Give Think-Act a clear outcome, the files or screenshots it needs, and any limits it should follow. It works through the task with the tools you enable, keeps the activity visible, and lets you save a proven process for later.
Describe the task in plain language
Pause before gated actions
Reuse workflows that prove reliable
Available for 64-bit Windows 10 and Windows 11. macOS is planned.
Automation run
Illustrative workflow overview
Task brief
Outcome, source material, boundaries
Enabled tools
Only available capabilities are exposed.
Approval
Gated actions can pause before execution.
Activity
A workflow that proves reliable can become a reusable local skill.
The best first automation is one that produces a result you can verify in under two minutes. Pick one task from your week that happens over and over again.
Security policy can require confirmation before a workflow writes files, sends messages, or changes an external system. Keep those approval gates enabled while you test a workflow. Start bounded. Expand later.
First Automation Screenshot
Recommended: 900 × 700
The agent workspace keeps everything visible: your request, the optional success criteria, which tools are enabled, and what the agent did at each step. When security policy requires approval, the prompt appears before that gated action is taken. When the run is done, the history is there to inspect.
Desktop Agent Workspace Screenshot
Recommended: 1920 × 1080
Describe what you need and add files, folders, or screenshots as context. No command syntax required.
Add a completion condition when you need to verify whether the task was actually finished correctly.
Review recorded tool calls, their results, and the observations captured during the run.
Once a workflow proves reliable, save it locally with its prerequisites, required connectors, and expected output.
A useful workflow says what Think-Act should work from, what it should produce, what it must leave alone, and how you will know the task is complete.
Example task brief
“Use these invoices to create a CSV with vendor, date, subtotal, tax, and total. Do not rename or move the source files. Flag any field you cannot read.”
Input
What should it work from?
Output
What should it produce?
Limits
What should stay untouched?
Check
What does “done” look like?
Give Think-Act the source material, expected result, and any boundaries that matter. Specific inputs make the result easier to review.
The agent works with the capabilities available to that run. If a required connector is missing, the workflow should report the limitation instead of pretending the action happened.
Compare the result with the original material and inspect supported execution evidence when it matters. Do not turn an unverified first run into a permanent workflow.
Once the process consistently produces the result you expect, document its prerequisites and save it as a reusable local skill for later runs.
OCR and document intelligence turn scanned files, screenshots, and image-locked PDFs into structured, workable content. The result feeds directly into the automation that needs it.
Extract from any format
Scans, screenshots, image-based PDFs, and forms.
Structure the output
Names, dates, totals, clauses, line items, and action tasks.
Chat with extracted content
Summarize, question, and analyze using OCR output as context.
Prepare follow-up content
Turn extracted data into emails, checklists, or summaries ready to send.
OCR and AI interpretation can miss context. Check important names, amounts, dates, and obligations against the original.
Document Automation Screenshot
Recommended: 1200 × 900
Some procedures are easier to demonstrate than describe from memory. Capture the workflow while you do it, then use Think-Act to prepare steps you can check, edit, share, and save as a local skill.
Step 01
Capture the workflow once while you do it
Step 02
Agent extracts context and numbered instructions
Step 03
Add warnings, context, and clarifications
Step 04
Store the workflow as a local reusable playbook
Step 05
Run the reviewed playbook again from the agent workspace
Screen Recorder & SOP Screenshot
Recommended: 1200 × 900
Support, HR, QA, and product teams use screen recording to document onboarding procedures, software walkthroughs, and recurring manual processes they would otherwise re-explain repeatedly.
From a single recording, generate a step-by-step SOP, a checklist, a bug report, a tutorial draft, or a reusable skill playbook saved locally for later runs.
Think-Act can prepare information without a connector, but changing files, querying services, or interacting with external systems requires the appropriate enabled tool.
When the filesystem or GitHub connector is enabled and healthy, the agent can read files, list directories, and query repositories as allowed by the configured tool permissions and security policy.
Example
Read last week's PR description and summarize changes using the GitHub connector.
Security policy can hold file modifications, external messages, and database writes behind an approval prompt. The agent explains the planned action before it executes.
Approval prompt
Create issue in pixlab/think-act with the draft below. Confirm to continue.
When a workflow needs a connector that is not installed or not enabled, the agent stops and explains what is required. It does not simulate the missing capability.
Not installed
This task requires the Jira connector. Install and enable it to continue.
Local MCP packages run code on your device and are not controlled by PixLab. Review third-party packages before installing them and use only sources you trust.
Think-Act can pause consequential actions before they happen, show you what the workflow intends to do, and wait for your decision when the configured security policy requires approval.
Allow
Trusted actions continue automatically
Ask
Pause for review before continuing
Block
Prevent actions the workflow should not take
Keep sensitive writes, sends, deletes, and external changes behind approval while testing a workflow.
Straight answers about what Think-Act can and cannot automate, how approvals work, and how to stay in control. More questions on the full FAQ page.
Try Your First Workflow with Think-ActNot automatically. Automating actions in a desktop application requires an enabled connector or tool for that application. Without one, Think-Act can still produce instructions, summaries, drafts, and structured output, but it cannot click or interact with the application directly. See how integrations and connectors work.
Think-Act can show an approval prompt before a write or external action when the configured security policy requires confirmation. Read-only steps may run without a prompt, depending on policy, and their tool observations and results remain available for review afterward. Review the approval and recovery controls before automating consequential actions.
No. Start with bounded, reversible workflows where you can check the output against the source in a minute or two. Keep confirmation enabled for anything that modifies files, sends messages, or calls external systems until you trust the workflow. The security guide explains the Allow, Ask, and Block model.
Think-Act identifies the missing connector or capability instead of pretending the action was completed. It explains what the workflow needs so you can review the available integration options, enable a trusted tool, and try again.
Once you have reviewed the steps and the output proves reliable, save the workflow as a local skill. Document the prerequisites, required connectors, and expected result so the agent can reference it correctly the next time you run it. You can also generate an SOP from a screen recording.
Yes. Think-Act is free to install and try. The free tier includes writing assistance, OCR, document summarization, and agent tasks within the free-plan limits. See the Think-Act downloads page for current requirements and plan details.
Choose one bounded task, add the source material and a clear finish condition, then review the work before you reuse the workflow.