First Interactions

Telegram / QQ Bot / WeChat ClawBot are recommended for interacting with ESP-Claw.

Send a message on a configured IM channel and ESP-Claw replies automatically.
No reply? ➡️FAQ
Prefer serial Console? ➡️Console usage

Chat and Q&A

You can ask ESP-Claw questions directly, just like chatting with a person:

Hi, what capabilities do you currently have?
Calculate this for me: at 12% annual interest, what will 10,000 become after 3 years?

ESP-Claw supports natural language understanding, basic Q&A, calculations, and task assistance.

Web Search

ESP-Claw can retrieve up-to-date information and online resources:

Any AI news lately?
Visit https://github.com/espressif/esp-claw and check the latest progress

Besides normal search, it can read web pages and summarize or analyze their content.

Cloud Skills

ESP-Claw can fetch and install new Skills from the cloud to extend its capabilities.

Fetch Cloud Skills

Can you fetch some cloud Skills?

Install Cloud Skills

Install the weather Skill
Install the GPIO control Skill

After installation, the Skill is automatically connected to the current system capabilities.

Memory System

ESP-Claw supports long-term memory and device context records.

User Memory

My name is xxx, remember me
I like xxx

Hardware Memory

I connected a temperature and humidity sensor to your XX pin, remember that
A WS2812 LED strip is connected to GPIO4

This information will continue to be used in later conversations.

Session Management

Interrupt the Current Session

Say:

Stop
Cancel
Never mind

ESP-Claw will immediately terminate the current task or conversation flow.

Switch Sessions

ESP-Claw supports multiple sessions.

View sessions:

/session

Create a new session:

/session new

Use this to isolate different projects, devices, or scenarios.

Scheduled Tasks

ESP-Claw supports recurring and one-shot task scheduling.

Recurring Triggers

Send me a message every five minutes
Summarize today’s sensor information and send it to me every day at 8am

One-shot Triggers

Remind me about the meeting tomorrow at 10am
Automatically start XXX after boot

Automation Rules and Shortcuts

Custom Shortcuts

When a shortcut matches, ESP-Claw can execute it without waking the large model, so response is faster and power consumption is lower.

For example:

Add an automation rule: next time I say “movie mode”, immediately set the light to blue

These rules are useful for:

Smart home workflows
Fixed action chains
Frequent device control
Local offline automation

Startup Tasks

ESP-Claw can automatically run scripts or features after device startup:

Automatically start sensor monitoring after boot
Automatically connect MQTT after boot
Automatically run a display animation after boot

This is useful for long-running and unattended scenarios.

Programming and Chat Coding

ESP-Claw follows the Vibe Coding idea.

Compared with traditional embedded development, you no longer need to repeatedly:

Write large amounts of boilerplate code
Set up projects manually
Compile, flash, and debug in loops

You can collaborate with ESP-Claw through natural language to build features.

To help ESP-Claw deliver results more efficiently and accurately, follow these principles.

1. Intent First, Not Syntax First

Your core asset is not code, but a clear goal.

Focus on describing:

What feature you want
How the user interacts with it
How the device should respond

Use natural, clear, and logical language.

For example:

When the human presence sensor detects someone, turn on the light automatically
Build a desktop screen that shows weather and time

Instead of:

Help me write a while(1) GPIO control loop

2. Run First, Refine Later

ESP-Claw emphasizes fast prototyping and iterative development.

Avoid starting with:

Perfect architecture
Extreme performance
Optimal resource usage
Complex abstraction design

Build a working version first.

After seeing the result, gradually:

Fix issues
Improve the experience
Adjust the structure
Improve performance

Fast validation is often more important than long design cycles.

3. Keep the Feedback Loop

Chat Coding is not:

Ask once and get a perfect application automatically

It is a continuous iteration process:

Describe the goal
    ↓
AI generates the feature
    ↓
Run and observe the result
    ↓
Provide feedback
    ↓
Improve again

If the result does not match expectations, provide:

Screen output
Performance issues
Behavior differences

ESP-Claw can combine logs with this feedback to improve later generation quality.

4. Split Tasks

For complex requirements, avoid asking for an oversized goal in one step.

Split the work into stages, for example:

Light up the screen first
Add touch input
Build the UI
Connect networking last

Small steps are usually more stable than trying to finish everything at once.

5. Understand ESP32 Capability Boundaries

One major difference between ESP-Claw and cloud-only AI is that ESP-Claw runs on real hardware.

It has hardware capabilities, but it is also limited by hardware resources.

Before designing a feature, clarify:

Which features ESP-Claw can do directly
Which features require installing Skills
Which features require extra peripherals
Which features exceed the current hardware capability

For example:

Feature	Directly available
GPIO control	✅
Local automation	✅
Camera recognition	Depends on hardware
Local large-model inference	Usually not suitable for ESP32
Cloud AI collaboration	✅

When Unsure, Ask It Directly

If you are not sure whether a feature is feasible, ask ESP-Claw:

What features do you currently support?
Which hardware resources can you access?
How should this requirement be implemented?
Which Skills need to be installed?
Which peripherals need to be added?

ESP-Claw will suggest an implementation plan based on current device status, installed capabilities, and hardware resources.

Next Steps

Console usage Configuration