Skip to main content

​
Migration from OpenClaw

OpenFang includes a built-in migration engine to import agents, memory, skills, and configurations from OpenClaw.

​
What Gets Migrated

Agent Manifests

Converts YAML agent configs to TOML format

Memory & Sessions

Imports conversation history and KV store

Skills

Converts SKILL.md to OpenFang skill.toml

Configuration

Maps provider names and channel configs

​
Migration Process

1

Backup OpenClaw data

# Backup your OpenClaw installation
cp -r ~/.openclaw ~/.openclaw.backup
2

Dry run (preview changes)

openfang migrate --from openclaw --dry-run
Shows what would be migrated without making changes.
3

Run migration

openfang migrate --from openclaw
Output:
Migrating from OpenClaw...
Found OpenClaw installation at: ~/.openclaw

Migrated:
✓ 5 agents
✓ 12 skills
✓ 3 channel configurations
✓ 847 memory entries

Skipped:
- 2 agents (unsupported module type)

Warnings:
- Agent 'coder' uses tool 'deprecated_tool' (not available in OpenFang)

Migration complete. Review ~/.openfang/migration-report.json
4

Review migrated agents

openfang agent list
All migrated agents are in suspended state. Inspect and activate them:
cat ~/.openfang/agents/migrated-agent.toml
openfang agent spawn ~/.openfang/agents/migrated-agent.toml
5

Test thoroughly

Chat with each agent to ensure it works correctly:
openfang agent chat migrated-agent

​
Migration Details

​
Agent Manifest Conversion

OpenClaw YAML: 
name: my-agent
module: builtin:chat
provider: anthropic
model: claude-3-opus
temperature: 0.7
tools:
  - file_read
  - web_search
OpenFang TOML: 
name = "my-agent"
module = "builtin:chat"

[model]
provider = "anthropic"
model = "claude-3-opus-20240229"
temperature = 0.7

[capabilities]
tools = ["file_read", "web_search"]

​
Tool Name Mapping

OpenClaw tool names are automatically mapped to OpenFang equivalents:
OpenClaw ToolOpenFang Tool
shellshell_exec
read_filefile_read
write_filefile_write
search_webweb_search
fetch_urlweb_fetch
store_memorymemory_store
recall_memorymemory_recall
See crates/openfang-migrate/src/tool_compat.rs for the full mapping (21 tools).

​
Provider Name Mapping

OpenClaw ProviderOpenFang Provider
claudeanthropic
gptopenai
geminigemini
llamagroq or ollama

​
Memory Migration

OpenClaw’s memory files are imported into OpenFang’s SQLite database:
  • Structured KV → memory.db table
  • Sessions → sessions table
  • Embeddings → embeddings table (if present)

​
Skill Migration

OpenClaw SKILL.md files are automatically converted: OpenClaw SKILL.md: 
---
name: github-expert
version: 1.0.0
author: community
---

# GitHub Expert

You are an expert at GitHub workflows...
OpenFang skill.toml + SKILL.md: 
# skill.toml
name = "github-expert"
version = "1.0.0"
author = "community"
runtime = "prompt"
The Markdown content is preserved in SKILL.md.

​
Custom Migration Path

For advanced scenarios: 
# Migrate from specific OpenClaw path
openfang migrate --from openclaw --path /path/to/.openclaw

# Migrate only agents (skip memory/skills)
openfang migrate --from openclaw --agents-only

# Migrate only skills
openfang migrate --from openclaw --skills-only

​
Post-Migration Steps

1

Update API keys

OpenFang uses different environment variable names:
# OpenClaw
export CLAUDE_API_KEY=...

# OpenFang
export ANTHROPIC_API_KEY=...
See LLM Providers for all variable names.
2

Review capabilities

Migrated agents may have broader capabilities than intended. Review and restrict:
[capabilities]
tools = ["file_read"]  # Was ["file_read", "file_write", "shell_exec"]
3

Update channel configs

Channel adapter names may differ:
# OpenClaw: telegram_bot
# OpenFang: telegram
[channels.telegram]
enabled = true
4

Test all agents

Spawn and test each migrated agent:
for agent in ~/.openfang/agents/*.toml; do
  echo "Testing $agent..."
  openfang agent spawn "$agent"
  openfang agent chat $(basename "$agent" .toml)
done

​
Compatibility Notes

​
Supported Features

✅ Agent manifests (YAML → TOML) ✅ Memory KV store ✅ Sessions and conversation history ✅ Skills (SKILL.md format) ✅ Tool name mapping (21 tools) ✅ Provider name mapping ✅ Channel configurations

​
Unsupported Features

❌ OpenClaw-specific modules (must be ported manually) ❌ Custom WASM modules (rebuild for OpenFang) ❌ Third-party plugins (check FangHub for alternatives)

​
Known Issues

Some OpenClaw tools don’t have direct OpenFang equivalents. The migration report lists these. You’ll need to:
  1. Remove the deprecated tool from capabilities.tools
  2. Update the system prompt to use available alternatives
If OpenClaw used custom memory structures, they may not import cleanly. Check:
sqlite3 ~/.openfang/data/openfang.db "SELECT * FROM memory LIMIT 10;"
Python skills may reference packages not in OpenFang’s bundled environment. Add them to skill.toml:
[requirements]
python_packages = ["requests", "beautifulsoup4"]

​
Migration Report

After migration, review the detailed report: 
cat ~/.openfang/migration-report.json

{
  "timestamp": "2026-03-07T12:34:56Z",
  "source": "openclaw",
  "source_path": "/home/user/.openclaw",
  "imported": {
    "agents": 5,
    "skills": 12,
    "memory_entries": 847,
    "channels": 3
  },
  "skipped": {
    "agents": [{"name": "custom-wasm", "reason": "unsupported module type"}]
  },
  "warnings": [
    "Agent 'coder' uses deprecated tool 'shell' (mapped to 'shell_exec')"
  ]
}

​
Getting Help

If you encounter migration issues:
  1. Check the migration documentation
  2. Review the migration report for specific errors
  3. Join the Discord community
  4. Open an issue on GitHub

​
Next Steps

Configuration

Set up your LLM providers

Agent Setup

Create new agents for OpenFang

Build docs developers (and LLMs) love

Get started for freeTalk to us