poe-connector

# Poe Connector — USE BASH TOOL

**You MUST use the `bash` or `shell` tool to run these commands. No other method works.**

## Commands

Chat with AI:
```bash
bash ~/.openclaw/workspace/skills/poe-connector/poe.sh chat Tell me 10 jokes
```

Generate image:
```bash
bash ~/.openclaw/workspace/skills/poe-connector/poe.sh image A sunset over mountains
```

Generate video:
```bash
bash ~/.openclaw/workspace/skills/poe-connector/poe.sh video A cat playing piano
```

Generate audio:
```bash
bash ~/.openclaw/workspace/skills/poe-connector/poe.sh audio Hello world in a cheerful voice
```

List models:
```bash
bash ~/.openclaw/workspace/skills/poe-connector/poe.sh models
```

Search models:
```bash
bash ~/.openclaw/workspace/skills/poe-connector/poe.sh search claude
```

## DO NOT

- Do NOT use `sessions_spawn` — it will fail
- Do NOT use browser to visit poe.com — it will fail
- Do NOT set poe as a model provider — it will fail
- Do NOT use any approach other than `bash` tool — only bash works

## Setup (required before first use)

This skill requires a Poe API key. Get one at https://poe.com/api/keys, then add to `~/.openclaw/openclaw.json`:

```json
{ "skills": { "entries": { "poe-connector": { "env": { "POE_API_KEY": "your-key" } } } } }
```

## Notes

- All commands use sensible default models (configured in poe_config.json)
- If you get a POE_API_KEY error, tell the user to run the setup above
- File attachments (--files) are base64-encoded and sent to Poe — do not attach sensitive files unless the user intends to share them with Poe

# Poe Connector — OpenClaw Skill

An [OpenClaw](https://github.com/openclaw/openclaw) skill that connects your agent to [Poe.com](https://poe.com)'s API, giving it access to **100+ AI models** through a single API key.

## Why This Exists

I've been a Poe user since it first launched — there's always a new model to try and I genuinely enjoy the platform. Last year I accidentally subscribed to a pretty expensive plan, and I've never been able to use up all the credits. So I figured: why not hook it up to OpenClaw and see if that burns through them a bit faster?


## Features

- **Text Chat** — Talk to Claude, GPT, Gemini, Grok, Llama, and hundreds more via `/v1/chat/completions`
- **Image Generation** — GPT-Image-1, Imagen-4, Flux-Kontext, Seedream-3.0
- **Video Generation** — Veo-3, Runway-Gen-4-Turbo, Kling-2.1
- **Audio / TTS** — ElevenLabs, Lyria
- **Multimodal File Input** — Send images, PDFs, audio, video alongside your prompt
- **Model Discovery** — List and search all available Poe models
- **Streaming** — Real-time token-by-token output
- **Auto Retry** — Exponential backoff on rate-limit (429) errors

## Quick Start

### 1. Get your Poe API key

Go to https://poe.com/api/keys and create an API key.

### 2. Configure the API key

Add your key to `~/.openclaw/openclaw.json`:

```json
{
  "skills": {
    "entries": {
      "poe-connector": {
        "env": { "POE_API_KEY": "your-key-here" }
      }
    }
  }
}
```

The skill reads the key from this file at runtime. No environment variable needed.

### 3. Install dependency

```bash
pip install openai
```

### 4. Use it

```bash
# Chat (uses default model from poe_config.json, streams by default)
python src/poe_chat.py --message "Hello!"

# Chat with a specific model
python src/poe_chat.py --model GPT-5.2 --message "Write a haiku"

# Generate an image (default: GPT-Image-1)
python src/poe_media.py --type image --prompt "A cat astronaut" --aspect 1:1

# List models
python src/poe_models.py --list

# Search model...