neverdown

Automatic failover across generative AI providers. Never let your app go down when Fal or Replicate does.

When Fal is down, neverdown automatically switches to Replicate. Your users never see an error.

from neverdown import generate

result = await generate(
    prompt="cinematic portrait of a woman in paris",
    model="flux-dev",
    providers=["fal", "replicate"],  # tries fal first, auto-switches on failure
)

print(result.image_url)   # the image
print(result.provider)    # which provider served it
print(result.latency_ms)  # how long it took

The problem

Your app depends on generative AI providers that go down — Fal, Replicate, RunPod. When they do, your app breaks. You either wait it out or manually switch providers. Neither is acceptable in production.

How it works

1. Submit job to first provider
2. Poll asynchronously — no blocking
3. If provider fails or times out → automatically retry on next provider
4. Failed providers enter a 60s cooldown so they're not hammered while down
5. Return result + which provider served it

Installation

pip install neverdown

Setup

Set your API keys as environment variables:

export FAL_KEY=your_fal_key
export REPLICATE_API_TOKEN=your_replicate_token

Usage

Basic

import asyncio
from neverdown import generate

async def main():
    result = await generate(
        prompt="cinematic portrait of a woman in paris, golden hour",
        model="flux-dev",
    )
    print(result.image_url)

asyncio.run(main())

Custom provider order

result = await generate(
    prompt="...",
    model="flux-schnell",
    providers=["replicate", "fal"],  # try replicate first
)

Concurrent generations

import asyncio
from neverdown import generate

results = await asyncio.gather(
    generate(prompt="photo 1", model="flux-dev"),
    generate(prompt="photo 2", model="flux-dev"),
    generate(prompt="photo 3", model="flux-dev"),
)

Error handling

from neverdown import generate
from neverdown.exceptions import AllProvidersFailedError

try:
    result = await generate(prompt="...", model="flux-dev")
except AllProvidersFailedError as e:
    print(e.errors)  # {provider: exception} for each failure

Supported models

| Model | Key | |-------|-----| | FLUX.1 Dev | flux-dev | | FLUX.1 Schnell | flux-schnell | | FLUX.1 Pro | flux-pro |

Supported providers

| Provider | Env var | |----------|---------| | Fal.ai | FAL_KEY | | Replicate | REPLICATE_API_TOKEN |

GenerationResult

result.image_url    # str   — URL of generated image
result.provider     # str   — "fal" or "replicate"
result.model        # str   — model used
result.latency_ms   # float — end-to-end latency in milliseconds

Exceptions

| Exception | When | |-----------|------| | AllProvidersFailedError | Every provider in the list failed | | ProviderUnavailableError | Provider is down or unreachable (triggers cooldown) | | JobFailedError | Job was submitted but failed on provider side | | JobTimeoutError | Job exceeded the timeout (triggers cooldown) |

Contributing

Pull requests are welcome. To add a new provider, implement BaseProvider in neverdown/providers/ and register it in core.py.

License

Apache 2.0 — see LICENSE