Together Code Interpreter

Together Code Interpreter (TCI) enables you to execute Python code in a sandboxed environment.

The Code Interpreter currently only supports Python. We plan to expand the language options in the future.

ℹ️
MCP Server
TCI is also available as an MCP server through Smithery. This makes it easier to add code interpreting abilities to any MCP client like Cursor, Windsurf, or your own chat app.

Run your first query using the TCI

from together import Together

client = Together()

# Run a simple print statement in the code interpreter
response = client.code_interpreter.run(code='print("Welcome to Together Code Interpreter!")', language="python")


print(f"Status: {response.data.status}")


for output in response.data.outputs:
    print(f"{output.type}: {output.data}")

import Together from 'together-ai';

const client = new Together();

const response = await client.codeInterpreter.execute({
  code: 'print("Welcome to Together Code Interpreter!")',
  language: 'python',
});

if (response.errors) {
  console.log(`Errors: ${response.errors}`);
} else {
  for (const output of response.data.outputs) {
    console.log(`${output.type}: ${output.data}`);
  }
}

curl -X POST "https://api.together.ai/tci/execute" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "language": "python",
    "code": "print(\"Welcome to Together Code Interpreter!\")"
  }'

Output

Status: completed
stdout: Welcome to Together Code Interpreter!

ℹ️
Pricing information
TCI usage is billed at $0.03/session. As detailed below, sessions have a lifespan of 60 minutes and can be used multiple times.

Example Use Cases

Reinforcement learning (RL) training: TCI transforms code execution into an interactive RL environment where generated code is run and evaluated in real time, providing reward signals from successes or failures, integrating automated pass/fail tests, and scaling easily across parallel workers—thus creating a powerful feedback loop that refines coding models over many trials.
Developing agentic workflows: TCI allows AI agents to seamlessly write and execute Python code, enabling robust, iterative, and secure computations within a closed-loop system.

Response Format

The API returns:

session_id: Identifier for the current session
outputs: Array of execution outputs, which can include:
- Execution output (the return value of your snippet)
- Standard output (stdout)
- Standard error (stderr)
- Error messages
- Rich display data (images, HTML, etc.)
  Example

{
  "data": {
    "outputs": [
      {
        "data": "Hello, world!\n",
        "type": "stdout"
      },
      {
        "data": {
          "image/png": "iVBORw0KGgoAAAANSUhEUgAAA...",
          "text/plain": "<Figure size 640x480 with 1 Axes>"
        },
        "type": "display_data"
      }
    ],
    "session_id": "ses_CM42NfvvzCab123"
  },
  "errors": null
}

Usage overview

Together AI has created sessions to measure TCI usage.

A session is an active code execution environment that can be called to execute code, they can be used multiple times and have a lifespan of 60 minutes.

Typical TCI usage follows this workflow:

Start a session (create a TCI instance).
Call that session to execute code; TCI outputs stdout and stderr.
Optionally reuse an existing session by calling its session_id.

Reusing sessions and maintaining state between runs

The session_id can be used to access a previously initialized session. All packages, variables, and memory will be retained.

from together import Together

client = Together()

# set a variable x to 42
response1 = client.code_interpreter.run(code="x = 42", language="python")

session_id = response1.data.session_id

# print the value of x
response2 = client.code_interpreter.run(
    code='print(f"The value of x is {x}")', language="python", session_id=session_id
)

for output in response2.data.outputs:
    print(f"{output.type}: {output.data}")

import Together from 'together-ai';

const client = new Together();

 // Run the first session
const response1 = await client.codeInterpreter.execute({
  code: 'x = 42',
  language: 'python',
});

if (response1.errors) {
  console.log(`Response 1 errors: ${response1.errors}`);
  return;
}

// Save the session_id
const sessionId = response1.data.session_id;

// Resuse the first session
const response2 = await client.codeInterpreter.execute({
  code: 'print(f"The value of x is {x}")',
  language: 'python',
  session_id: sessionId,
});

if (response2.errors) {
  console.log(`Response 2 errors: ${response2.errors}`);
  return;
}

for (const output of response2.data.outputs) {
  console.log(`${output.type}: ${output.data}`);
}

curl -X POST "https://api.together.ai/tci/execute" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "language": "python",
    "code": "x = 42"
  }'
  
  curl -X POST "https://api.together.ai/tci/execute" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "language": "python",
    "code": "print(f\"The value of x is {x}\")",
    "session_id": "YOUR_SESSION_ID_FROM_FIRST_RESPONSE"
  }'

Output

stdout: The value of x is 42

Using the TCI for Data analysis

Together Code Interpreter is a very powerful tool and gives you access to a fully functional coding environment. You can install Python libraries and conduct fully fledged data analysis experiments.

from together import Together

client = Together()

