ATaC-team/ATaC

What it solves

ATaC allows users to turn successfully completed tasks into reusable programs. Instead of an AI agent repeating the same complex sequence of tool calls and logic every time, ATaC captures these patterns as programs that can be stored, tested, and reused with new inputs.

How it works

ATaC integrates with the pi platform to monitor tasks. When a task is completed, the system can extract a candidate program. These candidates are placed in an "Inbox" where the user reviews the code, manifest, and tests. Once the user tests and publishes the program, it becomes available to the AI agent via a set of dedicated tools (atac_search, atac_get, atac_run). The agent is then instructed to prefer these published programs over manual tool-by-tool execution.

Who it’s for

Developers and power users of the pi platform who want to automate repetitive AI agent workflows by creating a library of verified, reusable program components.

Highlights

• Human-in-the-loop verification: Programs must be reviewed and pass tests in the Inbox before being published.
• Sandbox execution: Programs run in a Codemode sandbox, ensuring that file and network access are controlled via declared tools.
• TUI Management: Includes a full-screen terminal board to manage untested, in-progress, and published programs.
• Automated extraction: Background analysis can automatically identify and propose new programs based on completed tasks.