the archive's contents are not known until it is unpacked inside the sandbox at run time, so only `entrypoint`'s presence is checked when the step is saved — a wrong path is discovered when the run actually executes, not before. **Archive limits:** | Limit | Value | | ----------------------------- | ------------------------------------ | | Uploaded archive size | 21 MiB, before decompression | | Unpacked (decompressed) size | 200 MiB total | | Entries (files + directories) | 20 000 | | Size of any single file | 50 MiB | | Compression ratio | 100:1 per entry (rejects a zip bomb) | Exceeding any of these refuses the upload or the run with a named error — an oversized archive is refused before it is unpacked, before an unpack slot or an execution pod is spent on it. A real `node_modules` routinely has more than 20 000 entries; prune `devDependencies`, `.bin`, docs and test fixtures out of the vendored tree if you hit the entry-count limit. ### Pricing Billed per second of sandbox execution time, measured as an exact fraction — a run of 0.42 s is billed for 0.42 s, with no rounding up to a whole second. See pricing dashboard for current rates. ### Related capabilities * `code.run_node` — execute Node.js code in a sandbox * `web.search` — retrieve live data to process with this capability ### Examples **Add 1 to x** ```json theme={null} { "type": "code.run_python", "params": { "source": "result = x + 1", "timeout_seconds": 5 } } ``` **A two-file project** ```json theme={null} { "type": "code.run_python", "params": { "files": { "main.py": "import lib\nresult = lib.double(x)\n", "lib.py": "def double(n):\n return n * 2\n" }, "entrypoint": "main.py", "timeout_seconds": 5 } } ```