Provider Guide#

OpenMontage connects to external providers only through dedicated tools. The agent never calls general-purpose LLM APIs at runtime. Each provider is configured via environment variables in .env. Keys are optional—add only those you need. The tool registry discovers available providers automatically.

See Configuring Providers for full .env setup and Tool Registry for how selectors route requests.

Quick Start#

Start with free options and add paid providers as needed.

Environment Variables#

Copy the template and edit:

cp .env.example .env

Key variables (add only what you use):

# Free stock
PEXELS_API_KEY=
PIXABAY_API_KEY=

# Google (TTS + Imagen)
GOOGLE_API_KEY=

# Voice and music
ELEVENLABS_API_KEY=
OPENAI_API_KEY=
XAI_API_KEY=
DOUBAO_SPEECH_API_KEY=
DOUBAO_SPEECH_VOICE_TYPE=zh_female_vv_uranus_bigtts

# Multi-model gateway
FAL_KEY=

# Video
HEYGEN_API_KEY=
RUNWAY_API_KEY=
SUNO_API_KEY=

# Local GPU
VIDEO_GEN_LOCAL_ENABLED=true
VIDEO_GEN_LOCAL_MODEL=wan2.1-1.3b

Cloud Providers#

xAI (Grok)#

Tools unlocked: grok_image, grok_video

Setup:

1. Create an xAI developer account.
2. Generate an API key in the developer console.
3. Add XAI_API_KEY=xai-... to .env.

Best for: Image editing, style transfer, short reference-image video.

Pricing:

• grok-imagine-image: $0.02 per image
• Input images (edits): $0.002 per image
• Video at 480p: $0.05/sec
• Video at 720p: $0.07/sec

fal.ai#

Tools unlocked: flux_image, recraft_image, kling_video, veo_video, minimax_video

Setup:

1. Sign up at fal.ai (GitHub or Google).
2. Create a key at fal.ai/dashboard/keys.
3. Add FAL_KEY=your-key-here to .env.

Pricing (pay-as-you-go):

• FLUX Pro v1.1: $0.05/image (20 images per $1)
• FLUX Dev: $0.03/image (33 images per $1)
• Kling 2.5 Turbo Pro: $0.07/sec (14 seconds per $1)
• Veo 3: $0.40/sec (2.5 seconds per $1)

ElevenLabs#

Tools unlocked: elevenlabs_tts, music_gen

Setup:

1. Sign up at elevenlabs.io.
2. Go to Profile > API Keys and create a key.
3. Add ELEVENLABS_API_KEY=xi_your-key-here to .env.

Pricing:

• Free: 10,000 characters/month (2-3 minutes narration)
• Starter: $5/mo for 30,000 characters

Google#

Tools unlocked: google_tts, google_imagen

Setup:

1. Go to Google AI Studio and create an API key.
2. Enable Text-to-Speech API and Generative Language API in the Cloud console.
3. Add GOOGLE_API_KEY=AIza... to .env.

TTS pricing (free tier 1M chars/month per voice type):

• Standard: $4.00 per 1M chars after free tier
• WaveNet/Neural2: $16.00 per 1M chars

Imagen pricing:

• Fast: $0.02/image
• Standard: $0.04/image

New accounts receive $300 in credits.

OpenAI#

Tools unlocked: openai_tts, openai_image

Setup:

1. Create an account at platform.openai.com.
2. Add a payment method.
3. Create a secret key and add OPENAI_API_KEY=sk-... to .env.

Pricing:

• TTS: $15.00 per 1M characters (tts-1)
• DALL-E 3: $0.040 per 1024x1024 standard image

Runway#

Tools unlocked: runway_video

Setup:

1. Create a developer account at dev.runwayml.com.
2. Subscribe to Standard or higher.
3. Generate an API key and add RUNWAY_API_KEY=key_... to .env.

Pricing:

• Standard: $12/mo for 625 credits (~25 seconds Gen-4)
• Gen-4 Turbo: ~$0.05 per second

Free tier: 125 one-time credits.

HeyGen#

Tools unlocked: heygen_video

Setup:

1. Register at app.heygen.com.
2. Generate an API key in settings.
3. Add prepaid balance and HEYGEN_API_KEY=your-key-here to .env.

Pricing:

• Avatar video (Engine III): $0.017/sec
• Prompt to Video: $0.033/sec

Suno#

Tools unlocked: suno_music

Setup:

1. Create a Suno account.
2. Obtain an API key from sunoapi.org.
3. Add SUNO_API_KEY=your-key-here to .env.

Pricing: 1 credit = $0.005. Free tier: 50 credits/day (non-commercial).

Pexels and Pixabay#

Tools unlocked: pexels_image, pexels_video, pixabay_image, pixabay_video

Setup:

1. Create free accounts at pexels.com and pixabay.com.
2. Copy the displayed API keys.
3. Add PEXELS_API_KEY= and PIXABAY_API_KEY= to .env.

Pricing: Completely free. No attribution required. Commercial use allowed.

Local Options#

Local providers require no API keys and incur no cost.

Piper TTS#

Tool: piper_tts

Setup:

pip install piper-tts

Download a voice model on first use (e.g., en_US-lessac-medium). Fully offline.

Remotion#

Tool: video_compose (auto-routes when needed)

Setup:

cd remotion-composer && npm install

Requires Node.js 18+. Renders animated text cards, charts, and image scenes.

HyperFrames#

Tool: hyperframes_compose or video_compose with render_runtime=hyperframes

Setup:

npx --yes hyperframes doctor

Requires Node.js ≥ 22 and FFmpeg. Used for kinetic typography and character rigs.

Local Video Generation#

Tools: wan_video, hunyuan_video, cogvideo_video, ltx_video_local

Setup:

make install-gpu

Then set in .env:

VIDEO_GEN_LOCAL_ENABLED=true
VIDEO_GEN_LOCAL_MODEL=wan2.1-1.3b

Requires NVIDIA GPU. Models vary by VRAM (6GB+ for entry-level).

Local Diffusion#

Tool: local_diffusion

Setup:

pip install diffusers transformers accelerate torch

First run downloads the model (~4GB). Runs offline on GPU.

Verification#

Run preflight before production:

make preflight

Or:

python -c "
from tools.tool_registry import registry
import json
registry.discover()
print(json.dumps(registry.provider_menu_summary(), indent=2))

This reports configured capabilities, missing keys, and available runtimes.

All paid operations go through cost estimation and reservation. See Cost Tracking for details.