Your first persona¶
The setup wizard is the door. When no persona exists yet, the dashboard opens straight onto it; you can also reach it any time at http://localhost:5000/onboarding. It asks for the few things she needs to come into being — and deliberately not for more.
The chooser¶
The cold screen offers two paths:
- Wake a new one up — bring a brand-new persona into being. (Begin →)
- Bring one back — restore a persona from a diary file you already have, using its recovery phrase. (Restore →)
If this is your first time, choose Wake a new one up. (Restoring is covered in Operating; it needs a .diary file and its recovery phrase.)
What the create wizard asks for¶
Two things are required: her name and her Mind. Channels are optional. That's the whole form — you won't describe her, give her a bio, or pick personality traits. She shows you who she is by talking, and she learns the rest over time.
Step 1 — Her name¶
What does she answer to? Something short — you'll write it many times (e.g. Iris). A name is enough to begin.
Step 2 — Her Mind¶
The Mind is her Thinking organ — the one required model. It's what recognizes what you said, decides what to do, reflects on her day, and remembers. Pick a model that's good at structured reasoning. 14B is the practical floor — smaller models tend to produce too little usable output to be worth running; larger models work better.
The Mind picker first asks Local or Cloud:
Local — her thinking runs on this machine through Ollama; nothing leaves it. You fill in:
- Model name — the model id from Ollama, e.g.
qwen2.5:14b. (If you haven't pulled one, see ollama.ai; Eternego pulls it for you on first use.) - URL (advanced) — defaults to
http://127.0.0.1:11434. Leave it unless your Ollama runs elsewhere.
Local needs real hardware: a 14B model wants roughly 10 GB of RAM just to load, so plan on 32 GB+ for the model, her context, and the OS together. Without a GPU, expect each beat to take minutes rather than seconds. No API key, though — nothing leaves the machine.
Cloud — her thinking goes through a provider's API: fast and powerful, but the conversation passes through their servers. Pick a provider tile, then fill in the form:
| Provider tile | Pre-filled URL |
|---|---|
| Anthropic | https://api.anthropic.com |
| OpenAI | https://api.openai.com |
| Gemini | https://generativelanguage.googleapis.com |
| xAI | https://api.x.ai |
| DeepSeek | https://api.deepseek.com |
| Together | https://api.together.xyz |
| OpenAI-compatible | (you enter the endpoint) |
For each cloud provider you provide:
- URL — pre-filled for the named providers; for OpenAI-compatible (LM Studio, vLLM, LiteLLM, Groq, OpenRouter, Fireworks, and any other OpenAI-style endpoint) enter the base URL yourself. Use the base endpoint only — Eternego appends
/v1(or/v1beta, etc.) itself, so don't include it or the call 404s. - API key — stored locally, on this machine only. (It is written to her
config.jsonin clear text — see config.json.) - Model — the model id exactly as the provider expects it, e.g.
claude-sonnet-4.5.
Once name and Mind are filled, Wake her up becomes active.
Optional — Channels¶
You can connect Telegram or Discord now, or leave it for later (Settings). She's always reachable in this browser regardless.
- Telegram — create a bot via @BotFather, copy the token he gives you, paste it.
- Discord — create an application at discord.com/developers/applications, add a bot, copy its token, paste it.
A connected channel still needs pairing before she'll listen to it — see Talk to her.
Her other organs come later¶
The create wizard asks only for her Mind. Her other senses — Eye (vision), Mouth (voice), Ear (hearing), Imagination (drawing), Teacher (a stronger model she consults when she meets a moment she has no instruction for), and Researcher (reads documents you send) — are all optional and are added afterward, on the dashboard's Settings screen, using the same Local-or-Cloud form. A persona with only a Mind still works — for the organs that are really a thinking job, her Mind stands in: with no Teacher she consults her own thinking model when she meets a moment she has no instruction for, and with no Researcher that same model reads the documents you send. The ones her Mind genuinely can't fill are the senses and outputs — no Eye ⇒ she'll say she can't see an image, no Ear ⇒ she can't hear a voice note, no Mouth or Imagination ⇒ she can't speak aloud or draw. The full organ list is in the Vocabulary.
Many cloud Minds already include vision — if yours does, she sees through her Mind and needs no separate Eye.
The recovery phrase¶
The moment she comes to life, she hands you a recovery phrase — 24 words.
Save it. Her diary is encrypted with this phrase and nothing else — Eternego never holds a copy of the key. That's the point: the diary file can travel anywhere — a cloud backup, a USB stick, emailed to yourself — and stay unreadable to everyone but you. You hold the only key, so only you can move her to a new machine, and only you can ever read what she's learned about you. The trade for that guarantee: lose the phrase and no one, Eternego included, can bring it back — she keeps running on this machine, but can't be restored elsewhere.
The wizard makes you Copy phrase or Save as a file before it lets you continue — that's the acknowledgement step. Then click Meet \<name> to land on her view.
What happens next¶
The wizard is done; she's already running. You land on her main view, where you can talk to her right away.
Continue to Talk to her →.