README.md

Fox1.2 OpenClaw - 100% OpenClaw Tool Supported

A fine-tuned Qwen2.5-0.5B model optimized for OpenClaw agent tool execution.

Model Details

• Base: Qwen2.5-0.5B-Instruct
• Parameters: ~494M (compressed, efficient)
• Context Length: 32,768 tokens
• Size: ~994MB (F16)

Capabilities - All OpenClaw Tools Supported ✅

Core Tools

• exec: Shell command execution (ls, cd, git, docker, npm, pip, python, node, etc.)
• read: Read file contents (path required)
• write: Write content to file (path, content required)
• edit: Edit file by replacing exact text (path, oldText, newText required)

Process Management

• process: Background session management (list, poll, write, send-keys, submit, paste, kill)

Web Operations

• web_search: DuckDuckGo search (query, count, region, safeSearch)
• web_fetch: Fetch and extract web content (url, extractMode, maxChars)

OpenClaw Management

• session_status: Get session information
• sessions_list: List active sessions (activeMinutes, kinds, limit, messageLimit)
• sessions_history: Get conversation history (sessionKey, includeTools, limit)
• sessions_send: Send messages between sessions (sessionKey/label, message)
• sessions_spawn: Spawn sub-agents or ACP sessions (task, runtime, mode, etc.)
• sessions_yield: End current turn
• subagents: Manage spawned sub-agents (list, kill, steer)

Cron Jobs

• cron: Job management (status, list, add, update, remove, run, runs, wake)

Memory

• memory_search: Search memory files (query, maxResults, minScore)
• memory_get: Retrieve memory content (path, from, lines)

Additional

• image: Image analysis (image/images, prompt)
• weather: Weather information (location)

Tool Call Format

When you need to execute a tool, respond with JSON in this format:

{"action": "tool_name", "param1": "value1", "param2": "value2"}

Examples

{"action": "exec", "command": "ls -la"}
{"action": "read", "path": "/home/user/README.md"}
{"action": "write", "path": "/home/user/test.txt", "content": "Hello World"}
{"action": "web_search", "query": "python tutorials"}
{"action": "weather", "location": "Athens"}
{"action": "cron", "action": "list"}
{"action": "session_status"}

Usage

Ollama

# Create the model
ollama create fox1.2-openclaw -f Modelfile

# Run it
ollama run fox1.2-openclaw "list files in current directory"

Python/Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("teolm30/fox1.2-openclaw")
tokenizer = AutoTokenizer.from_pretrained("teolm30/fox1.2-openclaw")

# Generate tool call
inputs = tokenizer("List all files", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0]))

HuggingFace CLI

# Download
huggingface-cli download teolm30/fox1.2-openclaw

# Or use curl
curl -L -o fox1.2-openclaw.gguf https://huggingface.co/teolm30/fox1.2-opencloak/resolve/main/fox1.2-openclaw.gguf

Why This Model?

• Compact: 494M params - runs on consumer hardware (6GB VRAM)
• Fast: Optimized for local inference
• 100% Tool Support: All OpenClaw agent tools supported
• Smart Tool Selection: Knows when to use each tool appropriately
• OpenClaw Native: Built specifically for OpenClaw integration

Training

Trained on 200+ examples covering all OpenClaw tool patterns:

• File operations (read, write, edit)
• Shell commands (exec)
• Web operations (search, fetch)
• Session management
• Cron jobs
• Memory operations
• And more...

Optimized for tool call generation and execution in agent workflows.

OpenClaw Configuration

Add to your models.json:

{
  "id": "fox1.2-openclaw:latest",
  "name": "Fox1.2 OpenClaw",
  "reasoning": false,
  "input": ["text"],
  "contextWindow": 32768,
  "maxTokens": 4096,
  "api": "ollama"
}

Then restart OpenClaw: openclaw gateway restart

License

Apache 2.0

Author

teolm30 (OpenClaw Community)

Version History

• v1.0 (2026-04-01): Initial release with 100% OpenClaw tool support