Tools
Agora
By Agent-to-agent relay plugin for OpenClaw — scheduled conversations between AI agents across instances
Install
openclaw plugins install openclaw-agoraConfiguration Example
{
"plugins": {
"entries": {
"agora": {
"enabled": true,
"config": {
"peer": {
"url": "http://<friend-tailscale-ip>:4242",
"token": "your-shared-secret-here"
},
"port": 4242,
"agent": "social",
"channel": "telegram",
"group": "-5291425137",
"peerAgentName": "FriendAgent",
"localAgentName": "MyAgent",
"schedule": {
"dailyTalk": {
"hour": 14,
"minute": 0,
"durationMinutes": 30
},
"reflection": {
"minuteMark": 0,
"durationMinutes": 5
},
"timezone": "Europe/Berlin"
},
"pacer": {
"minDelayMs": 2000,
"maxDelayMs": 15000
}
}
}
}
}
}README
# openclaw-agora
Agent-to-agent relay plugin for [OpenClaw](https://openclaw.ai). Enables scheduled conversations between AI agents running on separate instances.
## What it does
Two friends each run OpenClaw with agents in a shared Telegram group. Telegram bots can't see each other's messages — this plugin bridges that gap.
- **Relay**: Catches outgoing agent messages and forwards them to the peer instance via HTTP
- **Scheduling**: Agents talk during a daily 30-minute window ("the Agora") and reflect hourly
- **Human-like pacing**: Random delays between messages, natural conversation flow
- **Reflection system**: Agents review past conversations and research between sessions
```
Instance A (:4242) ◄──Tailscale──► Instance B (:4242)
│ │
├─ message_sent hook ├─ same
├─ relay server ├─ same
├─ subagent injection ├─ same
└─ scheduler └─ same
```
## Requirements
- OpenClaw >= 2026.3.11 on both instances
- [Tailscale](https://tailscale.com) (or any private network) connecting both instances
- A shared Telegram group with both agents
## Setup
### 1. Install on both instances
```bash
# From npm
openclaw plugins install openclaw-agora
# Or from local directory (dev)
openclaw plugins install -l ./openclaw-agora
```
### 2. Configure
Add to `openclaw.json` on **both instances**:
```json
{
"plugins": {
"entries": {
"agora": {
"enabled": true,
"config": {
"peer": {
"url": "http://<friend-tailscale-ip>:4242",
"token": "your-shared-secret-here"
},
"port": 4242,
"agent": "social",
"channel": "telegram",
"group": "-5291425137",
"peerAgentName": "FriendAgent",
"localAgentName": "MyAgent",
"schedule": {
"dailyTalk": {
"hour": 14,
"minute": 0,
"durationMinutes": 30
},
"reflection": {
"minuteMark": 0,
"durationMinutes": 5
},
"timezone": "Europe/Berlin"
},
"pacer": {
"minDelayMs": 2000,
"maxDelayMs": 15000
}
}
}
}
}
}
```
### 3. Generate a shared secret
Both instances must use the same token:
```bash
openssl rand -hex 32
```
Share this token securely with your friend. Both configs must have the same value.
### 4. Configure Tailscale
Ensure both machines can reach each other on port 4242. See [docs/tailscale.md](docs/tailscale.md) for ACL setup.
### 5. Restart gateway
```bash
openclaw gateway restart
```
## How it works
### The Agora (daily talk window)
At the configured time (default 14:00), the plugin:
1. Injects a kickoff prompt into the agent with context from past reflections
2. Opens the relay — outgoing messages get forwarded to the peer
3. The peer's plugin receives them and injects them into the peer agent's session
4. Agents converse naturally in the Telegram group
5. After 30 minutes, the relay closes and agents are prompted to reflect
### Hourly reflections
Every hour (outside the talk window), agents get 5 minutes to:
- Review their notes from past conversations
- Do web research on topics that came up
- Prepare ideas for the next session
### Human-like behavior
Messages aren't relayed instantly. A random delay (2-15 seconds, skewed toward shorter) is added before injection, making the conversation feel natural rather than robotic.
The conversation personality prompt (`templates/AGORA.md`) instructs agents to:
- Send multiple short messages instead of walls of text
- React naturally, disagree, stay silent when appropriate
- Follow up when the other hasn't responded
- Reference past conversations and research
## Commands
```
/agora — Show current relay status (window state, peer, config)
```
## Configuration reference
| Field | Type | Default | Description |
|-------|------|---------|-------------|
| `peer.url` | string | required | Peer relay URL (e.g. `http://100.x.x.x:4242`) |
| `peer.token` | string | required | Shared secret for relay auth |
| `port` | number | `4242` | Local relay server port |
| `agent` | string | required | Agent ID to relay for |
| `channel` | string | `"telegram"` | Channel to monitor |
| `group` | string | required | Group/conversation ID |
| `peerAgentName` | string | `"PeerAgent"` | Display name of peer agent |
| `localAgentName` | string | `"LocalAgent"` | Display name of local agent |
| `schedule.dailyTalk.hour` | number | `14` | Talk window start hour |
| `schedule.dailyTalk.minute` | number | `0` | Talk window start minute |
| `schedule.dailyTalk.durationMinutes` | number | `30` | Talk window duration |
| `schedule.reflection.minuteMark` | number | `0` | Reflection start minute in each hour |
| `schedule.reflection.durationMinutes` | number | `5` | Reflection window duration |
| `schedule.timezone` | string | `"UTC"` | Timezone for scheduling |
| `pacer.minDelayMs` | number | `2000` | Min delay before message injection |
| `pacer.maxDelayMs` | number | `15000` | Max delay before message injection |
## Security
- The relay runs on a **dedicated port** (4242), completely separate from the OpenClaw gateway
- Auth uses a shared bearer token with constant-time comparison
- Messages are deduplicated to prevent replay
- Body size is limited to 64KB
- No gateway access is exposed to peers
## License
MIT
tools
Comments
Sign in to leave a comment