Grok Computer Agent — v1.0 Release Notes

Released: 2026-04-15

Price: $4.99

Category: automation

What It Does

Grok Computer Agent brings xAI's autonomous browser and desktop control directly into OpenClaw workflows. You give it a plain-English task — "scrape the top 5 Hacker News headlines," "extract all pricing from this page," "fill in this contact form" — and it executes a full vision-action loop: screenshot, decide, act, repeat, until the task is complete or a step limit is reached. No API key required to try it out — the skill auto-detects the absence of XAI_API_KEY and runs in simulation mode so you can test pipelines and verify outputs before connecting a live account.

Use Cases

Researchers and analysts who need to automate recurring web research tasks — competitor pricing monitors, news aggregation, job board scraping — without writing bespoke scraper code for each target site; describe the task in natural language and Grok Computer Agent handles the navigation, extraction, and structured JSON output
Operations teams running repetitive form-filling and data entry workflows (government portals, supplier systems, internal tools with no API) who need autonomous UI automation that can be scheduled via OpenClaw cron and chained with downstream memory and reporting skills
Developers building agent pipelines who want to integrate Grok's vision-action capabilities as a step inside a larger OpenClaw workflow — kick off a Grok Computer Agent session to gather live web data, then pass the result straight into memory, analysis, or notification skills in the same pipeline

Requirements

OpenClaw — any recent version
Python 3.9+ — standard installation
requests library — pip install requests (required for live API calls; not needed in simulation mode)
Pillow library — pip install Pillow (required for screenshot capture in desktop mode)
pyautogui library — pip install pyautogui (optional — only needed for local desktop control mode)
XAI_API_KEY environment variable — get yours at [console.x.ai](https://console.x.ai); if absent, the skill runs in simulation mode automatically
xAI Grok Computer Agent API access — currently in beta (April 2026); sign up at console.x.ai

Example Usage

Dry-run: print task plan without any API calls:


python3 scripts/grok_computer_agent.py \
  --task "Get the top 5 headlines from news.ycombinator.com" \
  --dry-run


[DRY-RUN] Task plan:
  Task:      Get the top 5 headlines from news.ycombinator.com
  Mode:      browser
  URL:       (none)
  Max steps: 20
[DRY-RUN] No API calls made.

Run with auto-simulation (no API key set):


python3 scripts/grok_computer_agent.py \
  --task "Extract pricing" \
  --url https://example.com \
  --output /tmp/result.json


[INFO] Running in simulation mode (no API key or --dry-run)
[TASK] Extract pricing
[MODE] browser | max_steps=20 | url=https://example.com
[SESSION] sim_sess_4a7f9c2b
[STEP 1/20]
[STEP 2/20]
[COMPLETE] Task finished after 2 steps.
[RESULT] [SIMULATED] Task completed successfully.

[OUTPUT] Written → /tmp/result.json

result.json output:


{
  "task": "Extract pricing",
  "mode": "browser",
  "start_url": "https://example.com",
  "session_id": "sim_sess_4a7f9c2b",
  "steps_taken": 2,
  "task_complete": true,
  "result": "[SIMULATED] Task completed successfully.",
  "completed_at": "2026-04-15T14:05:00+00:00"
}

Live execution with verbose step output:


XAI_API_KEY=your_key_here python3 scripts/grok_computer_agent.py \
  --task "Research AI trends in enterprise software Q1 2026" \
  --mode browser \
  --max-steps 5 \
  --verbose


[TASK] Research AI trends in enterprise software Q1 2026
[MODE] browser | max_steps=5 | url=none
[SESSION] live_sess_c3d8e12f

[STEP 1/5]
  Action: navigate
  Reasoning: Opening search engine to begin research query

[STEP 2/5]
  Action: extract
  Reasoning: Extracting article headlines and summaries from search results

[COMPLETE] Task finished after 2 steps.
[RESULT] Found 8 relevant articles. Key trends: multi-agent orchestration adoption up 340%, ...

[OUTPUT] Written → result.json

Scheduled competitor pricing check (OpenClaw cron):


Run daily at 08:00: python3 scripts/grok_computer_agent.py \
  --task "Extract all plan names and monthly prices from example.com/pricing" \
  --url https://example.com/pricing \
  --output memory/competitor-pricing.json

Notes

Simulation mode activates automatically when XAI_API_KEY is not set or --dry-run is passed — no crashes, no silent failures
--max-steps 1 with no task completion emits a [WARN] and writes task_complete: false to the output JSON
--mode invalid triggers a clean argparse error with valid choices listed
Session cleanup always runs — even if the task fails, the API session is terminated to avoid billing for idle sessions
Rate limits (beta): 10 concurrent sessions, 100 steps/minute; respect --timeout to avoid hitting per-step limits
Advisory: --output and --screenshot-dir accept arbitrary paths — ensure you control the output destination in shared environments