Your AI Agent Forgot Why It Did That. PacketForge Is the Fix.

Description and use

PacketForge stores task decisions and debt as semantic embeddings, enabling agents and humans to quickly find and understand why choices were made.

Problem

AI coding agents make decisions that are not persisted, making it hard for future agents or humans to understand why choices were made.

Outcome

PacketForge enables AI agents to persist and surface decision rationale, reducing guesswork and improving collaboration across projects.

Workflow

• Create a task via POST /graph/tasks
• Record a decision via POST /decisions
• Search decisions via GET /graph/search
• View tasks and decisions in the Kanban dashboard

Evidence

Each one makes real decisions: "chose Postgres over SQLite for the MVP," "left auth storing tokens in plaintext, that's tracked debt," "picked this library because the other one had no TypeScript types."

Source

PacketForge is shared memory for a task graph: a small, self-hosted API any agent, CLI, or editor plugin can read from and write to, so "why did the last task do it this way" has a real answer instead of a re-read-the-diff-and-guess.

Source