Demonstration-to-skill automation for macOS

AI Mime turns one recorded workflow into a runnable skill.

Record a task once. AI Mime reflects the trace into semantic workflow artifacts, optimizes the execution path, and packages it as readable code with inputs, fallback context, and repair hooks.

Download for macOS Browse marketplace Watch the demo View source

AI Mime tasks dashboard with ready workflows and a start recording button

AI Mime menu bar record controls — Record from the menu bar

AI Mime skill build chat confirming workflow inputs and outcome — Confirm and build the skill

See it in action

Record once. Build the skill. Run it forever.

1. Record Do the task once — clicks, typing, and screens are captured automatically. 2. Build An agent compiles it into a fast, deterministic skill — code for the spine, an LLM only at decision points. 3. Run Replay with new inputs. If the environment changed, it self-heals instead of failing.

How it works

Record → Reflect → Optimize → Build Skill → Run / Heal

Record

Capture screenshots, clicks, keystrokes, extracts, and user notes from the demonstrated workflow.

Reflect

Convert the raw trace into coordinate-free schema artifacts with task parameters and subtasks.

Optimize

Choose the most reliable executor: deterministic script first, browser harness second, native UI agent last.

Build Skill

Package run scripts, input templates, and fallback notes into a portable skill directory.

Run / Heal

Run with new inputs. If execution fails, an agent can inspect logs, complete the run, and patch the skill.

What is novel

The recording is evidence, not the automation.

Semantic trace compilation

A UI trace becomes task intent, targets, parameters, subtasks, and success criteria.

Intent optimization

Manual UI paths can be replaced with APIs, CLI calls, file parsing, browser CDP, or AppleScript.

Portable executable skills

Finished automations are readable packages that Claude Code or Codex can call directly.

Agentic healing

Broken runs can be triaged, completed, and repaired instead of being rerun from scratch.

Current app surfaces

From task library to replay and recovery.

Tasks dashboard with workflow list, statuses, and run actions — **Tasks dashboard** Scan available workflows, start recording, and launch ready skills.

Menu bar dropdown with Start Recording and dashboard actions — **Record button** Start a new demonstration from the macOS menu bar.

Skill build chat explaining inputs, outcome, and irreversible side effects — **Skill build** The agent confirms inputs, outputs, and high-level execution before packaging code.

Replay form for entering skill inputs before running a workflow — **Run input form** Provide new inputs and run the deterministic skill package.

Autohealing screen showing failed run logs and agent recovery instructions — **Autohealing** When a run fails, the agent reads logs, triages the issue, and repairs the skill when needed.

Developer inspection

Every stage leaves files you can read.

recordings/<id>/manifest.jsonl Raw event stream: screenshots, actions, extracts, and notes.

workflows/<id>/schema.json Semantic workflow produced from the recording.

workflows/<id>/optimized_plan.json Executor plan for script, browser harness, and UI-agent steps.

workflows/<id>/skills/<slug>/ Portable skill package with runnable code and fallback references.

workflows/<id>/runs/ Run logs, copied assets, outputs, and replay summaries.

Open source automation you can inspect

Build from source or download the macOS app.

Download app Browse marketplace Read architecture Developer guide