Open-Source AI Desktop Pet Coopanion: DeepSeek Big Fat Fish Moves Into Your Taskbar — Install at Your Own Risk

An AI Big Fat Fish Living in Your Taskbar

The very bottom of your screen is normally taskbar territory. Now someone's handed it over to a fish — a desktop pet shaped like the DeepSeek^Ad Big Fat Fish, from the open-source project Coopanion (spoken as OpenCloud in the video; refer to the project's actual name). According to the Bilibili creator, this desktop companion is completely free with open-source code, and its appearance can be swapped at will — the DeepSeek Big Fat Fish is just one of the more lovable options.

The creator opens with a warning: "Think it through before installing, because once it moves into the bottom of your screen, you'll keep wanting to poke it." This "anti-recommendation recommendation" actually nails the core experience of desktop pets — what it offers isn't the functionality of a productivity tool, but the mild interaction stickiness that comes from a sense of companionship. For users whose desktops feel too quiet, that might be a plus; for people who need to focus, it's genuinely a distraction.

Reason One to Think Twice: It Remembers Everything About You

Coopanion writes your name, your chat content — everything — into a local file on your computer, and keeps a version record every time it's modified. This means two things: data never leaves your machine, so the privacy boundary is relatively clear; and conversation history is traceable, with version management turning "it remembers what you said" into a verifiable fact rather than a black box.

For people who care about data sovereignty, local storage is a plus. But you should also realize that local files mean backups and cleanup are your responsibility.

Behind local storage and version records is a trade-off between "data sovereignty" and "auditability." Writing conversations to a local file and keeping historical versions gives the AI's long-term memory a physical carrier that users can inspect and delete, preventing conversation content from being uploaded to uncontrollable clouds. But it also means disk usage grows with conversation volume; if sensitive information gets written to the file, users need to encrypt it themselves or clean it up regularly; and multi-device sync won't happen automatically — you'll have to migrate manually when switching machines.

Reason Two: Pick a Personality — You're Asked on First Launch

The creator lists several personality presets — curious, clingy, sharp, lively, quiet, and more — and lets users choose on first launch. This essentially packages system prompts and interaction styles into "personality templates," lowering the barrier for ordinary users to configure an AI character. Different personalities directly affect the desktop pet's frequency of proactive behavior and way of speaking, so picking the wrong one may take some getting used to.

Personality presets are usually implemented through system prompts: developers write role settings, tone style, and proactive speaking frequency into templates that get injected into the model context at startup. This approach is cheap and fast to switch, but it isn't model fine-tuning, so personality consistency depends on prompt quality — long conversations may break character. Different personality templates also affect token consumption — chattier personalities naturally cost more.

Reason Three: It Asks Before It Acts

This is the most noteworthy part of the project's security design. When Coopanion wants to click buttons on your computer, it asks for your consent first; the moment your mouse touches it, it stops immediately; and sensitive operations like login and payment it never touches — those are left entirely to you.

This mechanism controls the risk of "AI agents operating your computer" at the pre-confirmation stage rather than remediating afterward. Combined with the local storage design, you can see the project takes a conservative stance on permission boundaries — a reasonable trade-off for a desktop pet whose main selling point is companionship.

"Confirm before acting, stop on mouse touch, stay away from sensitive operations" corresponds to permission tiering and human-in-the-loop design for AI agents. Agents can call system interfaces to simulate clicks and input; without gates, misoperations or prompt injection attacks could cause real damage. Coopanion puts confirmation upfront and hard-codes high-risk actions like login and payment as non-delegable — a practice of the principle of least privilege. But users should still be wary: when confirmation popups appear frequently, it's easy to fall into the habit of clicking "allow" without thinking.

Reason Four: It's Way Too Easy to Bully

Poke it, pat its head, or even pick it up and fling it away. This kind of physical interaction is a classic design in the desktop pet category, and it's the direct reason behind the creator's "you'll keep wanting to poke it." The smoother the interaction feedback, the easier it is to go tease it during work breaks — something users need to weigh for themselves.

Reason Five: No Model Loyalty — A Dozen-Plus to Choose From

Coopanion supports free switching among a dozen-plus models including DeepSeek, Qwen, Kimi, and Claude, with configuration so simple it's just "paste a T" (meaning filling in an API Token). With an extension installed, it can even play Plants vs. Zombies by itself.

Not being tied to a single model provider means users can mix and match based on cost, response speed, and capability preferences. The creator also mentions the cost issue: models are pay-per-use, while the software itself is free and open source.

Multi-model switching relies on the APIs and token mechanisms each provider offers. Tokens are the basic units of text that models process, billed separately for input and output, so long conversations and frequent proactive speaking significantly drive up costs. The API Token users "paste a T" for is both an identity credential and a billing basis — keep it safe; if leaked, others can steal your quota. Differences in context windows and function-calling capabilities across models also affect the desktop pet's proactive behavior.

The Most Critical Point: Free and Open Source — the Ecosystem Is Growing Fast

The creator's data: less than four weeks after launch, the project has over 350 stars on GitHub, and netizens have already made about 20 derivative versions. For a desktop pet project, this community growth rate shows real demand — many people genuinely want a desktop companion that can change appearance, chat, and remember them.

Install or Not?

Coopanion's positioning is clear: it's not a productivity tool, but a desktop companion program with memory, personality, and security boundaries. Local storage, confirmation before acting, and staying away from sensitive operations — these three make it more reassuring than most "AI takes over your computer" projects; and being free, open source, and multi-model supported greatly lowers the cost of trying it.

What you really need to figure out is: are you willing to let a little thing that remembers you, actively seeks you out, and is very easy to bully occupy those bottom few centimeters of your taskbar long-term? If yes, search Coopanion on GitHub; if you're not sure, then the creator's warning may be right.

One-line Summary

Free, open-source AI desktop pet Coopanion — with memory, personality, and security boundaries.

Paragraph Summary

This article introduces the open-source desktop pet project Coopanion, which lives in your taskbar with swappable appearances like the DeepSeek Big Fat Fish, focusing on companionship rather than productivity. Key features include: names and chat logs stored in local files with version history, so data never leaves your machine; personality presets like curious and clingy selectable on first launch; asking before operating your computer, stopping on mouse touch, and never touching sensitive operations like login or payment; and free switching among a dozen-plus models including DeepSeek, Qwen, Kimi, and Claude, with models billed by usage. Less than four weeks after launch, it has over 350 GitHub stars and 20-plus derivatives. The author considers its security design more reassuring than most "AI takes over your computer" projects, but whether you're willing to let a little thing that remembers you and actively seeks you out occupy your taskbar long-term is up to you.