Keen Code: Open Source Coding Agent
AI-powered coding assistant developed in Go now available.
1What is Keen Code?
Keen Code is an AI assistant developed from scratch in Go by mochow13, designed for developers' daily tasks. With its minimal and simple UI, the tool is open-sourced on GitHub. It supports multiple providers and can be used for real software engineering work at production level, offering features expected from a useful coding agent.
2Turn Memory feature
A standout feature of Keen Code is Turn Memory, which removes tool outputs in multi-turn conversations, retaining only tool call traces. This approach prevents unnecessary tool results from filling up the context window in subsequent turns. As a result, the context window fills up much more slowly compared to other agents during multi-turn conversations.
3Context optimization
The biggest advantage of Turn Memory is that the context window can reduce to as low as 1%. However, there's a tradeoff: if the agent needs a result from a previous tool call, it won't have access to it. The developer notes that tool calls like read, bash, and web_fetch are inexpensive, and files can be re-read if needed - similar to how Claude Code or Codex frequently re-reads files.
4Skill-Driven MCPs
Another innovative feature is Skill-Driven MCPs (Model Control Programs). In this approach, each MCP server receives a skill, but unlike traditional MCP usage guidance, these skills are generated by Keen. The major advantage is that no server comes pre-loaded with complete tool schemas by default.
5How MCPs work
In the Skill-Driven MCP system, the agent only receives skill frontmatter for the server. When a server is needed, the agent loads the full skill file listing tool descriptions, then reads the specific schema file for the tool it wants to invoke. While each MCP call requires a file read operation, everything is cached locally upon discovery, making this approach practical.
6Other optimization efforts
Beyond Turn Memory and Skill-Driven MCPs, the developer is experimenting with other well-known context optimization ideas like hashline edits. These innovations aim to make AI assistants more efficient with their context windows. Being open source, the project welcomes community contributions to further enhance these features.
7Usage and contributions
Keen Code's CLI usage guidelines are available in the project documentation. Written in Go, the project has matured enough for real production workloads. The developer uses it both for actual work and for its own development. Community feedback and contributions are actively encouraged, with the project hosted on GitHub for easy collaboration.
8Impact on software development
The proliferation of open-source AI assistants like Keen Code could significantly boost productivity in software development. Context window optimizations in particular may enable large language models to work with longer code segments more effectively. The open-source nature suggests ongoing community-driven development and future feature enhancements.
Original article: Hacker News: Show HN — Keen Code: Açık Kaynak Kodlama Asistanı