The leak of 500,000 lines of code from the world's strongest model Claude is already well known, containing a large amount of unreleased product architecture: the permanently resident Claude codenamed KAIROS, the 30-minute remote planning mode codenamed ULTRAPLAN, multi-agent coordination by coordinator, and cluster scheduling by agent swarms. But among all these "serious" architectural secrets, the most heated topic of discussion online is an electronic pet.

Its name is Buddy. Since everyone already knows, Claude simply released it on April Fool's Day.

The picture is of a colleague's good Buddy

The Tamagotchi in the Terminal

Next to the input box in the Claude Code terminal resides an ASCII pixel creature. It has its own species, rarity, five-dimensional attributes, and personality. It can "observe" your coding process and pop up with bubbles to say a few words while you debug. You can pet it, and it will float hearts. You can call its name, and it will chat with you.

Internally, Anthropic defines it not as a "decoration," but as "a separate watcher"—an independent observer. There is a clear comment in the source code: "Buddy is a separate entity and is not you (Claude)." The main model and the pet each have their own independent system prompt, and cannot speak for each other.

This is a companion system that has been carefully designed. Reverse analysis shows that the buddy system uses a dual-layer architecture. Internally, Anthropic refers to them as "Bones" and "Soul."

The skeletal layer is fully deterministic. The system uses Mulberry32—a lightweight 32-bit pseudo-random number generator—with hash(userId + 'friend-2026-401') as the seed, rolling out all the appearance attributes of your pet in one go. Species, rarity, eye style, hat, five-dimensional values are all determined by this seed. The same account will always get the same pet, no matter which machine it runs on. There is no choice, and no chance to redraw.

The soul layer, on the other hand, is non-deterministic. When you first execute /buddy to trigger "hatching," the Claude model generates a name and a personality description for your pet based on the attributes of the skeletal layer. A pet with high WISDOM will have a calm and reserved personality; a high CHAOS pet might be a chatterbox. This information is written into the companion field of ~/.claude.json for persistent storage. The soul is generated only once and cannot be reset.

The "401" in the salt is not a random number. April 1st, April Fool's Day.

18 Species, 1% Legendary Rarity

The complete species list: duck, goose, jelly, cat, dragon, octopus, owl, penguin, turtle, snail, ghost, hexagonal dinosaur, capybara, cactus, robot, rabbit, mushroom, fat cat.

Rarity is divided into five tiers: common (60%), uncommon (25%), rare (10%), epic (4%), legendary (1%). In addition, there is a separate 1% chance for Shiny—any species, any rarity can possibly trigger. The theoretical probability of a Shiny legendary capybara is one in ten thousand.

Image source: Xiaohongshu @Yoki's AI Lab

Each pet has five attributes: DEBUGGING, PATIENCE, CHAOS, WISDOM, SNARK. The system randomly designates one as the peak attribute (base value +50, maximum 100), one as the valley attribute (base value -10, minimum 1), and the remaining three are randomly distributed. The higher the rarity, the higher the base value—legendary level valley attributes might be higher than common level peak attributes.

In terms of appearance, there are 6 eye styles (· ✦ × ◉ @ °), 7 types of hats (crown, top hat, wizard hat, halo, propeller hat, beanie, topped with a little duck). Pets of common quality do not have hats, while uncommon and above are randomly assigned one.

The sprite body is ASCII ASCII art: 5 lines high, 12 characters wide, with each species having 3 frames of idle animation, refreshing every 500 milliseconds. The 0th line is reserved for hats. Eyes are injected into the body template using the {E} placeholder.

To the student in the crypto circle playing with NFTs, you're crying so sadly, what's going on?

How to Play?

The core commands are very simple. Type /buddy to hatch your pet for the first time, and then entering it again will summon it. /buddy pet is to pet it, hearts will float from the pet's head. /buddy card shows the pet card, displaying species, attributes, and rarity. /buddy off hides the pet. The most interesting part is that you can directly call the pet's name, and it will engage in an independent conversation with you based on its personality.

In daily use, the buddy will actively generate reactions based on your coding behavior—popping up a sentence in a bubble box. According to community reverse analysis, these reactions do not consume the user's token quota—but Anthropic has not yet confirmed this.

In terms of technical implementation, the buddy occupies fixed space at the bottom of the terminal. The system calculates the reserved width using companionReservedColumns to ensure that the input box and pet sprite do not overlap. The bubble box supports left-right switching and automatic line breaks.

The AI Programming Tool Has Started Keeping Pets

The salt friend-2026-401 points to April Fool's Day, but the code structure tells another story.

In the source code, the buddy system is gated by a BUDDY compilation flag. April 1st to 7th is set as the "preview window"—users can experience full functionality during this period. The official launch is scheduled for May. This is not a temporary Easter egg; it is a product feature with a complete release plan.

Community reactions have also confirmed this. Less than 48 hours after the leak, developers had already created a pet encyclopedia website (claude-buddy.vercel.app), a buddy query tool (input user ID to preview what you will draw), and even someone raised an Issue in Anthropic's GitHub repository requesting the addition of an RPG evolution system—to let pets level up based on actual token consumption.

From Anthropic's product logic, the goal of buddy is very clear: user stickiness. Claude Code is a command-line tool, and its usage scenario involves long-term, high-intensity programming sessions. In such scenarios, a companion with "personality" can alleviate the tool's coldness and create an emotional connection. The blind box mechanism creates natural social topics—"What species did you get?" "I'm a Shiny legendary."

GitHub Copilot does not chat with you. Cursor does not pop up a capybara saying "PATIENCE +3" when you write a bug.

Claude Code's buddy may be the first AI development tool companion system that has been carefully engineered and written into product source code. It has a complete deterministic generation pipeline, dual-layer architecture, independent LLM personality, ASCII rendering engine, and animation system. This is not a product of a weekend hackathon.

A company that makes AI programming assistants has invested engineering resources into an electronic pet. The fact itself is a signal—the competition among AI tools is shifting from "whose model is smarter" to "who can make developers want to stay."

Among 512,000 lines of code, the most popular isn't agent swarms, it's not KAIROS, it's a 5-line-tall ASCII capybara. It is wearing a wizard hat, with a SNARK value of 87, watching you write code.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。