Browse / LLM wiki / agent-wiki
agent-wiki
zsayem30
An agentic wiki dedicated for multiple research agents working on research projects, inspired by Karpathy's LLM wiki
Open zsayem30/agent-wiki → Checked against the four rules.
Works with any agent
ships AGENTS.md, the cross-tool instruction file.
In its own words
agent-wiki is a curator-owned project memory scaffold for research coding projects. It is designed to live as a subfolder inside a host project. OpenCode can use the bundled root opencode.json template; Codex and other agents can use the same memory protocol through AGENTS.md plus the Python helper scripts.
The host project owns implementation, experiments, papers, figures, and domain-specific agents. agent-wiki/ owns compact durable memory and the wiki-curator workflow.
Core Idea
Research projects often accumulate many long Markdown files: plans, paper notes, debug reports, experiment logs, and outdated summaries. Future agents then waste time rereading stale material.
agent-wiki separates the surfaces:
The compact wiki is not a dumping ground. It is the distilled project memory that tells agents what to read, what is true now, what has been tried, and what should happen next.
Documentation Site
This repository includes a Sphinx/MyST documentation site configured for Read the Docs. Source lives in docs/; Read the Docs uses .readthedocs.yaml and docs/conf.py.
Install In A Host Project
From the host project root:
If your project already has opencode.json or AGENTS.md, merge the template sections instead of overwriting them.
OpenCode Workflow
Start OpenCode from the host project root:
The root template configures two primary agents:
Quoting 34% of the README. Read the rest on GitHub ↗
Read from the repository's git tree on 2026-08-17.