Automate repetitive browser tasks like logging in, filling forms, or searching without calling an AI model for each step.
Automate native desktop actions such as launching apps or typing, with human approval required for sensitive steps.
Handle ambiguous browser tasks by optionally asking GPT-5.6 to propose the next step, which is then validated and verified locally.
| chengmatt416/lhic | 000madz000/payload-test-api-route-handler | akash-sr/memobsidian | |
|---|---|---|---|
| Stars | 3 | 3 | 3 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | hard | easy | moderate |
| Complexity | 4/5 | 2/5 | 3/5 |
| Audience | developer | developer | general |
Figures from each repo's GitHub metadata at analysis time.
Requires Node.js 24, a local Playwright Chromium install, and running from source since the published npm package is outdated.
LHIC, short for Local Human Intent Controller, is a runtime for automating browser and desktop tasks that runs locally on your machine rather than relying on a cloud service to decide what to do. Its goal is to take a task a person wants done and turn it into precise, checkable computer actions, such as clicking, typing, or navigating a web page. Most common tasks, like logging into a site, filling out a form, or searching, run through what the project calls the Fast Path, which uses the Playwright browser automation library and pre-built skills without ever calling an AI model, so there is no per-action cost or delay from waiting on a model. For trickier, less defined tasks, an optional Slow Path can call GPT-5.6 to propose a next step, but LHIC still checks that proposal against its own rules, requires human approval for sensitive actions, and verifies the result actually happened before moving on. The project's documentation is explicit that this AI model is a helper for planning, not the thing actually in charge of execution. The project also controls native desktop actions across macOS, Windows, and Linux, such as opening apps, typing, or clicking, again requiring approval and a check afterward that the action really occurred. Security features include encrypted storage for cookies and session data, automatic removal of passwords and personal information from logs, and support for managing approval keys through cloud key management services. The code is organized as a monorepo with separate packages for schemas, browser automation, verification, tracing, memory storage, security, and a command line tool. Setup requires Node.js 24, installing dependencies, and downloading a local Chromium browser through Playwright. The README notes that the published npm package is an older version and points users to run the project from a checkout of this repository instead until a newer release is published. The Slow Path's use of GPT-5.6 is optional and only activates if the user supplies their own API key.
A local-first automation runtime that turns browser and desktop tasks into verified, approval-gated actions, using fast rule-based execution for common tasks and an optional GPT-5.6 helper for ambiguous ones.
Mainly TypeScript. The stack also includes TypeScript, Playwright, Node.js.
No license is stated in the visible README content.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.