deAPI

deAPI is an OpenAI-compatible inference API for open-source AI models. It's built for developers who want to keep their OpenAI SDK and ship features OpenAI doesn't provide — image generation, image edits, video generation, music, voice cloning, transcription from URLs, OCR, and embeddings — without rewriting their integration code.

OpenAI-compatible by default. The gateway lives at [https://oai.deapi.ai/v1](https://oai.deapi.ai/v1). Switching is two lines:

client = OpenAI( api_key="dpn-sk-...", base_url="[https://oai.deapi.ai/v1"](https://oai.deapi.ai/v1"), )

The request body, response schema, and error envelope follow the OpenAI specification. images.generate, audio.speech.create, audio.transcriptions.create, embeddings.create, and an OpenAI-style POST /v1/videos extension all work without changes to call signatures. Works with the OpenAI Python SDK, OpenAI Node SDK, LangChain, LlamaIndex, Vercel AI SDK, n8n, and anything else that takes a base_url.

What it covers (non-LLM modalities, by design):

• Image: text-to-image, image-to-image, OCR, background removal, upscaling
• Video: text-to-video, image-to-video, audio-to-video, character replacement, upscaling
• Audio: text-to-speech with preset voices, voice cloning, voice design; text-to-music
• Transcription: unified endpoint, audio + video, file upload up to 80 MB, or URL ingest from YouTube, X/Twitter, Twitch, Kick, TikTok, X Spaces
• Embeddings: text-to-vector for RAG, semantic search, clustering
• Prompt enhancement endpoints for image, video, speech, and image-to-image prompts

Chat completions are intentionally out of scope — deAPI does not serve LLMs. It's the complementary half to your LLM provider, not a replacement.

Decentralized GPU network. Inference runs on a verified worker network rather than a single hyperscaler. Workers receive payloads in RAM only — no disk writes, no UI, encrypted in transit, with request hashing, binary verification, and optional dual-worker verification for sensitive workloads. Pricing reflects this: a Flux.1 schnell 512×512 image at 4 steps costs $0.00136 (Flux.2 Klein at the same size is $0.00187); transcription is $0.021 per hour; TTS is $0.77 per million characters; embeddings are $0.000068 per 1k tokens; background removal at 1024×1024 is $0.00035 per image.

Pricing model. Pay-as-you-go, no subscription. New accounts get $5 in free credits with conservative Basic rate limits. Any top-up unlocks Premium: 300 RPM across every endpoint, unlimited daily requests. The remaining bonus balance carries over.

For builders. Native REST v2 API, official Python SDK with sync + async clients and exponential backoff, MCP server for Claude Desktop / Cursor / ChatGPT, n8n community node, Claude Code skills, webhooks with HMAC-SHA256 signatures, and Pusher-compatible WebSockets for real-time job progress.