Memory Star Map That Forgets: An Open-Source Plugin Where Every Memory Is a Star

In an era where AI memory management tools are heavily homogenized, an open-source project from Bilibili redefines how memories are presented using a "star map": each memory becomes a star, its size determined by weight, and its brightness reflecting how long it has gone unrecalled. This memory plugin visualizes the abstract mechanism of forgetting and provides a complete engineering implementation.

Design Philosophy: Turning Memory Decay into an Observable Starry Sky

Traditional memory plugins usually manage content with lists or tags, making it hard for users to intuitively perceive which memories are "fading." This plugin introduces an astronomical metaphor: memory weight determines star size, and recent access frequency determines brightness. Over time, low-weight memories that have long been unused gradually dim and drift outward along their orbits.

According to the creator, the project simulates tens of days of memory evolution—memories with little presence dim and slowly drift to the edges. This dynamic decay mechanism lets users see at a glance which memories are being "marginalized" by the brain, helping them decide whether to reinforce or let them fade.

Memory decay is a common mechanism in cognitive science and AI memory systems, referring to the gradual weakening of memory strength over time and with lack of access. In AI applications, this is typically implemented through time-decay functions, such as the Ebbinghaus forgetting curve or exponential decay models, which downweight, archive, or delete low-weight or long-unaccessed memories. The star map visualizes this abstract process, allowing users to intuitively perceive the "freshness" and importance changes of memories. This design draws on the brain's memory consolidation and forgetting mechanisms, helping the system automatically manage memory priorities and reducing the burden of manual cleanup.

Three Orbital Rings and the Fixed Star Mechanism: Stable, To-Do, and Archive

The star map uses three orbits corresponding to three memory tiers: the inner ring is the stable zone, holding core and frequently used memories; the middle ring is the to-do zone, for content that needs attention but has not yet been consolidated; the outer ring is the archive zone, for low-frequency memories. These tiers are not static labels but dynamically migrate with memory activity.

The most distinctive feature is the "fixed star" at the center. According to the creator, the fixed star never forgets and does not participate in elimination. This provides a "pinning" mechanism for critical memories, preventing important information from being accidentally caught by the decay algorithm. The fixed star is also only displayed in the star map but does not enter the context, meaning it does not interfere with the AI's real-time reasoning and exists only as a long-term reference.

The fixed star mechanism is similar to the "pinning" or "favorite" function in memory systems, used to mark critical memories as unforgettable. In AI memory management, such mechanisms typically allow users or the system to set certain memories as permanently retained, bypassing the decay algorithm. The star map's fixed star is not only exempt from elimination but is also displayed only in the visualization interface without entering the context—similar to the separation between "long-term reference" and "real-time working memory." The context window is the limited text input during AI reasoning; the fixed star not entering the context means it does not consume reasoning resources and exists only as an external knowledge base, avoiding interference with real-time decisions. This design balances memory protection and reasoning efficiency.

Engineering Implementation: Zero Dependencies, 66 Tests, and an HTTP Interface

This plugin is also noteworthy at the engineering level. According to the creator, the project achieves zero dependencies—after cloning, a single command initializes it, lowering the deployment barrier. Functionality covers 66 tests, including concurrent writes, command injection, and bad files, all of which passed actual testing. Such test coverage is uncommon in personal open-source projects, especially the consideration of security boundaries (command injection) and exception handling (bad files), indicating the author has clear requirements for production readiness.

The plugin also provides an HTTP interface that can connect to other programs. This means the star map is not just a visualization panel but can also serve as a memory service called by external applications, expanding its use cases. The repository address is in the video description and comments, making it easy for developers to access directly.

Zero-dependency means the project does not rely on any third-party libraries or frameworks; all functionality is implemented based on the standard library. This reduces deployment complexity and version conflict risks, allowing it to run with a single command after cloning. The HTTP interface exposes memory services through RESTful APIs or similar protocols, allowing other programs (such as chatbots or automation scripts) to call the star map's memory management functions for cross-application integration. The tested scenarios of concurrent writes, command injection, and bad files respectively verify multi-thread safety, security vulnerability protection, and abnormal input handling—key quality metrics for production-grade software.

An Observation: Can Visualization Truly Improve Memory Management?

From a product perspective, the value of the star map lies in turning "forgetting" from an implicit process into explicit feedback. Users no longer need to manually clean up memories; the system automatically prioritizes through decay and orbital migration. But this also raises a question: once memories are visualized, will users over-intervene in the decay process, thereby undermining the originally automated filtering mechanism?

Additionally, while the fixed star mechanism solves the problem of protecting critical memories, how to define a "fixed star" still depends on user judgment. If there are too many fixed stars, the star map's dynamism is compromised; if too few, important information may be missed. Finding the balance requires practice.

Overall, this open-source plugin offers a novel approach to memory visualization, and its engineering completeness exceeds expectations. For developers who need to manage large amounts of AI memories, it is worth a try.