sigpanic/goink

What it solves

Goink is a desktop AI writing system designed for long-form fiction writing. It solves the problem of "memory loss" in general-purpose LLMs, where the AI forgets character names, plot points, and foreshadowing across dozens of chapters. It replaces the need for manual tracking of story elements and manual searching through previous chapters to maintain consistency.

How it works

Goink uses an Agent-based architecture with a custom ReAct loop and 31 structured tools. Instead of a linear pipeline, the AI agent autonomously decides which tools to use to query character profiles, update plot arcs, or read and write text.

Key technical components include:

• Structured Memory: Dedicated tracking for character relationships (as directed graphs), foreshadowing status, narrative arcs, world-building (location hierarchies), and reader knowledge.
• Local Semantic Search: A local engine using ONNX Runtime and sqlite-vec to perform meaning-based searches across the entire manuscript without requiring an internet connection.
• Skill System: A modular system where writing methodologies are defined in Markdown files, allowing for three levels of override (Novel > User > Built-in) and three trigger modes (auto, manual, always).
• Approval Workflow: To prevent unwanted changes, the system generates a Diff preview for any text edits, requiring user confirmation before writing to disk.
• ListNode: Built-in Git integration for automatic commits and version history.

Who it’s for

Fiction writers and novelists who need an AI assistant that can maintain strict narrative consistency over hundreds of thousands of words and manage complex story structures.

Highlights

• Autonomous Maintenance: Automatically triggers character and plot updates after a chapter is written, with an optional independent "Review Agent" to ensure consistency.
• Style Distillation: Can analyze a sample text to extract writing style patterns (sentence structure, tone, etc.) and generate a custom "Imitation Skill" for the AI to follow.
• Zero-Configuration Setup: Distributed as a standalone desktop app (Wails/Go) that doesn't require Python or a GPU.
• Sandboxed Security: Uses a regex whitelist and SafePath to ensure the AI cannot access or modify unauthorized files on the system.