Agent-Base
A Chinese-language, source-grounded research handbook that dissects seven coding-agent products, such as Claude Code, Codex, Gemini CLI and OpenCode, comparing their agent loops, context handling, tools, safety and multi-agent orchestration.
Agent-Base is a research handbook, read online, that studies coding agents by tracing their actual source code rather than their marketing. It compares seven product families across eight implementation surfaces, including Claude Code, Codex, Gemini CLI, Kimi CLI and Code, OpenCode, SWE-agent and Qwen Code, and marks each conclusion as either verified against source or inferred.
It organizes the research into a consistent set of about thirteen facets applied to every project, such as CLI entry, session runtime, agent loop, tools, MCP integration, memory and context, UI interaction, and multi-agent orchestration through subagents, teams and swarms, so a reader can either study one product end to end or compare how every product handles the same facet. A cross-project architecture atlas reads the same questions across all seven families side by side.
The GitHub repository holds the research documents and a source index; the material is meant to be read through its own hosted site rather than the raw repository.