FGP OpenAI Daemon

Fast OpenAI API access with connection pooling for GPT, DALL-E, Whisper, and more.

Why FGP?

FGP daemons maintain persistent connections and avoid cold-start overhead. Instead of spawning a new API client for each request, the daemon stays warm and ready.

Benefits:

• No cold-start latency
• Connection pooling
• Persistent authentication

Installation

brew install fast-gateway-protocol/tap/fgp-openai

# Or
bash ~/.claude/skills/fgp-openai/scripts/install.sh

Setup

export OPENAI_API_KEY="sk-..."

Usage

Chat Completions

# Simple completion
fgp openai chat "What is the capital of France?"

# With model
fgp openai chat "Explain quantum computing" --model gpt-4o

# System message
fgp openai chat "Write a poem" --system "You are a creative poet"

# With temperature
fgp openai chat "Generate a story" --temperature 0.9

# Streaming
fgp openai chat "Tell me a story" --stream

# JSON mode
fgp openai chat "List 3 colors" --json

# From file
fgp openai chat --file prompt.txt

Embeddings

# Single text
fgp openai embed "Hello world"

# Multiple texts
fgp openai embed "Text 1" "Text 2" "Text 3"

# From file
fgp openai embed --file texts.txt

# With model
fgp openai embed "Hello" --model text-embedding-3-large

# Output to file
fgp openai embed "Hello" --output embedding.json

Image Generation (DALL-E)

# Generate image
fgp openai image "A cat wearing a hat"

# With size
fgp openai image "Mountain landscape" --size 1792x1024

# With quality
fgp openai image "Portrait" --quality hd

# With style
fgp openai image "Abstract art" --style vivid

# Multiple images
fgp openai image "Sunset" --n 4

# Save to file
fgp openai image "Robot" --output robot.png

Image Analysis (Vision)

# Analyze image
fgp openai vision ./image.png "What's in this image?"

# From URL
fgp openai vision "https://example.com/image.jpg" "Describe this"

# Detailed analysis
fgp openai vision ./chart.png "Explain this chart in detail"

Audio (Whisper)

# Transcribe audio
fgp openai transcribe ./audio.mp3

# With language hint
fgp openai transcribe ./audio.mp3 --language en

# Translate to English
fgp openai translate ./spanish-audio.mp3

# With timestamps
fgp openai transcribe ./audio.mp3 --timestamps

# Output formats
fgp openai transcribe ./audio.mp3 --format srt
fgp openai transcribe ./audio.mp3 --format vtt

Text-to-Speech

# Generate speech
fgp openai speak "Hello, welcome to our service" --output welcome.mp3

# With voice
fgp openai speak "Breaking news" --voice nova

# Available voices: alloy, echo, fable, onyx, nova, shimmer

# With speed
fgp openai speak "Slow speech" --speed 0.8

Assistants

# List assistants
fgp openai assistants

# Create assistant
fgp openai assistant create --name "Helper" --model gpt-4o --instructions "You are a helpful assistant"

# Run assistant
fgp openai assistant run <assistant-id> "Help me with this task"

# With file
fgp openai assistant run <assistant-id> "Analyze this" --file data.csv

Files

# List files
fgp openai files

# Upload file
fgp openai file upload ./data.jsonl --purpose fine-tune

# Get file info
fgp openai file <file-id>

# Download file
fgp openai file download <file-id> --output data.jsonl

# Delete file
fgp openai file delete <file-id>

Available Commands

Command	Description	Example
chat	Chat completion	fgp openai chat "Hello"
embed	Get embeddings	fgp openai embed "text"
image	Generate image	fgp openai image "prompt"
vision	Analyze image	fgp openai vision img.png "?"
transcribe	Audio to text	fgp openai transcribe audio.mp3
speak	Text to speech	fgp openai speak "text"
assistants	List assistants	fgp openai assistants
files	List files	fgp openai files

Models

Type	Models
Chat	gpt-4o, gpt-4o-mini, gpt-4-turbo, gpt-3.5-turbo
Embeddings	text-embedding-3-small, text-embedding-3-large
Image	dall-e-3, dall-e-2
Audio	whisper-1
TTS	tts-1, tts-1-hd

Example Workflows

Generate embeddings for RAG

# Embed documents
cat docs/*.txt | fgp openai embed --model text-embedding-3-small --output embeddings.json

Transcribe meeting

fgp openai transcribe meeting.mp3 --format srt > meeting.srt

Create image asset

fgp openai image "Professional logo for tech startup, minimal, blue" --size 1024x1024 --output logo.png

Troubleshooting

Invalid API key

Error: Invalid API key

Check OPENAI_API_KEY is correct.

Rate limited

Error: Rate limit reached

Wait or upgrade your tier.

Context too long

Error: Maximum context length exceeded

Reduce input length or use a model with larger context.

Architecture

• OpenAI REST API
• Connection pooling for reduced latency
• UNIX socket at ~/.fgp/services/openai/daemon.sock
• Streaming support for chat completions