Everything You Need in a $5 AI Agent
A full-stack AI agent platform that fits on a microcontroller. No compromises.
Streaming AI Agent Engine
ESP-Claw processes AI responses token-by-token using a custom streaming JSON parser. This means the agent can start acting on tool calls before the full response arrives — using just 400KB of SRAM.
Streaming JSON Parser
Process tool calls mid-stream without buffering the entire response
Multi-Provider Support
Claude, OpenAI, Ollama, or any OpenAI-compatible API endpoint
SOUL.md Personality
Define behavior in Markdown — readable, shareable, version-controlled
Persistent Memory
USER.md and MEMORY.md survive reboots, making your agent remember context
User → "Turn on the porch light"
Stream starts...
├─ Parse: tool_call detected
├─ Execute: mqtt_publish
│ topic: home/porch/light
│ payload: {"state": "on"}
├─ Result: published OK
└─ Generate response
AI → "Done! Porch light is on. 💡"
───────────────────────────
Tokens: 89 | Latency: 0.8s
RAM: 12KB peak | Tools: 1Write Once, Run on Any ESP32
The hardware abstraction layer automatically detects your chip variant and configures pin mappings, memory limits, and peripheral support. Switch boards without changing a line of code.
- Auto-detect chip variant at boot
- Unified GPIO/I2C/SPI/I2S API
- Memory-aware tool scheduling
- OTA updates over Wi-Fi
40+ Built-In Tools
The AI calls tools automatically based on conversation context.
Sensors
- DHT22 Temp/Humidity
- BME280 Barometer
- PIR Motion
- Ultrasonic Range
- ADC Voltage
Actuators
- Servo Motors
- Relay Control
- PWM / LED
- NeoPixel RGB
- IR Blaster
Communication
- Telegram Bot
- Discord Bot
- MQTT Pub/Sub
- HTTP Client
- Webhooks
System
- OTA Update
- Deep Sleep
- Cron Scheduler
- Persistent Memory
- Wi-Fi Scan
Talk Through Any Channel
Your AI agent connects to the messaging platforms you already use. One device, many interfaces.
Start Building Today
Get your AI agent running in 5 minutes with a $5 board and a browser.