openclawopenclaw

A technique where write operations are collected and processed later in the background, so the program doesn't have to wait for each save to finish.

A component that translates messages from a specific platform (like Discord or Telegram) into a standard format the system can understand.

The technique of controlling how much conversation history is fed to the AI model, often by summarizing or trimming old messages to stay within memory limits.

A mechanism that temporarily marks a failed model as unavailable for a set period, preventing repeated attempts to use it immediately after an error.

A technique where high-level modules don't directly depend on low-level modules; instead, both depend on abstractions, making the system more flexible.

A mechanism that prevents multiple parts of a program from accessing the same file at the same time, avoiding data corruption.

A system's ability to automatically switch to a backup model or service when the primary one fails, ensuring the conversation continues without interruption.

A design pattern that isolates the core logic from external systems using ports and adapters, making it easy to swap out databases or APIs without changing the core.

A mechanism that lets plugins run custom code at specific moments in the program's lifecycle, like before a message is processed or after a model responds.

A storage system that keeps recently used data in memory and automatically discards the least recently used items when space runs out.

A design where data flows through a series of steps (functions) in a fixed order, each step transforming the data before passing it to the next.

A central list that keeps track of all available plugins and their capabilities, acting as the single source of truth for what the system can do.

A design where the main system is built around a central core, and additional features are added through plugins that extend but cannot replace core parts.

A pattern where an AI model repeatedly decides which tool to use, executes it, and uses the result to decide the next step until a final answer is reached.

A separate, isolated process that runs code with restricted permissions so it cannot harm the main system or access sensitive data.

The process of checking that data matches a predefined structure or format, ensuring messages are correctly formed before processing.

A component that saves conversation data to disk using files instead of a database, with features like async writing and locks to prevent conflicts.

A system that automatically finds and loads available tools or skills from specific folders, making them available for the AI to use.

A communication method that sends data in chunks as it becomes available, allowing the AI's response to appear gradually rather than all at once.

A central list that manages the lifecycle of smaller, specialized AI agents that can be spawned and controlled by the main agent.

The process of condensing a long conversation into a shorter version that keeps the important information, so the AI can continue without exceeding its memory limit.

The maximum number of words or pieces of text an AI model can process in one go, beyond which the conversation must be shortened or summarized.

A strategy where cached data is automatically updated after a set amount of time (Time To Live) to ensure it stays current.

A way for one service to send real-time data to another service via a URL, often used to receive messages from chat platforms.