OpenAlgo posted a tweet: "OpenAlgo now has a brain."

It's not Obsidian, it's not a vector database. It's a real-time knowledge graph built on Google's Open Knowledge Format (OKF). The knowledge is stored entirely as plain Markdown, which any AI agent can read and traverse.

The video demonstrates a node graph where various concepts are connected by relationships to form a network. Every node is a Markdown file, and AI can navigate through edges to jump between nodes.

OKF's code and specification are already open source on GitHub: https://github.com/marketcalls/openalgo/tree/main/okf

When one user asked if there was a team supporting retail customers with setup, OpenAlgo replied that the getting started guide is designed for complete beginners, and most people can set it up themselves after watching the tutorial. The guide is available here: https://openalgo.in/getting-started

Another user, Tom Ballard, mentioned a project called Lore, which he says was released a few weeks before OKF was published, converged on a similar design to OKF while adding formal structure. Lore's repository is here: https://github.com/itsthelore/rac-core

What is Lore, exactly? In one sentence: it structures your team's existing decisions, requirements, designs and roadmaps as typed Markdown files stored in your repository, and exposes them read-only to AI agents like Claude Code and Cursor via the MCP protocol. There's no RAG, no vector embeddings, no model inference to determine relevance — retrieval is deterministic and reproducible.

Lore's core engine is called RAC (Requirements as Code), which enforces validation at write time: `rac validate` and `rac gate` will reject incorrectly formatted files, broken links, and entries referencing obsolete decisions in CI. By contrast, OKF only defines the storage format (Markdown + YAML frontmatter in a Git tree) and leaves validation up to consumers. Lore's `rac export --okf` can convert any RAC repository into an OKF-compliant package — the two projects are complementary.

Looking at these two projects side by side is fascinating.

One is a knowledge graph built for a quantitative trading platform, enabling AI agents to understand relationships between trading strategies, market data, and configurations. The other is a decision logging system designed for software engineering teams, that keeps coding agents from forgetting architecture decisions you already locked in.

But their core logic is identical: **Use human-readable, machine-parsable Markdown as the knowledge carrier, organize relationships with graph structures, so AI no longer has to rely on fuzzy vector similarity, and can instead reference and traverse knowledge precisely.**

This isn't a completely new idea. Programmers have been writing ADRs (Architecture Decision Records) for years, but historically these were written for human colleagues. Now agents are the new audience, so we need a format they can read too.

OKF and Lore are both working toward the same goal: freeing knowledge from black boxes (vector databases, RAG pipelines) and turning it into plaintext, auditable, version-controlled files.

For AI practitioners, this means you no longer need specialized infrastructure to run a knowledge graph. A Git repository, a collection of Markdown files, and a simple parser are all you need.

For casual users, this may sound abstract. But you can think of it this way: Previously, when you fed data to AI, it might misremember, mix up information, or fail to find what it needs entirely. Now you write your knowledge as a map, AI follows the map step by step, and can tell you exactly which file it read and which decision it referenced at every step.

Isn't that more reliable?

Of course, both projects are still young. OpenAlgo's OKF just launched, and Lore describes itself as early and evolving quickly. But the direction is already clear: **Standardization of knowledge graphs is underway, and Markdown is the standard.**

If you're working on AI agent-related projects, it's worth checking out both repositories. They demonstrate a compelling possibility: letting AI stop "guessing" and start "reading"