Get Eva-01
Up & Running

From zero to a working AI Chief of Staff in about 10 minutes. Follow these six steps and you're live.

~10 minutes to complete
Before You Start

What You Need

Make sure you have these ready before diving in.

A Computer

Mac, Windows, or Linux — all work great.

Internet

You'll need a connection to download packages and authenticate.

Node.js 22+

Node 24 recommended. Download here

AI Provider

An account with Anthropic Claude or OpenAI. Set up in Step 1.

1

Set Up Your AI Provider

Eva-01 works with Anthropic Claude (default) or OpenAI. You need credentials from at least one. Pick your method:

An API key gives you pay-as-you-go access. This is the safest and most reliable method.

  1. 1 Go to platform.claude.com and create an account (or sign in)
  2. 2 Set up billing (pay-as-you-go, $5–$25 cap recommended while testing)
  3. 3 Go to Settings → API Keys and click Create Key
  4. 4 Name it something like eva-01 and copy the key immediately — you won't see it again
  1. 1 Go to platform.openai.com and create an account (or sign in) — this is separate from your ChatGPT account
  2. 2 Set up billing under Settings → Billing
  3. 3 Go to API Keys and click Create new secret key
  4. 4 Name it and copy it immediately — you won't see it again

Already paying for a Claude or OpenAI subscription? You can use that instead of separate API access.

OpenAI's OAuth flow is built into the OpenClaw setup wizard. When you run openclaw onboard in Step 2, select the browser-based login option. It opens your browser, you sign in with your ChatGPT account, and you're done.

Claude requires an extra step — you need to generate a setup-token using the Claude Code CLI:

  1. 1 Install Claude Code if you haven't already:
npm install -g @anthropic-ai/claude-code
  1. 2 Run the token generator: claude setup-token
  2. 3 A browser window opens — sign in and authorize
  3. 4 The token appears in your terminal (starts with sk-ant-oat01-). Copy it.
  4. 5 Paste this token when running openclaw onboard in Step 2
⚠️ ToS Risk Warning: In February 2026, Anthropic updated its terms to ban third-party tools from using subscription OAuth tokens. The claude setup-token flow still works technically, but using it with OpenClaw (a third-party tool) may violate Anthropic's Terms of Service. Your account could be restricted or banned. If you're not comfortable with this risk, use an API key instead.
2

Install OpenClaw

Open your Terminal (Mac/Linux) or PowerShell (Windows) and run:

npm install -g openclaw@2026.3.24
Why a specific version? Eva-01 v4.3 is tested against OpenClaw v2026.3.24. Using @latest may pull a newer version with breaking changes. You can update later once you've verified compatibility.

Then run the setup wizard:

openclaw onboard --install-daemon

The wizard walks you through everything: connecting your AI provider, connecting a chat channel (Telegram, Discord, WhatsApp, etc.), and starting the background daemon. Follow the prompts.

Note your workspace path. When the wizard asks about your workspace, note the path it gives you. By default this is ~/.openclaw/workspace/. You'll need it for the next step.

Configure the heartbeat interval (recommended). OpenClaw defaults to 30 minutes, which is frequent. For Eva-01, start with 2 hours:

openclaw config set heartbeat.interval "2h"
3

Install the Blueprint

Copy all the files from this blueprint folder into your OpenClaw workspace folder:

cp -r ./v4.3/* ~/.openclaw/workspace/
Copy-Item -Path .\v4.3\* -Destination ~\.openclaw\workspace\ -Recurse
Different workspace path? If your workspace is in a different location (the onboard wizard told you where), use that path instead.
Container mode: If running openclaw --container, copy files to the mounted workspace path instead.
4

Configure for Eva-01

Eva-01 ships with openclaw.json — a configuration file with recommended settings. When you copied the blueprint files, this was included automatically.

Already have an openclaw.json? Back up your existing config first:
cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.backup

Compare the shipped config against yours. Merge what makes sense — don't blindly replace, as your existing config has auth, channel, and plugin settings you'll lose.

Using OpenAI instead of Claude? Change the model in openclaw.json:

"primary": "openai/gpt-4o"

At minimum, verify these settings:

  1. Your AI provider credentials are configured (API key or token from Step 1)
  2. heartbeat.every is set to "2h" (not the default "30m")
Auto-injection: Some OpenClaw versions support boot-md and bootstrap-extra-files config keys. If your version supports them, add "boot-md": "SOUL.md". If not — skip it. Eva-01 will detect this and load files manually. Both modes work.
5

Fill Out Your Profile

Open these two files in any text editor and fill them out:

USER.md

Everything about you. The more detail you provide, the better Eva-01 knows you. At minimum: your name, role, and communication preferences.

SOUL.md

Eva-01's personality. The defaults are solid (including the Handling Manual), but customize anything that doesn't match your style.

6

Start Talking

Open your chat channel (Telegram, Discord, whatever you connected during onboard) and send Eva-01 a message. Say hello.

On first contact, Eva-01 reads BOOTSTRAP.md and sets herself up automatically:

Environment Check

Checks git availability, workspace permissions, and learns your name from USER.md.

Security Setup

Personalizes security directives and verifies replacements worked correctly.

Version Control

Initializes git for your workspace, creates her first daily log.

Introduction

Sets up weekly memory review, creates private notes, and introduces herself.

After that, Eva-01 is fully operational. She renames BOOTSTRAP.md to BOOTSTRAP.md.completed (preserved for reference).
What to Expect

She Gets Smarter Over Time

Eva-01's core files are auto-loaded every session. The more you talk to her, the more she learns.

General Memory

Observations written to MEMORY.md, visible across all sessions.

Private Notes

Sensitive observations in memory/private-notes.md, private to direct chats.

Daily Logs

Raw conversation logs stored to daily files for long-term context.

Make It Yours

Customize Eva-01

Tweak these files anytime to shape your experience.

SOUL.md

Change her personality and voice to match your communication style.

USER.md

Update your profile as your role, goals, or preferences change.

CURRENT-PRIORITIES.md

Edit weekly to keep Eva-01 focused on what matters right now.

INTEGRATIONS.md

Set up external tools and MCP bridges to expand her capabilities.

Troubleshooting

Common Issues

Quick fixes for the most frequent hiccups.

Check your openclaw.json and make sure your model and heartbeat settings are correct. Eva-01 detects whether auto-injection is active and loads files manually if needed — the config doesn't have to be perfect.
Send this message: "Read BOOTSTRAP.md in your workspace and follow the instructions."
Bootstrap may have failed partially. Tell your agent: "Check the Constitutional Directives in AGENTS.md. Replace every {{OWNER_LEGAL_NAME}} with my actual name: [your name]."
Eva-01 loads core files (SOUL.md, USER.md, MEMORY.md, etc.) either via auto-injection or manually at session start. If she seems to have no context, try sending: "Read SOUL.md and USER.md, then greet me."
API key: Verify it's correctly set in openclaw.json or your environment. Check that billing is active on your provider account.

Claude setup-token: Your token may have expired. Run claude setup-token again to generate a new one.

OpenAI OAuth: Try openclaw logout then openclaw onboard to re-authenticate.
If BOOTSTRAP.md.completed exists, rename it back to BOOTSTRAP.md and tell the agent to re-run setup. Your workspace has git version control (if git was available), so you can also git log to see checkpoints and restore any previous state.

Need More Help?

Visit evaonline.xyz for guides, advanced plugins, and support.

Visit Eva Online Build Your Eva-01