Skip to main content
This guide assumes you have already completed the steps in local setup guide. You should have already recorded your first automation and saved it as a recording in the Optexity dashboard.

​
Start the inference child process server

The primary way to run browser automations locally is via the inference child process server in optexity/inference/child_process.py. From the repository root: 
optexity inference --port 9000 --child_process_id 0
Key parameters:
  • --port: HTTP port the local inference server listens on (e.g. 9000).
  • --child_process_id: Integer identifier for this worker. Use different IDs if you run multiple workers in parallel.
When this process starts, it exposes:
  • GET /health – health and queue status
  • GET /is_task_running – whether a task is currently executing
  • POST /inference – main endpoint to allocate and execute tasks (see next section)

​
Call the /inference endpoint

With the server running on http://localhost:9000, you can allocate a task by sending an InferenceRequest to /inference.

​
Request schema

InferenceRequest (from optexity/schema/inference.py) has this shape:
  • endpoint_name: Name of the automation endpoint to execute. This must match a recording/automation defined in the Optexity dashboard.
  • input_parameters: dict[str, list[str]] – all input values for the automation, as lists of strings.
  • unique_parameter_names: list[str] – subset of keys from input_parameters that uniquely identify this task (used for deduplication and validation). Only one task with the same unique_parameter_names will be allocated. If no unique_parameter_names are provided, the task will be allocated immediately.
A minimal JSON example: 
{
    "endpoint_name": "extract_price_stockanalysis",
    "input_parameters": {
        "search_term": ["NVDA"]
    },
    "unique_parameter_names": []
}

​
Example curl request

curl -X POST http://localhost:9000/inference \
  -H "Content-Type: application/json" \
  -d '{
    "endpoint_name": "extract_price_stockanalysis",
    "input_parameters": {
      "search_term": ["NVDA"]
    },
    "unique_parameter_names": []
  }'
On success, the inference server:
  1. Forwards the request to your control plane at api.optexity.com using INFERENCE_ENDPOINT (defaults to api/v1/inference).
  2. Receives a serialized Task object from the control plane.
  3. Enqueues that Task locally and starts processing it in the background.
  4. Returns a 202 Accepted response like:
{
    "success": true,
    "message": "Task has been allocated"
}
Task execution (browser automation, screenshots, outputs, etc.) happens asynchronously in the background worker. You can see it running locally in your browser.

​
Monitor health and execution

You can monitor the task on the dashboard. It will show the status, errors, outputs, and all the downloaded files. Task runs

​
Video Tutorial