For software teams using multiple AI coding agents

Keep every coding agent aligned.

Write your project instructions once, review the exact file each tool will receive, and let every teammate connect the targets to their own development machine.

  • One shared sourceMaintain named, reusable instruction blocks with your team.
  • Tool-correct outputsRender tested files for Codex, Claude Code, Gemini CLI, Copilot, Cursor, and Ollama.
  • Local controlEach person chooses where each target file lives in their own browser.
Start guided setup Sign in Contact us about the pilot

No command-line installer is required. Start from a template, import one existing instruction file, or begin empty.

Coggy want to help

Coggy, the LLMConfig AI assistant, welcoming you

Compatibility

Six tool profiles, rendered from the same rules

Each profile has a documented filename, repository suggestion, output format, and tested renderer.

Tool Suggested repository file Available delivery
Codex / AGENTS.md Repository instructions for Codex and other tools that read AGENTS.md. AGENTS.md Browser file association or download Codex can also use its dedicated MCP sync destination.
Claude Code Repository instructions for Claude Code. CLAUDE.md Browser file association or download
Gemini CLI Repository instructions for Gemini CLI. GEMINI.md Browser file association or download
GitHub Copilot Repository-wide custom instructions for GitHub Copilot. .github/copilot-instructions.md Browser file association or download
Cursor project rule A Cursor project rule with YAML frontmatter and Markdown instructions. .cursor/rules/llmconfig.mdc Browser file association or download
Ollama Modelfile An Ollama Modelfile with the selected instructions as its system prompt. Modelfile Browser file association or download

LLMConfig generates and writes the target content; it does not claim to control the coding tool itself. Direct browser writing depends on browser file-system support, and download remains available as the fallback.

Concrete output

One rule becomes the right file for each tool

The examples below are generated by the same renderers used by onboarding, previews, downloads, and sync.

Shared block

Verify changes before sharing

Category
Instruction
Activation
Always

Run the relevant tests before proposing a change.

Codex / AGENTS.md AGENTS.md 
# AGENTS.md

## Metadata semantics

Category:
- Safety: Constraints, permissions, and things the model must avoid.
- Domain Context: Project facts, architecture, naming, business context, ownership boundaries, and conventions.
- Instruction: Task behavior and workflow rules.
- Reference: Background material the model can consult when relevant.
- Model Configuration: Model or runtime tuning notes.

Topic:
- Topic identifies the block's primary subject area.

Activation:
- Always: Apply this block whenever it is relevant to the task.
- Situational: Apply this block only when its Condition matches the current task.
- Reference: Use this block as supporting context, not a direct command.

Condition:
- Condition describes the task context where a Situational block applies.
- If Condition is absent, use Activation as the only applicability signal.

## Instruction

### Verify changes before sharing

> Activation: Always

Run the relevant tests before proposing a change.
Claude Code CLAUDE.md 
# CLAUDE.md

## Instruction

### Verify changes before sharing

Run the relevant tests before proposing a change.
Cursor project rule llmconfig.mdc 
---
description: "Project instructions managed by LLMConfig."
globs: ""
alwaysApply: true
---

# llmconfig.mdc

## Instruction

### Verify changes before sharing

Run the relevant tests before proposing a change.

Product walkthrough

From existing instructions to a team-ready workspace

A short, accessible walkthrough of the actual first-run flow.

  1. Choose your starting point

    Import AGENTS.md, CLAUDE.md, GEMINI.md, Copilot Markdown, or a Cursor rule. You can also use a starter template or begin empty.

    Import instructions Use a template Start empty

  2. Review blocks and every target

    Edit titles, categories, activation, conditions, and content. Then inspect the rendered files and the source diff before anything is saved.

    3 blocks reviewed 6 target previews ready Nothing saved yet

  3. Invite a teammate

    Send an optional Editor or Viewer invitation. Each accepted teammate gets the same workspace but keeps independent local file associations.

    teammate@example.com Editor Invitation ready

  4. Connect or download each file

    Choose local destinations in a supported browser, use the download fallback, or connect Codex through its dedicated MCP workflow.

    AGENTS.mdSynced in this browser CLAUDE.mdChoose a local file

Where your data goes

Shared instructions in LLMConfig, local destinations under your control

  1. 1Your team edits shared blocksWorkspace content and target settings are stored in LLMConfig.
  2. 2Tested renderers build each targetThe same renderer output powers preview, download, browser sync, and MCP reads.
  3. 3Your browser receives the contentOnly target content and an opaque browser identifier are exchanged for browser sync.
  4. 4You choose the local fileThe local path and file handle remain specific to that person and browser.
A write in one browser does not mark another browser, teammate, or Codex destination as current.

First pilot

Try LLMConfig with a real multi-agent project

We are inviting a small number of software teams to test the full workflow and tell us where it creates—or fails to create—real value.

Contact us about the pilot Start guided setup