Autohand implements a CLI-based AI coding assistant utilizing the ReAct (Reason + Act) reasoning pattern. The system processes requests through sequential phases: reasoning, tool invocation (file system, Git operations, shell commands), result interpretation, and JSON-formatted response generation. Implemented in TypeScript with Bun runtime for optimal performance.

• ReAct Execution Model: reason → tool calls → observation → final response
• Toolset: 20+ operations including read_file, write_file, git_status, run_command, search, etc.
• Response Protocol: Structured JSON {"thought":"...","toolCalls":[...],"finalResponse":"..."}
• Workspace Integration: Operates within the current project directory with Git state awareness
• Runtime: Bun-native with zero external dependencies, native TypeScript execution
• Safety Controls: Destructive operations require explicit justification in reasoning trace

• Bun ≥1.0 (curl -fsSL https://bun.sh/install | bash)

bun install

# Development mode
bun run dev

# Production build and execution
bun build
./dist/cli.js

# Global installation
bun add -g .

Create a .autohandrc.json file in your project root or home directory:

{
  "openrouter": {
    "apiKey": "your-api-key-here",
    "model": "anthropic/claude-3.5-sonnet"
  },
  "workspace": {
    "defaultRoot": ".",
    "allowDangerousOps": false
  },
  "ui": {
    "theme": "dark",
    "autoConfirm": false,
    "readFileCharLimit": 300
  }
}

For telemetry and feedback submission, create a .env file in the project root:

# Autohand API Configuration
AUTOHAND_API_URL=https://api.autohand.ai
AUTOHAND_SECRET=your-company-secret-here

Note: The Autohand feedback and telemetry API backend has been extracted to a separate repository: 👉 https://github.com/autohandai/api

The CLI includes client libraries with offline queueing, retry mechanisms, and privacy features. API authentication uses a compound identifier: device_id.company_secret.

• readFileCharLimit (default: 300): Maximum characters to display when reading files. Larger files will be truncated with a clear indicator.
• theme: UI theme ("dark" | "light")
• autoConfirm: Skip confirmation prompts for destructive operations

See config.example.json for a full example.

Create run-microvm.sh:

#!/bin/bash
firecracker --api-sock /tmp/fc.sock &

cat > config.json <<EOF
{
  "boot-source": {
    "kernel_image_path": "vmlinux.bin",
    "boot_args": "console=ttyS0 reboot=k panic=1 pci=off"
  },
  "drives": [{
    "drive_id": "rootfs",
    "path_on_host": "rootfs.ext4",
    "is_root_device": true,
    "is_read_only": false
  }],
  "machine-config": { "vcpu_count": 2, "mem_size_mib": 1024 }
}
EOF

curl --unix-socket /tmp/fc.sock -d @config.json http://localhost/actions

# Within MicroVM: bun install && bun run dev

Execute: chmod +x run-microvm.sh && ./run-microvm.sh

Dockerfile:

FROM oven/bun:1
WORKDIR /app
COPY . .
RUN bun install
RUN bun build
CMD ["./dist/cli.js"]

Build and run:

docker build -t autohand .
docker run -it autohand