# Create a code interpreter instance
code_interpreter = client.code_interpreter

code = '''
!pip install numpy
import numpy as np

# Create a random matrix
matrix = np.random.rand(3, 3)
print("Random matrix:")
print(matrix)

# Calculate eigenvalues
eigenvalues = np.linalg.eigvals(matrix)
print("\\nEigenvalues:")
print(eigenvalues)
'''

response = code_interpreter.run(
  	code=code,
    language="python"
)

for output in response.data.outputs:
    print(f"{output.type}: {output.data}")
if response.data.errors:
    print(f"Errors: {response.data.errors}")

import Together from 'together-ai';

const client = new Together();

// Data analysis
const code = `
  !pip install numpy
  import numpy as np

  # Create a random matrix
  matrix = np.random.rand(3, 3)
  print("Random matrix:")
  print(matrix)

  # Calculate eigenvalues
  eigenvalues = np.linalg.eigvals(matrix)
  print("\\nEigenvalues:")
  print(eigenvalues)
`;

const response = await client.codeInterpreter.execute({
  code,
  language: 'python',
});

if (response.errors) {
  console.log(`Errors: ${response.errors}`);
} else {
  for (const output of response.data.outputs) {
    console.log(`${output.type}: ${output.data}`);
  }
}

curl -X POST "https://api.together.ai/tci/execute" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "language": "python",
    "code": "!pip install numpy\nimport numpy as np\n# Create a random matrix\nmatrix = np.random.rand(3, 3)\nprint(\"Random matrix:\")\nprint(matrix)\n# Calculate eigenvalues\neigenvalues = np.linalg.eigvals(matrix)\nprint(\"\\nEigenvalues:\")\nprint(eigenvalues)"
  }'

Uploading and using files with TCI

from together import Together

client = Together()

# Create a code interpreter instance
code_interpreter = client.code_interpreter

script_content = "import sys\nprint(f'Hello from inside {sys.argv[0]}!')"

# Define the script file as a dictionary
script_file = {
    "name": "myscript.py",
    "encoding": "string",
    "content": script_content,
}

code_to_run_script = "!python myscript.py"

response = code_interpreter.run(
    code=code_to_run_script,
    language="python",
    files=[script_file],  # Pass the script dictionary in a list
)

# Print results
print(f"Status: {response.data.status}")
for output in response.data.outputs:
    print(f"{output.type}: {output.data}")
if response.data.errors:
    print(f"Errors: {response.data.errors}")

import Together from 'together-ai';

// Initialize the Together client
const client = new Together();

// Create a code interpreter instance
const codeInterpreter = client.codeInterpreter;

// Define the script content
const scriptContent = "import sys\nprint(f'Hello from inside {sys.argv[0]}!')";

// Define the script file as an object
const scriptFile = {
  name: "myscript.py",
  encoding: "string",
  content: scriptContent,
};

// Define the code to run the script
const codeToRunScript = "!python myscript.py";

// Run the code interpreter
async function runScript() {
  const response = await codeInterpreter.run({
    code: codeToRunScript,

curl -X POST "https://api.together.ai/tci/execute" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "language": "python",
    "files": [
      {
        "name": "myscript.py",
        "encoding": "string",
        "content": "import sys\nprint(f'\''Hello from inside {sys.argv[0]}!'\'')"
      }
    ],
    "code": "!python myscript.py"
  }'

Output

Status: completed
stdout: Hello from inside myscript.py!

Pre-installed dependencies

TCI's Python sessions come pre-installed with the following dependencies, any other dependencies can be installed using a !pip install command in the python code.

- aiohttp
- beautifulsoup4
- bokeh
- gensim
- imageio
- joblib
- librosa
- matplotlib
- nltk
- numpy
- opencv-python
- openpyxl
- pandas
- plotly
- pytest
- python-docx
- pytz
- requests
- scikit-image
- scikit-learn
- scipy
- seaborn
- soundfile
- spacy
- textblob
- tornado
- urllib3
- xarray
- xlrd
- sympy

List Active Sessions

To retrieve all your active sessions:

curl -X GET "https://api.together.ai/tci/sessions" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json"

Output:

{
  "data": {
  "sessions": [
    {
      "id":"ses_CM3zokEcfkdh5G8UiKApw",
      "started_at":"2025-03-12T10:34:22.125248Z",
      "execute_count":2,
      "last_execute_at":"2025-03-12T10:37:24.145685Z",
      "expires_at":"2025-03-12T11:04:22.125248Z"
    }
    ]
  },
  "errors": null
}

Troubleshooting & questions

If you have questions about integrating TCI into your workflow or encounter any issues, please contact us.

Together Code Interpreter

ℹ️
MCP Server

Run your first query using the TCI

ℹ️
Pricing information

Example Use Cases

Response Format

Usage overview

Reusing sessions and maintaining state between runs

Using the TCI for Data analysis

Uploading and using files with TCI

Pre-installed dependencies

List Active Sessions

Further reading

Troubleshooting & questions

ℹ️MCP Server

Run your first query using the TCI

ℹ️Pricing information

Example Use Cases

Response Format

Usage overview

Reusing sessions and maintaining state between runs

Using the TCI for Data analysis

Uploading and using files with TCI

Pre-installed dependencies

List Active Sessions

Further reading

Troubleshooting & questions

ℹ️
MCP Server

ℹ️
Pricing information