Meta's new Muse AI companion uses a configuration file called Soul.md to define its personality and behavior, following the same markdown-based pattern that OpenClaw, an open-source personal agent project, established for AI agent identity. The discovery, made public on September 13, 2026, by Peter Steinberger, OpenClaw's creator, highlights a broader shift in how developers and companies are now building AI agents: instead of burying identity and behavior deep in code or system prompts, they're writing them into plain-text files that humans can read, edit, and version control.

What Is Soul.md and Why Does It Matter?

Soul.md is a markdown file inside Meta's Muse app that configures the AI companion's persona and identity. Unlike traditional approaches where an AI system's behavior was hardcoded into its underlying model weights or buried in a long system prompt, Soul.md makes that identity explicit and human-readable. The name itself signals Meta's intent: where a coding agent might have a file describing technical conventions, a consumer AI companion has a file describing its "soul".

Steinberger, who works directly on agent identity files for a living, spotted the file and posted about it on X (formerly Twitter) with the message: "Anyone got an invite code for Meta's Muse? I saw their Soul.md file and now i'm curious." The post accumulated 235,000 views by lunchtime, followed by a second message: "all sorted out. they defo been cookin!". That reaction from a builder who understands the technical landscape is significant; it suggests Meta's implementation is thoughtfully designed.

Is This Pattern New, or Has It Been Around?

Soul.md is not a new idea in the agent-tooling world. The underlying convention has been established for years, but it's now spreading from developer tools into consumer products. Similar patterns already exist in the coding-agent space:

• CLAUDE.md: A project-level file that gives Claude Code persistent instructions about your codebase's stack, conventions, and what to remember between sessions.
• AGENTS.md: A vendor-neutral, cross-tool version that allows Codex, OpenCode, Gemini CLI, and other agent harnesses to read one committed file instead of multiple vendor-specific configurations.
• Other specialized files: SKILL.md, MEMORY.md, DESIGN.md, and WORKFLOW.md each carve out a specific slice of an agent's behavior into human-readable, version-controllable text files.

What makes Soul.md noteworthy is that it extends this pattern to a new category: a consumer AI companion's persona, rather than a coding agent's engineering conventions. The mechanism is the same (a plain-text file the runtime reads), but the job is different. Where CLAUDE.md answers "how should this agent write code in this repo," Soul.md answers "who is this agent, personality-wise?".

What's the Connection to OpenClaw?

One of the most specific claims in the discussion came from Michael Guo, who noted that Muse ships "almost the exact same 5 Markdown files" as OpenClaw. Guo also suggested that Meta, led by Mark Zuckerberg (nicknamed "Little Z" in the post), reportedly tried and failed to acquire OpenClaw outright but nonetheless converged on a similar configuration pattern. However, this claim remains unverified beyond Guo's own observation through his invite access to Muse.

What can be confirmed with confidence is narrower but more useful: the underlying convention Guo described (splitting an agent's identity across multiple named markdown files rather than cramming everything into one system prompt) is a real, well-established pattern in 2026 agent tooling. Muse having a file literally named Soul.md is consistent with Meta adopting some version of this approach.

How to Apply This Pattern to Your Own AI Agents

• Write identity in plain text: Instead of burying agent behavior in code or a single long prompt, create a dedicated markdown file that describes who the agent is, what it should do, and what boundaries it should respect. Make it human-readable so anyone on your team can understand and modify it.
• Split concerns across multiple files: Rather than putting everything into one document, separate identity from memory, capability from tone. This makes each file easier to maintain, version control, and reason about independently.
• Version control your agent's personality: Treat agent identity files like you treat code. Commit them to git, review changes in pull requests, and maintain a history of how your agent's behavior has evolved over time.

Why This Shift Matters for the Future of AI

The headline here isn't really that Meta Muse has a file called Soul.md. The real story is that the practice of giving an AI agent its identity in a plain, human-readable markdown file has now crossed from developer tooling into a consumer product used by millions of people who will never open a terminal. A year ago, "prompt engineering" mostly meant crafting one long system prompt. In 2026, the more durable pattern is to split identity, memory, and capability into separate, named, versionable files.

This represents a maturation in how the industry thinks about AI agent design. Markdown-as-config is now genuinely cross-category. CLAUDE.md and AGENTS.md answer "how should this agent behave in this codebase." Soul.md answers "who is this agent, personality-wise." Same mechanism, different job. The lesson for any builder is the same: write down identity and behavior in something a human can read and diff, not just in a prompt buried in code.

The fact that Meta adopted this pattern for a consumer product suggests it's becoming the default way to specify agent identity, regardless of which company's product you're using or whether you're building a coding assistant, a customer-facing companion, or an internal operations bot.