About Pathrule: the context layer for AI coding teams

What Pathrule is

Pathrule is a shared AI coding context layer for engineering teams that use AI coding assistants. It stores three kinds of team knowledge in a path-indexed graph that mirrors your repository's folder structure: memories (what you've learned), rules (what to do or avoid) and skills (how to perform multi-step work).

When an AI assistant is about to act on a file, Pathrule runs at the hook layer that already exists in tools like Claude Code, Cursor, Codex, Windsurf and GitHub Copilot, and injects only the slice of knowledge that matches the path. The model never sees the parts of your team's rulebook that do not apply.

Same knowledge, every tool. A rule written for /apps/mobile applies whether the developer is using Claude Code today, switches to Cursor tomorrow, or runs Codex from the terminal. Pathrule is the layer that survives the tool churn.

Why we built it

Every team that adopts an AI coding assistant eventually hits the same wall. The CLAUDE.md, AGENTS.md or .cursorrules file that worked when it was 40 lines starts breaking down at 400. The model ignores parts of it. Token budgets balloon. New teammates do not know what is in it, and old teammates do not know what is stale.

The pattern shows up in very different shapes: a one-developer side project, a backend monorepo adding its tenth service, an agency juggling client work. The team changes; the problem does not. Team knowledge was being modelled as one giant file when the work itself is path-scoped.

Pathrule started as the answer to one question: what if rules lived next to the paths they apply to, and the model only saw the ones it needed?

How it works

• Author. Write memories, rules and skills in the web app, desktop app or CLI. Each is attached to a path in your workspace.
• Route. Pathrule indexes the graph by path. The relevant subtree is precomputed for fast lookup at hook time.
• Inject. When your AI tool fires a PreToolUse or UserPromptSubmit hook, Pathrule emits the path-scoped slice as JSON.
• Survive tool churn. The same content works for Claude Code, Cursor, Codex, Windsurf and GitHub Copilot through one MCP server. New tools do not reset team memory.

For the technical version, see the core concepts and the how hooks work docs.

What Pathrule is not

• Not another AI coding tool. Pathrule does not replace Claude Code, Cursor or Codex. It is the layer that makes any of them smarter about your team.
• Not a code indexer. Pathrule does not read, scan or upload your source files. It stores the typed content your team writes.
• Not a chat product. There is no Pathrule chat UI competing for your prompts. The AI tool you already use stays in charge of the prompt.
• Not a model provider. Pathrule does not run inference on your codebase or train models on your content.

Founder

Sertan Helvacı, Founder

Sertan spent years writing code, then moved into product design and worked formally as a product designer. Pathrule comes out of both sides of that path. It is not a vibe-coding side project.

Design has always had a design-system principle: shared tokens, shared components, shared rules that travel with the work. Pre-AI software teams had a quieter version of the same thing in shared conventions, shared review and shared memory. The first wave of AI coding assistants broke that quiet system. The rules a team used to share with itself collapsed into a single instruction file that the model half-followed and the team half-trusted.

Pathrule is what comes out of taking those older principles seriously again, for the AI coding era. He writes about the problem from a team's point of view in the Pathrule writing.

[email protected]

Trust and privacy

Pathrule never reads, scans or uploads your source code. The cloud stores the typed memories, rules and skills your team writes, plus the structural metadata needed to route them. Everything else stays on your disk.

Every claim on that boundary is independently verifiable in a couple of minutes. The hook script is a single unobfuscated Node.js file. The MCP tool surface is enumerable with a single tools/list call, so you can confirm there is no file-reading tool. The network surface is short and auditable. The full breakdown lives on the Security page.

FAQ

What is Pathrule? Pathrule is a context layer for AI coding teams. It stores path-scoped memories, rules and skills in a knowledge graph and injects only the slice that matches the current working directory into Claude Code, Cursor, Codex, Windsurf and GitHub Copilot at hook time, before the first tool call.

Who is Pathrule for? Engineering teams that already use AI coding assistants and feel the pain of AGENTS.md or CLAUDE.md files growing past the point where the model can follow them. Small teams, startups and agencies who want shared rules and memory across tools without uploading source code.

How is Pathrule different from CLAUDE.md or AGENTS.md? AGENTS.md and CLAUDE.md are single, flat files that ship every instruction on every turn. Pathrule is a routing layer above them: you store knowledge once, in a path-indexed graph, and only the relevant subtree is delivered into the model's context for the current task. Long read: AGENTS.md vs Pathrule.

Does Pathrule see my source code? No. Pathrule never reads, scans or uploads your source files. The cloud stores only the typed content your team writes (memories, rules and skills) and structural metadata. Source code stays on your disk. See the Security page for verifiable details.

Which AI coding tools does Pathrule support? Claude Code, Cursor, Codex, Windsurf and GitHub Copilot today, over a single MCP server with hook-time injection. The same memories, rules and skills work across all of them.

How do I get started? Read the quickstart, then install Pathrule Studio or CLI. Solo developers can use Pathrule for free.

Contact

• General: [email protected]
• Security: [email protected]
• Privacy: [email protected]