Koishi: a chatbot framework that grew a whole ecosystem

A framework named for an unconscious mind

Koishi is a cross-platform, extensible, high-performance chatbot framework, and its README explains the unusual name with a story: both the name and the icon design come from Komeiji Koishi, a character from the Touhou Project games who acts unconsciously, chosen to symbolize the theme of a chatbot while carrying the affection its developer poured into the work. The framework is written entirely in TypeScript and licensed MIT, with copyright lines running from 2019 under the author Shigma, and the project's homepage at koishi.chat hosts the documentation, the console and the plugin marketplace that the rest of the story is built on. The framing matters for expectations, this is a personal project that matured into infrastructure, not a corporate product with a support department.

A console for people who will never read the code

The first advertised property is being usable out of the box, and the specifics are aimed at a non-programmer: a highly convenient console lets you set up your own chatbot within minutes, without any foundation, and an online plugin marketplace means even users with no programming background can download and install plugins from inside the console. Once running, the same control panel monitors operational status at any time, steers the bot's behavior, and, in a feature the README describes with visible affection, even lets you log in and chat as the bot. This audience decision shapes everything else, the plugin system is not merely an architecture for developers but a distribution channel for operators, and the console is the storefront.

Four platforms, multiple accounts, shared data

Platform coverage is the framework's cross-platform claim made concrete: QQ, Telegram, Discord and Feishu are the mainstream chat platforms named, with multi-account support and cross-platform data interoperation, so a user's state, credentials or game progress can follow them between networks. The interoperation point deserves emphasis because it is the hard part of multi-platform bots, most frameworks can post to several APIs, but keeping one identity model coherent acrossQQ and Discord is where such projects usually degrade into parallel silos. Combined with the console's multi-account handling, the intended deployment is one bot living simultaneously in several communities, administered from one place, which matches how Chinese open-source communities in particular actually operate, spread across QQ groups, Discord servers and Telegram channels at once.

Three thousand plugins after four years of iteration

The ecosystem number is the headline feature: after four years of iteration, Koishi has developed a rich plugin ecosystem with more than 3000 official and community plugins, covering platform support, databases, resource storage, the web console, state management and concrete business functionality. The README is careful to frame this for two audiences at once, whether the goal is a large interactive application or a lightweight auxiliary bot, the framework provides the relevant practices, and detailed documentation stands ready for anyone afraid of getting lost in the complexity. For an evaluator, the plugin count functions as a compatibility statement, the parts of bot authoring that are undifferentiated work, session handling, storage, permissions, marketplaces, already exist, and the 3000 figure is the evidence that the abstraction has held up under other people's code, not just its authors'.

TypeScript all the way down, with hot reload

The developer story rests on three named pillars. Type support is the first: Koishi is fully developed in TypeScript with top-tier typing, and the README claims the code completion is rich enough that you rarely need to consult documentation while writing. Unit testing is the second, all core functionality passes unit tests, which the project presents both as reliability and as a worked example of how plugin authors should test and locate problems in their own work. Hot reload is the third and the most felt in daily use: while developing a plugin, saving the file reloads it without restarting the bot, an experience the README compares to frontend development. Together the three form a coherent pitch, the framework is pleasant to extend precisely because it was extended by its own maintainers first.

A monorepo with upstream satellites

The repository layout reveals the framework's true shape, a yarn 4 workspace monorepo whose workspace globs include not only packages and plugins but external trees under external/upstream/adapter, external/upstream/database and external/upstream/server, meaning the platform adapters, database drivers and server components live as satellite repositories woven into one build. Root scripts name the constellations, console, dialogue, minato for the database layer, satori for the protocol layer, and yakumo, the in-house build tool chaining esbuild and tsc with version bumping, dependency upgrading and publishing commands. Tests run through mocha with esbuild registration, coverage through c8 in text, JSON and HTML, and a crowdin.yml coordinates translations. The scope of the machinery matches the scope of the ecosystem it ships.

Release bursts, quiet summers, a donation link

Release history shows a distinctive rhythm: versions 4.18.9 and 4.18.10 landed within minutes of each other on 2026-03-08, with 4.18.11 following the next day, a patch burst typical of fixing something in a just-shipped fix, and the last repository push came on 2026-08-28. Getting started routes through the starter manual at koishi.chat rather than inline instructions, contributions have a guide and a discussions venue, license compliance is scanned by FOSSA, and the author accepts support through the afdian donation platform, the customary patronage channel for Chinese independent developers. Against single-platform bot libraries, the difference is architecture: those bind you to one network's idioms, while Koishi's adapters, shared data layer and console exist precisely so the same bot can live everywhere its users are.

Editorial conclusion

Choose Koishi when one bot must serve several chat platforms from a single codebase and console, or when non-programmers should manage the running instance through a web panel. Choose a single-platform library when you target exactly one network and want its native idioms without abstraction. Verify first that adapters exist for every platform you need, read the starter manual at koishi.chat before installing plugins from the marketplace, and check plugin maintenance dates, since a 3000-plugin ecosystem inevitably spans the actively kept and the long abandoned.

Frequently asked questions

Official sources

1. koishijs/koishi on GitHub
2. License: MIT
3. Project website
4. README
5. Releases