Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.universalbench.dev/llms.txt

Use this file to discover all available pages before exploring further.

These are the patterns that turn UniversalBench from a remote sandbox into a real execution platform. Each one is built into the workbench and costs nothing extra to use.

​
Persistent sessions

By default every call is independent. Pass a session_id and Python state, imports, and cached connections persist across calls. 
{
  "session_id": "my_pipeline_001",
  "code": "import pandas as pd; df = pd.read_csv('/tmp/data.csv'); print(len(df))"
}
The next call with the same session_id can reference df without reimporting or rereading. To clear state: 
{ "session_id": "my_pipeline_001", "clear_session": true }

​
Adaptive caching

Repeated identical reads return instantly from a 30 second TTL cache. Control this per call. 
{
  "db_select": {
    "table": "customers",
    "filters": [["status", "eq", "active"]],
    "limit": 100,
    "cache_control": "skip"
  }
}
cache_controlBehaviour
auto (default)Use cache normally, store fresh results
skipBypass cache for this call only
refreshBypass read, store fresh result
clearClear all caches before the call
clear_dbClear database caches only
clear_queryClear query result cache only
To clear caches without running anything else, send a standalone request: 
{ "cache_op": "all" }

​
Validated code pushes

For Python files, validate_and_push runs a five gate pipeline before anything reaches GitHub. 
{
  "validate_and_push": {
    "owner": "your-org",
    "repo": "your-repo",
    "path": "app/main.py",
    "content": "your code here",
    "message": "deploy: new feature",
    "branch": "main",
    "sha": "current-file-sha"
  }
}
The gates: ruff lint, AST syntax check, runtime dry run, auto install on missing imports, and rollback SHA capture. If any gate fails, nothing pushes and you receive a structured error pointing at the failing line.

​
Safe deploy with auto rollback

safe_deploy adds a smoke test on top of validate_and_push. After the push, it hits a URL you specify and checks the response. If the response does not contain your expected string, it automatically reverts to the previous commit. 
{
  "safe_deploy": {
    "owner": "your-org",
    "repo": "your-repo",
    "path": "app/main.py",
    "content": "your code here",
    "message": "deploy: bump to v2",
    "smoke_test_url": "https://your-app.com/health",
    "smoke_test_expect": "version 2.0.0",
    "wait_seconds": 30
  }
}

​
Encrypted secrets vault

Store API keys, tokens, and other secrets once. Your AI injects them at runtime by name without ever seeing the values. 
{
  "secrets_vault": {
    "action": "save",
    "customer_id": "cust_001",
    "secret_name": "STRIPE_SECRET_KEY",
    "secret_value": "sk_live_xxx"
  }
}
AES 256 GCM encryption at rest. The master key never appears in any response or log.

​
Parallel execution

Run up to eight code blocks concurrently. Results return in order. 
{
  "parallel_blocks": [
    { "code": "import time; time.sleep(2); print('A done')" },
    { "code": "import time; time.sleep(2); print('B done')" },
    { "code": "import time; time.sleep(2); print('C done')" }
  ]
}
Total wall time is roughly two seconds instead of six.

​
Where to next

The complete capability list with descriptions and live playground is in the API Reference tab. The Quickstart covers the basic connect and call flow.