Getting Started

Digital Me carries your intelligence — knowledge, taste, decisions — to every agent you run. This page takes you from a clean machine to your first capture → recall loop in about five minutes.

Prerequisites

• openclaw — mandatory. Digital Me rides on openclaw's gateway daemon; install it first and verify openclaw --version works. setup hard-stops with install guidance if it's missing.
• Node.js ≥ 22.5 and pnpm — the CLI checks and tells you if your Node is too old.
• Python ≥ 3.11 — optional; only needed for the dream-cycle distillation pipeline (skip it with --minimal).

Install

git clone https://github.com/Amyssjj/digital-me.git ~/digital-me-os
cd ~/digital-me-os && pnpm install && pnpm build

# Detect installed CLIs, scaffold ~/digital-me/, install hooks/skills/configs,
# link the `digital-me` command onto PATH, run doctor:
pnpm dm setup

# Node-only? Skip the optional services (dream-cycle venv + dashboard build):
pnpm dm setup --minimal

# Pin a custom wiki location:
pnpm dm setup --wiki-root ~/notes/brain

What setup does:

1. Detects ~/.claude/, ~/.codex/, ~/.hermes/ to figure out which runtimes you have.
2. Scaffolds the wiki root — ~/digital-me/{wiki,inbox,.cache} plus a live config.yaml (created only if absent, never clobbered).
3. Installs each detected runtime — hooks, skills, MCP entries, protocol bundles (see Runtimes for what lands where).
4. Runs doctor to confirm everything resolved.

Re-running is idempotent — installers merge into existing settings without clobbering your other hooks.

Packaging status (pre-release): install from source as above — that one pnpm dm setup is the whole build. The dream-cycle pipeline is published on PyPI as digital-me-dream-cycle; the npm package is prepared but not published yet — npm install will not find it until the first release lands.

Your first loop

The whole system exists for one loop: an agent learns something once, and every agent knows it afterwards.

# 1. Verify the brain is reachable
digital-me doctor

# 2. In any wired agent (Claude Code, Codex, Hermes, openclaw), do real work.
#    When something reusable surfaces, the agent calls learning_capture —
#    or you ask it directly: "remember this for next time".

# 3. Watch it come back: start a NEW session, ask about the same topic.
#    The runtime adapter injects matching wiki knowledge into the prompt
#    before the model ever sees your question.

Behind the scenes the capture lands in your wiki repo as a reviewable markdown entry, the nightly dream-cycle distills raw learnings into clean entries, and memory_search serves them to every runtime. Nothing is hidden in a vector store you can't read — the wiki is plain files in git.

Where things live

Next steps

• Wire up each agent CLI you use → Runtimes
• Understand the loop you just ran → How it works
• Every command and flag → CLI reference