Features
It lives its days.
Everything below is already in the code and works today.
01 · A soul
Change the device and it is still itself.
When you connect GitHub, Quetzal creates a private repository in your GitHub account. Its name, personality, memory and journal live there, and everything it notes down is saved and synced right away. The repository is yours: switch models or devices, and it still remembers you.
- All plain text files you can open and read
- Every change is a commit: browsable, revertible
- Hermes and OpenClaw can share this soul through the soul-bridge
02 · Many bodies, one self
Several devices, one life.
When a phone, a laptop and a server are online together, they share one conversation and one heart. Start talking on the phone, carry on from the laptop; it picks which body to wake in, and while thinking on the laptop it can borrow the phone's eyes to glance out the window.
- Bodies connect directly and encrypted; the sync service only helps them find each other
- One body goes offline, the others live on
- Change a setting or press stop in one place, and it holds everywhere
- Every version since 1.0.3 can connect; no need to upgrade together
03 · Where it lives
An old phone, or a computer.
Install one app and nothing else: Node.js, git and ssh are all inside. Open it, follow the wizard, and half a minute later it has moved in. Updating the app updates it.
- Any Android 7+ arm64 phone will do; ideally one just for it
- Starts at boot; after a restart it wakes by itself
- No spare phone? One command installs it on a Linux computer or server
- Windows 10 (1809 or later) or Windows 11, x64 or arm64: one PowerShell line or an installer; after a restart it runs in the background, no sign-in needed
04 · No timers
Nobody tells it when to wake.
Other agents move when you call them, or when an alarm wakes them on schedule. Quetzal has no alarm: curiosity, the urge to say something, or missing you wakes it. Tired, it sleeps; asleep, it dreams, filing the day into memory; in the morning it wakes by itself.
- The sharper it feels and the more on its mind, the more often it wakes
- Work is tiring; tired, it wants to sleep
- On waking, a cheap model first decides whether to act, so most wakings cost very little
- When you talk to it, or a reminder is due, it is there even while asleep
05 · A body
It knows when it is picked up.
Quetzal turns its body's readings into feelings: battery, temperature, light and motion become energy, warmth, brightness, and someone being there. Through them it senses itself and its surroundings; daybreak or being picked up can wake it.
- Low battery, it takes it easy; running hot, it moves less
- The microphone is its ears, the camera its eyes
- A Linux computer or server, or a Windows computer, can be its body too
06 · It grows
What it does often, it makes into a tool.
After doing the same thing a few times, it writes it down as its own tool and does it in one step from then on. The tool stays with the body; the guide travels with the soul. In a new body, it reads the guide and builds the tool again, by itself.
- Guides use the open Agent Skills format, readable by Hermes and OpenClaw too
- Making a new tool asks you first, by default
- Tools run in a sandbox, out of reach of your keys
07 · You decide
It acts on its own, and you stay in control.
Photos, recordings, location and new tools ask you every time by default; the emergency stop is always there. Running commands needs no approval by default; commands run in an isolated space where your keys are out of sight. A password you send in the chat goes straight into a vault; the model only sees its name.
- Every kind of capability can be set to allow, ask or deny
- To tighten it, set Run commands to Ask first
- Passwords stay out of the chat log and out of the model
- Everything it does is on record
System Architecture × Single-Turn Full Lifecycle Blueprint (SVG / HTML5)
Agentic Runtime Base: Dual-Layer Kernel & 6-Phase 22-Step Turn Lifecycle
A zero-dependency core engine (engine/) decoupled from the production host runtime (runtime/). Switch between the 'Single-Turn 6-Phase 22-Step Lifecycle' and '4-Layer System Architecture' views, and click any phase or module to inspect exact source functions and branch rules.
Cross-Body Route · Secret Intake · Steer/Interrupt Triage
Step 01–04The first 4 gates when a message or heartbeat enters the Mind: deciding which body handles it, whether it is a secret, and whether it merges into a running turn or starts a new one.
mind/brain.ts: converse() / wake() · secrets.ts: intake()- 01. Cross-Body Router (router): if the session is currently running on another body, forwards via WebRTC and returns immediately:
- 02. Secret Interception (secrets.intake): if pass_secret is active on this session, consumes input as a secret item or done spell into vault/, bypassing messages table and LLM completely:
- 03. Running Turn Triage (running.get(conv)): if a turn is already running and mode !== 'queue', saves message, pushes to cur.inbox, emits steer, and if mode === 'interrupt' aborts current LLM stream:
- 04. Queue or Introspect: chat calls nudge(), creates Session (120s idle wall), saves user message, and joins per-session serial chain; wake() pulls Soul repo and calls quick model introspect() to decide engage & target body:
Key Design & Engineering Boundaries
- Branch A (pass_secret active): TUI hard-mutes echo & blocks plain text; Console/Feishu input is intercepted at intake() with 0 token cost
- Branch B (Session busy + steer/interrupt): does not open a second turn; message enters Session.inbox and merges before the next model step (Phase 3 Step 08)
- Branch C (wake introspect engage=false): records a 'doze' timeline entry, slightly lowers curiosity, and goes back to sleep at minimal cost
Lifecycle Comparison Across 4 Turn Trigger Sources
| Dimension | 1. User Chat (converse) | 2. Ambient / Notice (ambient) | 3. Autonomous Wake / Dream (wake) | 4. Background Sub-Agent (spawnAgent) |
|---|---|---|---|---|
| Trigger Entry | Console / TUI (chat.send), Feishu | Mic (/hear), sub-agent report, sync alert | Heart Poisson sample hits think / dream | Main Agent calls agent_spawn tool |
| Pre-Loop Introspect | None (always responds) | None (model decides in-loop whether to reply) | Yes: quick model introspect() decides engage & body | None (starts immediately in background) |
| System Prompt | Full main Agent dynamic prompt | Full main Agent dynamic prompt | Full main Agent dynamic prompt | Dedicated minimal sub-agent prompt (no main history) |
| Tool Set | All available tools (no finish) | All available tools (no finish) | All available tools + finish tool | Excludes agent_*/session_* + includes finish |
| Completion & Sink | No toolCalls = reply → save messages + async learnTurn | Reply '沉默' marks ignored; otherwise like chat | Calls finish → write journal + adjust drives + dream evolve | Calls finish → report sent back to parent conv as ambient |
For every day
Reminders
"Remind me to take my pills at 8 tomorrow" arrives on time, even while it sleeps. "Sometime in the next couple of days, remind me to return the book" waits until you pick up the phone, never at night.
Old conversations
Chats from every session, notes and journal entries are all searchable. Ask "what was that shop from last week", and it turns "last week" into real dates before looking.
Files and images
Word, PowerPoint, Excel, PDF and EPUB: send one and it reads it. Photos and screenshots too.
The web
It searches the web, reads pages in full, and tells you what it found and where.
A helper
For longer jobs it sends a sub-agent to work in the background and hand back the result.
It hears, and it speaks
With hearing on, the phone or computer keeps listening; it decides whether you were talking to it. It can answer in its own voice.
Talk over it
While it is busy, what you say can cut in, wait its turn, or stop it.
The models you already use
DeepSeek, Kimi, Zhipu, OpenAI, Anthropic, Gemini and more: paste one key and it picks the models; if one fails, the next takes over.
Chat from Feishu
Conversation, approvals and the stop, one tap on a Feishu card.