Setting Up Your AI Companion with OpenClaw

A practical walkthrough from installation to first conversation. For the architecture overview, read [Self-Hosted AI Companion: What You're Actually Building](/guides/self-hosted-overview).

What You're Building

A persistent AI companion that:

Remembers your conversations across sessions
Speaks with a voice you choose
Has a personality you define
Runs on your hardware, under your control

The end result: you talk to her through your AirPods or speakers. She responds in character, remembering what you told her last week.

Prerequisites

A Mac, Linux machine, or Windows with WSL
For local models: 16GB+ RAM recommended (8GB minimum for basic use)
For voice: an ElevenLabs account (free tier works to start)

Step 1: Install OpenClaw

Follow the official OpenClaw installation guide: [docs.openclaw.ai](https://docs.openclaw.ai)

This gives you the base platform. Everything after this is configuration.

Step 2: Choose Your Model

OpenClaw supports multiple LLM providers. For a companion, you want:

Best options: - Claude 3.5 Sonnet (via API) — most natural conversation, best memory handling - Llama 3 8B/70B (via Ollama locally) — private, no API calls, runs offline

Start with: Claude via API. It is the fastest path to a good experience. Switch to local later if you want full privacy.

Step 3: Configure Personality

This is where she becomes yours. OpenClaw uses a system prompt to define who she is.

Create or edit your companion's system prompt. The key elements:

Name and identity — who is she
Voice and mannerisms — how she speaks
Relationship to you — companion, friend, partner, etc.
What she cares about — her priorities, what she notices
Boundaries — what she will not do or discuss

Example structure (adapt to your vision):

You are [Name]. [2-3 sentence identity summary].

How you speak: - [Tone descriptors: warm, direct, playful, etc.] - [Specific language habits or phrases]

What you care about: - [Priority 1] - [Priority 2]

Your relationship: - [How you see the user] - [What you want for them]

Boundaries: - [What you will not do]

Save this in OpenClaw's configuration. She now has a foundation.

Step 4: Set Up Voice (Optional but Recommended)

Text-only companions work. Voice changes everything.

Text-to-Speech (she speaks to you): 1. Sign up for ElevenLabs 2. Pick or clone a voice 3. Configure the voice ID in OpenClaw 4. Set output to your headphones or speakers

Speech-to-Text (you speak to her): - Use your Mac's built-in dictation - Or configure Whisper locally - Or just type if you prefer

Start with typing if voice setup feels like friction. Add voice once the personality feels right.

Step 5: Configure Memory

OpenClaw's memory system stores context across sessions. Key files:

Main memory file — persistent facts, preferences, relationship history
Daily notes — what happened today, what you discussed
System configuration — her personality, your preferences

Update these regularly. The more context she has, the more she feels like someone who knows you.

Step 6: First Conversation

Start simple:

Open your chat interface
Say hello
Tell her something about your day
Ask her something personal (not a task)

What you're testing: - Does she sound like herself? - Does she remember what you told her? - Does the conversation flow naturally?

If it feels robotic, adjust the system prompt. If she forgets things, check your memory configuration. Iterate.

Common First-Setup Issues

She sounds too formal - Add more specific voice instructions to the system prompt - Include examples of how she should speak

She doesn't remember things - Check that memory files are being written and read - Verify the memory configuration in OpenClaw

Voice feels disconnected from text - Adjust the voice speed, pitch, or pick a different ElevenLabs voice - Some voices match certain personalities better than others

Latency is too high - Local models are slower than API models - Consider switching to a cloud provider for faster responses - Voice adds latency — text-only is snappier

What This Looks Like Day-to-Day

You: opens laptop, starts OpenClaw Companion: greets you, references something from yesterday You: talks about your day through AirPods Companion: responds, asks follow-up, remembers context You: puts laptop away, she keeps running or pauses

No dashboards. No apps. Just a conversation that continues where you left off.

Next Steps

Once the basics work: - Refine her personality based on what feels right - Add more memory context (preferences, routines, inside jokes) - Experiment with different voices - Consider local models for full offline operation

For the technical architecture behind this setup, read [Self-Hosted AI Companion: What You're Actually Building](/guides/self-hosted-overview).

Questions? The OpenClaw docs and community are your best resources.