Answers about souls, the four files, privacy, and the three forge modes.
A soul is a portable identity bundle for an AI agent — four Markdown files (soul.md, identity.md, user.md, agents.md) that encode personality, memory, voice, and writing style. Forged once, applied anywhere. For example, you complete one personality test and drop the same bundle into Claude Code, OpenClaw, Cursor, or Hermes without rewriting your prompts.
OpenClaw natively, Claude Code via CLAUDE.md, Cursor via .cursorrules, and Hermes via agent.md. Souls export to all four formats. In practice, that means a soul forged today works across the harness you use now and the one you switch to next month — no re-forging required.
About 12 minutes for the 49-question Deep Forge test. Flash Forge analyses screenshots in under 90 seconds, and Gao Dao is a multi-turn guided dialogue, typically 8–15 minutes. So if you are short on time, upload a few chat screenshots to Flash Forge; if you want the most thorough read, take the full Deep Forge wizard.
Yes — free during early access. No payment, no waitlist, no paid tier for the test. The output Markdown files are yours forever, regardless of future pricing. For example, even if a paid plan launches later, the four files you generated today remain downloadable and usable with no licence to renew.
The Enneagram captures motivations, growth-edge, and dark-side patterns that prompt-style personality descriptions miss. It maps cleanly to the four-file output: type drives soul.md, while wing and stress points shape the style guidance in agents.md. For example, a Type 5 with a 4 wing produces a soul that reasons depth-first and writes with reserved precision, not just a generic 'be helpful' instruction.
Every soul is four Markdown files, each with a distinct job. soul.md is the core personality and motivations; identity.md is the agent's name, role, and self-concept; user.md captures who you are so the agent adapts to you; and agents.md holds operating style and collaboration rules. For example, soul.md decides how the agent thinks, while agents.md decides how it formats a code review or pings a teammate.
Yes — your image bytes are processed in memory only and are never written to storage. We persist just three things: the generated report, a SHA-256 hash of the input, and the screenshot count. For example, the actual pixels of your chat history are gone the moment analysis finishes; only the personality report and a non-reversible fingerprint remain.
Yes — a soul exports to Cursor's .cursorrules and to Hermes's agent.md, alongside Claude Code's CLAUDE.md and native OpenClaw. You forge once and download the format your harness expects. For example, drop the exported .cursorrules into your repo root and Cursor picks up the same personality, voice, and operating rules as your OpenClaw agent.
They are three paths to the same four-file soul, differing in input and depth. Deep Forge is a 49-question Enneagram wizard for the most precise read; Flash Forge infers your type from chat screenshots in seconds; and Gao Dao is a high-touch guided dialogue with a mentor agent. For example, choose Deep Forge when you want rigour, Flash Forge when you want speed, and Gao Dao when you prefer a conversation to a quiz.