Blog
OKF 0.2 explained: the Open Knowledge Format and validation
OKF 0.2 explained: the Open Knowledge Format spec, its frontmatter fields, the index.md and log.md rules, and how to validate an OKF 0.2 bundle with IWE.
Read more8 min read
November 8, 2026
Your agent hates walking your knowledge graph. So stop making it walk.
The N+1 criticism of knowledge graphs as agent memory is right, but it is a property of traversal in the agent loop, not of graphs. Move the walking into the engine.
Read more5 min read
September 1, 2026
Memory that lives in your project
The design behind the IWE plugin for Claude Code: memory stored in the repository as plain markdown, one graph for you and your agents, nothing written unasked.
Read more11 min read
August 30, 2026
Designing edit operations for AI agents
Four lessons from designing IWE's block-edit language for AI agents: declared blast radius, identity as a constraint, recoverable failures, and errors that teach.
Read more6 min read
August 1, 2026
The benchmark that built the tools
A plain-Markdown knowledge graph as agent memory, measured on LOCOMO: it lost to grep, rebuilt our tools, and ended at parity with published memory systems.
Read more35 min read
July 12, 2026
We benchmarked IWE on 20,000 documents. Here's what we found.
We generated a realistic 20,000-document Markdown knowledge base and measured load and query time: loading under a second, most queries under 20 milliseconds.
Read more3 min read
May 3, 2026
Structured questions need a query layer, not better retrieval
Full-text retrieval answers 'tell me about X' but fails at 'how many X are in state Y.' The fix is a frontmatter query layer the agent filters, counts, and sorts.
Read more6 min read
May 3, 2026
Markdown database: notes as memory for humans and AI agents
Treat a folder of Markdown notes as a queryable graph database: file-path keys, a frontmatter schema, inclusion and reference edges, and one filter language.
Read more9 min read
May 2, 2026
Five components of agent memory in plain Markdown
The five things agent memory must do — persistence, structure, retrieval, writeback, forgetting — implemented as plain Markdown files and CLI queries, no vector database.
Read more5 min read
April 25, 2026
How I turned three philosophy books into a 1,200-document knowledge graph
Marcus Aurelius, Machiavelli, and Nietzsche use one word to mean three things. A plain-Markdown knowledge graph traces the shift across seventeen centuries.
Read more4 min read
April 17, 2026
My agent kept "remembering" things wrong. The fix was embarrassingly simple.
Why structured Markdown beats embeddings for architectural knowledge: a vector store kept retrieving the similar-but-wrong note, so I moved to CLI-navigable notes.
Read more5 min read
April 5, 2026
Markdown knowledge graph for humans and agents
Your notes and your AI agent can share one knowledge base — plain Markdown files, queried through structure instead of embeddings.
Read more5 min read
March 21, 2026