himanshu748/vegas-amnesia — explained in plain English
Analysis updated 2026-05-18
Play a timed detective mini-game built on a real memory API
See how Cognee's remember, recall, memify, and forget operations work in practice
Study an example of an LLM-driven character reacting to a live knowledge graph
| himanshu748/vegas-amnesia | 0xallam/my-recipe | 0xhassaan/nn-from-scratch | |
|---|---|---|---|
| Stars | 0 | — | 0 |
| Language | Python | Python | Python |
| Last pushed | — | 2022-11-22 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 3/5 | 2/5 | 4/5 |
| Audience | general | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Needs a Cognee Cloud API key and, for full character dialogue, a Hugging Face or Anthropic API key.
Vegas Amnesia is a browser based detective game built for a hackathon centered on a memory platform called Cognee Cloud. The premise is playful: the player controls an AI assistant whose memory got wiped after a wild night in Las Vegas, and must piece together what happened, including tracking down a mysterious ring, before the owner's fiancee arrives at noon. What makes the game notable is that it is not just themed around memory, it is built directly on top of Cognee's real memory operations. Every piece of evidence the player chooses to "file" is actually written as a real memory into a live Cognee Cloud dataset, shown as a glowing node in a three dimensional knowledge graph the player can open at any time. Asking the in-game AI a question actually calls Cognee's recall function, and a "connect the dots" action triggers a real inference pass that derives new facts from what has already been stored. Deleting evidence that turns out to be false calls Cognee's forget operation, permanently removing that memory and visibly killing its node in the graph. Some clues found during play are outright false, and filing them poisons the AI's growing picture of the night, so part of the challenge is deciding what to trust before committing it to memory. Several characters in the game are driven by a language model and react based on what the memory graph already contains, occasionally contradicting themselves in ways a careful player can catch. The game runs as a FastAPI backend paired with a plain JavaScript and three.js frontend for the graph visuals, with each play session getting its own isolated Cognee dataset that is deleted on reset. Character dialogue is generated through a hosted Qwen model, and the whole project states it was built using Claude Code as part of the hackathon rules. A public, rate-limited version is playable online, alongside recorded demo videos.
A browser detective game where every clue filed, recalled, or deleted is a real operation against a live Cognee Cloud memory graph.
Mainly Python. The stack also includes Python, FastAPI, JavaScript.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly general.
This repo across BitVibe Labs
double-check against the repo, no cap.