Overview
Connection Error
No Recent Activity
Activity will appear here once agents start processing.
Select an agent to start chatting
Choose an agent from the sidebar or go to the Agents tab to create a new one.
Chat
Execution Approvals
No approvals
When agents request permission for sensitive actions, they'll appear here.
Workflows
Scheduler
*/5 * * * * (every 5 min) or
0 9 * * 1-5 (weekdays at 9am). You can also run any job
manually with the "Run Now" button.
| Name | Schedule | Agent | Status | Delivery | Last Run | Next Run | Actions |
|---|---|---|---|---|---|---|---|
|
Delivery Log
Loading delivery log...
Configured targets ():
No fan-out targets configured.
Recent deliveries ():
Delivery history will appear here after the next run. Per-run results are not yet
persisted — the fan-out engine emits tracing events you can view under
Logs.
|
|
|
|
No scheduled jobs
Create a cron job to run agents on a recurring schedule. Jobs are stored persistently and survive restarts.
| Agent | Pattern | Prompt | Fires | Enabled | Created | Actions |
|---|---|---|---|---|---|---|
No event triggers
Create event triggers on the Workflows page to fire agents in response to system events.
| Time | Name | Type | Status | Total Runs |
|---|---|---|---|---|
No run history yet
Run history will appear here after scheduled jobs or triggers execute.
Channels
No channels match your search.
Connecting to WhatsApp Web gateway...
Channel will activate automatically.
Or use the with a Meta developer account.
Configure via WhatsApp Cloud API (requires a Meta Business developer account).
How to get credentials
Skills
Skills & Ecosystem
Skills extend your agents with new capabilities. OpenFang supports the OpenClaw/ClawHub ecosystem (3,000+ community skills) plus local skills.
- Prompt-only — inject context and instructions into the agent's system prompt (most ClawHub skills)
- Python / Node.js — executable tools that agents can call during conversations
- MCP Servers — external tools via Model Context Protocol (GitHub, filesystem, databases, etc.)
No skills installed
Skills add new capabilities to your agents. Browse ClawHub for 3,000+ community skills or create your own.
ClawHub may be temporarily unavailable. The OpenClaw ecosystem is hosted at clawhub.ai.
No skills found for ""
Try a different search term or browse by category.
MCP Servers (Model Context Protocol)
MCP servers provide external tools to your agents — GitHub, filesystem, databases, APIs, and more. OpenFang is compatible with all OpenClaw MCP servers.
Configure MCP servers in your config.toml under [mcp_servers].
No MCP servers configured
MCP servers extend your agents with external tools. Add servers to your config.toml:
[[mcp_servers]] name = "filesystem" timeout_secs = 30 [mcp_servers.transport] type = "stdio" command = "npx" args = ["-y", "@modelcontextprotocol/server-filesystem", "/path"]
OpenFang supports all OpenClaw-compatible MCP servers.
Quick Start Skills
Create prompt-only skills with one click. These inject context into your agent's system prompt — no code required. Perfect for adding domain expertise or workflow guidelines.
Hands
Hands — Curated Autonomous Capability Packages
Hands are pre-configured AI agents that autonomously handle specific tasks. Each hand includes a tuned system prompt, required tools, and a dashboard for tracking work.
No hands available
Hands are curated AI capability packages. They will appear once the kernel loads bundled hands.
No active hands
Activate a hand from the Available tab to get started. Each hand spawns a dedicated agent.
Settings
LLM Providers
OpenFang supports 12 LLM providers out of the box. Configure API keys to unlock models from each provider. Set environment variables and restart, or use the form below to save keys directly.
Add Custom Provider
Connect any OpenAI-compatible API (vLLM, LiteLLM, LocalAI, etc.)
No providers found
Provider information could not be loaded. Check that the API is running.
Model Catalog
Browse all available models across providers. Models marked "Available" have their provider configured and ready to use.
Add Custom Model
| Model | Provider | Tier | Context | Input Cost | Output Cost | Status | |
|---|---|---|---|---|---|---|---|
| Tool | Description |
|---|---|
Runtime Configuration
View and edit the active configuration. Changes are applied immediately. For advanced edits, modify config.toml in your OpenFang data directory.
sha2 SHA-256
hmac HMAC-SHA256
subtle constant-time
ed25519-dalek signing
zeroize secret wiping
rand randomness
governor rate limiting
Peer Networking (OFP)
Link multiple OpenFang instances into a mesh via the OFP wire protocol.
| Node | Address | State | Agents | Protocol |
|---|---|---|---|---|
No peers connected
Add a [network] section to config.toml with shared_secret and peer addresses.
A2A External Agents
Discovered agents on other OpenFang/A2A-compatible instances that this node can communicate with.
| Name | URL | Description |
|---|---|---|
Budget & Spending Limits
Monitor and control spending across all agents.
Top Spenders (Today)
| Agent | Today | Hourly Limit | Daily Limit | Monthly Limit | Token Limit/hr |
|---|---|---|---|---|---|
- Converts agent.yaml to agent.toml with proper capabilities
- Maps tools (read_file → file_read, execute_command → shell_exec, etc.)
- Merges channel configs into config.toml
- Copies workspace files and memory data
Analytics
| Model | Calls | Input Tokens | Output Tokens | Cost | Usage |
|---|---|---|---|---|---|
No model usage data yet.
| Agent | Total Tokens | Tool Calls |
|---|---|---|
No agent usage data yet.
| Model | Provider | Tier | Input Tokens | Output Tokens | Calls | Cost | Cost Share |
|---|---|---|---|---|---|---|---|
Sessions
| Session | Agent | Messages | Created | Actions |
|---|---|---|---|---|
No sessions yet
Sessions are created when you chat with agents. Start a conversation to see session history here.
No sessions match your filter.
Agent Memory
Each agent has its own key-value memory store. Agents use memory to persist preferences, notes, and context between conversations.
| Key | Value | Actions |
|---|---|---|
|
|
|
No keys stored
This agent has no memory entries yet. Agents create memory entries automatically during conversations, or you can add them manually.
Select an Agent
Choose an agent from the dropdown above to browse and edit its memory store.
Logs
No log entries yet
Activity will appear here as agents run.
| # | Timestamp | Agent | Action | Detail | Outcome |
|---|---|---|---|---|---|
No audit entries yet
Activity will appear here as agents operate.
Agent Comms
Connection Error
Setup Wizard
Welcome to OpenFang
OpenFang is an open-source Agent Operating System. It lets you run AI agents that can chat, use tools, access memory, and connect to messaging channels — all from a single dashboard.
Takes about 2 minutes. You can skip any step and configure later.
Connect an LLM Provider
OpenFang needs at least one LLM provider to power your agents. Select a provider and enter your API key.
Provider Already Configured
You already have at least one provider set up. You can continue to the next step or configure additional providers.
npm install -g @anthropic-ai/claude-codeclaude authnpm install -g @anthropic-ai/claude-code && claude authCreate Your First Agent
Pick a template to get started quickly. You can customize the agent later or create more from the Agents page.
Try Your Agent
Send a quick message to test your new agent. Try one of the suggestions below or type your own.
Connect a Channel Optional
Channels let your agent communicate via messaging platforms. This is optional — you can always use the built-in web chat.
You can skip this step. The built-in web chat is always available from the Agents page. Add channels any time from Settings → Channels.
You're All Set!
OpenFang is configured and ready to go. Here is a summary of what was set up:
Runtime
| Platform | |
| Architecture | |
| API Listen | |
| Home Directory | |
| Log Level | |
| Network |
| Provider | Status | Models | Latency |
|---|---|---|---|