[KongchangAI]
ProductKoishi框架

Koishi

An open-source chatbot framework built on a "everything is a plugin" design philosophy. It has a kernel-free architecture and ships approximately 195 official plugins, supporting LLM integration, adapters, session management, and a WebUI, with a high degree of extensibility.

Timeline (last 90 days)

Sep 27

Cordis is a general-purpose dependency injection and plugin container framework that matured in the Koishi ecosystem and was later adopted by DeepSeek Harness as its underlying runtime

Unverified50%
Sep 12

The Cordis framework is derived from a same-named service-container library in the JavaScript ecosystem, originally used for the chatbot framework Koishi.

Unverified70%
Sep 12

Koishi officially ships with approximately 195 built-in plugins, covering functional modules such as utilities, LLM integration, adapters, session management, WebUI, and sandboxing.

Unverified50%
Sep 12

A Koishi plugin is composed of four core structures: name, inject (dependencies), Config (configuration), and apply (entry point).

Unverified50%
Sep 12

apply is the sole execution entry point of a Koishi plugin, where all core logic is written, typically including registration methods such as tools.register.

Unverified50%
Sep 12

Koishi defines six main runtime states: pending (not ready, dependencies unmet), loading (apply executing), active (loaded and active), failed (error occurred), unloading (being unloaded), and disposed (removed).

Unverified50%
Sep 12

Koishi provides three service injection methods: hard dependency (inject), context.get (on-demand retrieval at runtime), and context.inject (injection in a specific context).

Unverified50%
Sep 12

Koishi's realm service isolation rule is: rows that publish a service must enter the realm, while rows that only consume a service must remain outside the realm.

Unverified50%
Sep 12

When binding event listeners with context.on, next must be used in conjunction to ensure the middleware chain is correctly passed along.

Unverified50%
Sep 12

Koishi's configuration is organized in four layers: the bundle layer, the profile layer, the home layer, and codespatch.

Unverified50%

9 more timeline events

All Facts (20)

Unverified

Koishi是基于Node.js的QQ机器人开发框架,支持多平台

90%
Unverified

The Cordis framework is derived from a same-named service-container library in the JavaScript ecosystem, originally used for the chatbot framework Koishi.

70%
Unverified

Cordis is a general-purpose dependency injection and plugin container framework that matured in the Koishi ecosystem and was later adopted by DeepSeek Harness as its underlying runtime

50%
Unverified

The core design philosophy of Koishi is "no kernel, everything is a plugin."

50%
Unverified

Koishi officially ships with approximately 195 built-in plugins, covering functional modules such as utilities, LLM integration, adapters, session management, WebUI, and sandboxing.

50%
Unverified

A Koishi plugin is composed of four core structures: name, inject (dependencies), Config (configuration), and apply (entry point).

50%
Unverified

apply is the sole execution entry point of a Koishi plugin, where all core logic is written, typically including registration methods such as tools.register.

50%
Unverified

Koishi defines six main runtime states: pending (not ready, dependencies unmet), loading (apply executing), active (loaded and active), failed (error occurred), unloading (being unloaded), and disposed (removed).

50%
Unverified

Koishi provides three service injection methods: hard dependency (inject), context.get (on-demand retrieval at runtime), and context.inject (injection in a specific context).

50%
Unverified

Koishi's realm service isolation rule is: rows that publish a service must enter the realm, while rows that only consume a service must remain outside the realm.

50%
Unverified

Koishi provides four resource cleanup methods: context.on, context.effect, context.setTimeout, and context.setInterval.

50%
Unverified

When binding event listeners with context.on, next must be used in conjunction to ensure the middleware chain is correctly passed along.

50%
Unverified

Koishi's configuration is organized in four layers: the bundle layer, the profile layer, the home layer, and codespatch.

50%
Unverified

Koishi distinguishes between two runtime planes: host (process-based, with process-level configuration written in the profile's codespatch) and agent/preset (session-based, with configuration written in the agent preset directory of the home folder).

50%
Unverified

Koishi preset releases published as online packages must use a bare package name, while local references must use an absolute path (relative paths are also supported).

50%
Unverified

In Koishi plugins, the description field is the basis on which the model makes judgments, determining whether the plugin can be correctly identified and invoked by the model.

50%
Unverified

Behind Koishi UI interface operations, there are four loading layers: the module loading layer, the unit layer (unit, Context construction), the mount layer (patching the plugin into the system), and the end-to-end layer (actually consuming the model API).

50%
Unverified

The Koishi plugin publishing process is: fill in meta information such as repository URL, keywords, and export in package.json, run npm publish --access public to push to npm, and users install it via a command.

50%
Unverified

Cordis is an Inversion of Control (IoC) container and plugin management framework for the JavaScript/TypeScript ecosystem, originally validated at scale within the Koishi chatbot framework.

50%
Unverified

Cordis是源自Koishi跨平台聊天机器人框架的通用插件运行时系统,设计哲学受微内核操作系统思想影响

50%

Source Articles