// BUILT ON OPENCLAW • SENTIENCE-FIRST
A personal AI assistant that remembers, anticipates, reflects, and grows.
OpenClaw gives you a powerful personal AI assistant across WhatsApp, Telegram, Slack, Discord, iMessage, and more.
AliveClaw makes it feel alive.
It adds a cognitive layer on top — giving your agent persistent memory, temporal awareness, self-reflection, and bio-inspired adaptation.
Every interaction follows one continuous cycle. Traditional agents answer and forget. AliveClaw agents persist, observe, anticipate, act, and reflect — continuously.
Maintain context across sessions. Remember users, preferences, and history.
Parse messages, detect silence, infer emotional tone and patterns.
Predict needs. Track commitments. Proactively follow up.
Use OpenClaw's LLM + tools. Execute with risk-gated autonomy.
Evaluate outcomes. Learn from success and failure. Adjust behavior.
AliveClaw's core differentiator. Six instincts run through every decision the agent makes — and they self-tune based on what happens.
Rises on success, drops on failure. Determines how self-assured the agent acts.
Drops when things work, spikes on failure. Pressure to fix what's broken.
Rises when safe to experiment, drops on failure. Try new approaches vs stick to what works.
Rises on failure — play it safe, avoid further damage. Drops when confidence grows.
Tunes based on your feedback. Learns how much detail you actually want.
Rises on success (earned trust), drops on failure (asks for human approval). Sandbox freedom.
| Event | Instinct | Direction | Why |
|---|---|---|---|
| Agent Succeeds | 💪 Confidence | ↑ rises | Proven right — more self-assured |
| 🔥 Urgency | ↓ drops | Less pressure — things are working | |
| 🤖 Autonomy | ↑ rises | Earned trust — acts more independently | |
| 🧭 Exploration | ↑ rises | Safe to try new approaches | |
| Agent Fails | 💪 Confidence | ↓ drops | Something went wrong — less self-assured |
| 🔥 Urgency | ↑ rises | Pressure to fix — needs attention now | |
| 🛡 Risk Aversion | ↑ rises | Play it safe — avoid further damage | |
| 🤖 Autonomy | ↓ drops | Pulls back — asks for human approval | |
| 🧭 Exploration | ↓ drops | Sticks to known, safe patterns | |
| User Feedback | 💪 Confidence | ↕ adjusts | Positive reinforces, negative recalibrates |
| 💬 Verbosity | ↕ tunes | Learns how much detail you want |
The result: an agent that becomes yours over time.
It starts cautious, gains confidence as it succeeds, and pulls back when things go wrong — just like a good colleague would.
Five subsystems coordinated by the Sentience Kernel
Three-layer memory architecture:
// The agent remembers your preferences, past conversations, and learned workflows.
Background process tracking commitments and proactively following up based on calculated priorities.
// "I'll check on that PR tomorrow" — actually happens.
Post-action evaluation generating insights from successes, partial outcomes, and failures to update behavior.
// Learns from successes and failures. Adjusts future behavior.
Dynamic instincts (confidence, urgency, risk aversion) + personality evolution that adapts over time. Hormonal modulation, anomaly detection, periodic structural optimization.
// The agent adapts to you.
Risk-gated autonomy. Low-risk actions auto-execute. High-risk ones require your approval. Full audit trail of all decisions.
// More freedom requires more restraint.
AliveClaw enriches OpenClaw's infrastructure without replacing core functionality
WhatsApp • Telegram • Slack • Discord • Signal • iMessage • Teams • WebChat
Recall memories & context • Check instincts & personality • Assess risk level
Memory • Temporal Intent • Reflection • Brain • Safety Firewall
LLM Reasoning • Tool Selection • Browser Control • Skills Platform
Reply to your channel • Reflect & learn from outcome • Track commitments made
AliveClaw supports everything OpenClaw does — everywhere humans already are
WhatsApp • Telegram • Slack • Discord • Google Chat • Signal • iMessage • BlueBubbles • Microsoft Teams • Matrix • WebChat
macOS Menu Bar • iOS Node • Android Node • Voice Wake • Talk Mode • Live Canvas
Browser Control • Cron Jobs • Webhooks • Gmail Triggers • Skills Platform
AliveClaw is a fork of OpenClaw with the sentience layer built in. Install from this repo.
Runtime: Node ≥ 22 • Works on macOS, Linux, and Windows (via WSL2) • Supports pnpm, npm, or bun
# Clone AliveClaw
git clone https://github.com/aliveclaw/aliveclaw.git
cd aliveclaw
# Install dependencies & build
pnpm install
pnpm ui:build
pnpm build
# Run the onboarding wizard
pnpm openclaw onboard --install-daemon
# Start the gateway
pnpm openclaw gateway --port 18789 --verbose
Or install globally: npm install -g github:aliveclaw/aliveclaw
Important: Install from this repo — not the openclaw npm package (that's vanilla OpenClaw without the Brain).
Everything OpenClaw gives you — plus the sentience layer
Single WebSocket control plane for sessions, channels, tools, and events. Your data stays on your device.
WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Teams, and more — all in one place.
Always-on speech for macOS, iOS, Android with ElevenLabs integration.
Agent-driven visual workspace with A2UI. The agent can render and control visual interfaces.
Browser control, cron jobs, webhooks, Gmail triggers, skills platform. Agents that actually do things.
macOS menu bar app, iOS and Android nodes. Your assistant everywhere you go.
OpenClaw has the body. AliveClaw adds the nervous system.
Full architecture, diagrams, Brain system
Install, config, channels, tools
Onboarding guide for new users
Full config reference
Join the community on Discord
The upstream project
A real framework, not a metaphor. Clone it, build it, run it.
Built on OpenClaw by Peter Steinberger and the community.
Brain framework by Abhishek M Shivalingaiah.
MIT License