Show HN:AI Mime – 為電腦使用代理錄製並參數化工作流程
AI Mime 是一款新推出的原生 macOS RPA 工具,能夠錄製複雜的使用者工作流程,利用 LLM 將其精煉為參數化指令,並為電腦使用代理進行重播,旨在彌合 RPA 的採用差距。
Navigation Menu
Search code, repositories, users, issues, pull requests...
Provide feedback
We read every piece of feedback, and take your input very seriously.
Saved searches
Use saved searches to filter your results more quickly
To see all available qualifiers, see our documentation.
AI Mime: Record, refine, and replay complex workflows with computer-use agents.
License
Uh oh!
There was an error while loading. Please reload this page.
prakhar1114/ai_mime
Folders and files
Latest commit
History
Repository files navigation
![]()
AI Mime
Record, Refine and Replay on macOS.
AI Mime is a native macOS RPA tool designed to empower computer use agents with the context they need to succeed. It moves beyond simple text prompts by capturing the rich ground truth required for reliable automation.
Record: Capture every nuance of a workflow—mouse movements, keystrokes, and window states—providing the detailed ground truth that standard LLMs often lack.
Refine: Leverage LLMs to transform raw recording data into parameterized agent instructions. This process converts static recordings into dynamic, manageable subtasks, allowing you to define inputs and dependencies that keep the agent operating within strict, observable boundaries.
Replay: Dispatch computer use agents to execute complex, multi-step processes across native apps and the web on demand, ensuring high fidelity and repeatability.
Why AI Mime?
The Adoption Gap in Computer Use Models
While the latest computer-use models have demonstrated significant improvements in general-purpose reliability and accuracy, they have yet to see widespread adoption in the Robotic Process Automation (RPA) industry. The primary bottleneck is not the model's intelligence, but the interface of current products.
The Context Problem Existing tools typically rely on users to describe tasks using simple natural language. However, standard text prompts are often too brief to capture the nuance of complex workflows.
When an agent operates with this "low-context" instruction over a long horizon, the probability of error compounds. Without strict boundaries, models may attempt to be "creative" in their problem-solving, leading to unpredictable behaviors that are unacceptable in enterprise automation.
The AI Mime Solution - Rich Context & Repeatability
AI Mime bridges this gap by shifting from vague natural language to rich, demonstrated context. Instead of relying on a single long-horizon prompt, AI Mime allows you to:
Capture Detailed Workflows: "Show" the agent exactly how to perform a task, capturing the precise steps and decision paths.
Enforce Observability: Break down complex workflows into manageable subtasks, each grounded in high-quality examples.
Eliminate Unwanted Creativity: By providing a strict chain of steps and rich context, the agent is constrained to operate exactly as instructed, ensuring it executes the workflow rather than "interpreting" it.
In short, AI Mime transforms abstract instructions into repeatable, observable, and error-resistant workflows, making computer-use agents viable for production RPA.
Demo
Watch the demo on YouTube
Record
Start capturing a workflow by clicking AI Mime → Start Recording in the macOS menu bar. When you’re done, click AI Mime → Stop Recording to save it.


Replay
To run a saved workflow, open AI Mime → Replay, choose a workflow from the dropdown, enter the new parameters, and click Run. The workflow will execute automatically—no manual intervention needed.

Installation
Permissions (Critical)
To function correctly, ai-mime requires macOS permissions. Because you are likely running this from a Python virtual environment, you must grant permissions to the Python binary inside your .venv.
Use Cmd+Shift+G when trying to add the python binary in settings.
Note: If you see "Terminal" in the list but it still doesn't work, try removing it and re-adding it, or run the script from a dedicated terminal window.
Environment keys (.env)
Create a .env file (or export env vars) with:
and add it in the repo root
Usage
Start app (record + replay UI)
Start the menubar app:
PS: The app creates a lot of processes in the background. These might persist on quitting the app. The following command to force kills all these processes:
Output
Recordings are saved in recordings/<session_id>/:
Workflows are saved in workflows/<session_id>/:
CLI: Reflect
If you want to run reflect manually on an existing recording:
Glossary
Replay (Run a workflow)
To run an existing workflow:
About
AI Mime: Record, refine, and replay complex workflows with computer-use agents.
Resources
License
Uh oh!
There was an error while loading. Please reload this page.
Stars
Watchers
Forks
Releases
Packages
0
Languages
Footer
Footer navigation
相關文章