Firebase-native. Vertex AI ready. Veo 3 approved.

Prompt in. The right model out.

For Firebase app builders using Realtime Database + Vertex AI.

FireGen turns Realtime Database into your universal Generative AI API. AI Request Analyzer understands your prompt and automatically selects the best model and valid parameters. Write a string. Watch results stream back. Polling, LROs, GCS URLs, and auth — handled.

See how it works Join waitlist

One-liner: “Generative AI as easy as writing to Firebase Realtime Database.”

1) Install

Add the FireGen Extension to your Firebase project.

2) Write

push("firegen-jobs", "your prompt")

3) Subscribe

onValue(jobRef, ...) → response.url

requested


// AI‑assisted mode: just write a prompt string
import { getDatabase, ref, push, onValue } from "firebase/database";

const rtdb = getDatabase();
const jobRef = push(ref(rtdb, "firegen-jobs"), "Vertical video of a waterfall with ambient sound");

// Subscribe to updates
onValue(jobRef, (snap) => {
  const job = snap.val();
  if (!job) return;
  if (job.status === "succeeded") {
    // Media outputs provide a signed URL under response.url
    player.src = job.response.url;
  } else if (job.status === "failed") {
    console.error(job.response.error?.message);
  }
});


// Explicit mode: send a structured job
import { getDatabase, ref, push, onValue } from "firebase/database";

const rtdb = getDatabase();
const jobRef = push(ref(rtdb, "firegen-jobs"), {
  status: "requested",
  request: {
    type: "video",
    model: "veo-3.1-fast-generate-preview",
    prompt: "a neon city cat skateboarding at night",
    duration: 8,
    aspectRatio: "16:9",
    audio: true
  }
});

onValue(jobRef, (snap) => {
  const job = snap.val();
  if (!job) return;
  if (job.status === "succeeded") {
    player.src = job.response.url;
  } else if (job.status === "failed") {
    console.error(job.response.error?.message);
  }
});

Zero SDK juggling. No model guessing. No schema errors. Just RTDB.

Demo coming soon

Watch “write a string → stream the result” in ~15 seconds.

New AI Request Analyzer

Semantic prompt‑to‑model routing

FireGen understands your natural‑language prompt, picks the best model across Veo/Imagen/Gemini/Lyria/TTS, and fills valid parameters automatically. It preserves your prompt verbatim and saves the reasoning trail for transparency.

Two‑phase analysis (candidates → final) with validation & retry
Zod‑validated requests from a single model registry
URL‑aware (image‑to‑video) and multilingual phrasing
Reasoning chain stored under _meta.reasons

Input (RTDB write)

"Vertical video of a waterfall with ambient sound"

Transformed Job (server)

{
  "status": "requested",
  "request": {
    "type": "video",
    "model": "veo-3.1-fast-generate-preview",
    "prompt": "Vertical video of a waterfall with ambient sound",
    "duration": 8,
    "aspectRatio": "9:16",
    "audio": true
  },
  "_meta": {
    "aiAssisted": true,
    "reasons": ["…two-step reasoning chain…"]
  }
}

Reasoning Chain (_meta.reasons)

[
  "Step 1: Top 3 candidates with parameters and confidence…",
  "Selected Model: veo-3.1-fast-generate-preview with final parameters…",
  "Validation: (if any) Zod error corrected on retry…"
]

Stored at firegen-jobs/{jobId}/_meta/reasons for transparency and debugging.

Step 1

Install Extension

Deploy FireGen to your Firebase project (set region and bucket). Works with RTDB.

Step 2

Write to RTDB

AI‑assisted: write a string prompt. Explicit: send a structured request. Both to firegen-jobs .

Step 3

Listen for results

Use onValue to read response.url (media) or response.text (text/STT) when status is succeeded.

View on GitHub Join waitlist

How it works

A Firebase‑native pattern. Use AI‑assisted mode (string prompt) or explicit mode (structured request).

Client writes to RTDB

AI‑assisted: write a string prompt to firegen-jobs/{jobId} . Explicit: write a structured object.

AI analyzes & selects model

The AI Request Analyzer picks the best model and valid parameters, then transforms your node into a structured job.

Client listens for results

Functions v2 + Task Queue handle sync/async models. You subscribe and read response.url or response.text when succeeded.

Client         ──▶  RTDB /firegen-jobs  ──▶  Functions (onCreate)
  │                              │
  │         ◀────────────────────┘  analyze prompt → structured request
  │
  │                              ┌─▶ Task Queue Poller ──▶ Vertex Operations.get()
  │                              │          │
  │                              └──────────┴── (backoff, TTL, retries)
  │
  ◀──────── real-time updates ◀── RTDB /firegen-jobs (status/response/error)

Why teams choose FireGen

DX‑first, Firebase‑native, now with AI‑assisted routing.

AI Request Analyzer

Prompt‑to‑model routing with validation. No more guessing IDs or params; FireGen sets them correctly and saves the reasoning.

Firebase-native

Realtime triggers, Task Queue, Functions v2, secure rules. No extra infra.

LROs handled

Exponential backoff + jitter. Cancel, TTL, and dead-letter patterns baked in.

Big-file friendly

Veo writes to GCS via storageUri. We return signed URLs you can stream.

Universal schema

Same job shape for video, image, text, audio. Sync and async flows unified.

Secure by default

Rules restrict per-user reads/writes. App Check ready. Keys kept server-side.

Drop-in Extension

Install, set bucket & location, done. Write a string or an object — your choice.

Supported models (auto‑routed)

Veo 3 / Veo 3 Fast (Video) Imagen 4 (Image) Gemini 2.5 (Text / Multimodal) Lyria 2 (Music) Gemini TTS (Speech)

Why not DIY?

We’ve been there: multiple SDKs, outdated examples, polling edge cases, storage hand-offs, and auth. FireGen compresses days into minutes.

No OIDC or queue plumbing — Firebase Task Queue does it for you.
No guessing async vs sync — we normalize both flows.
No storage headaches — signed URLs ready to stream.

Approach	Dev Effort	Time-to-first-result
Direct Vertex SDKs	High (LROs, storage, auth, docs)	Days → weeks
Workflows / Scheduler DIY	Medium (new services, YAML/cron)	Days
FireGen (Extension)	Low (write node + subscribe)	Minutes

FAQ

Quick answers for builders.

Prompt in. The right model out.

Semantic prompt‑to‑model routing

Install Extension

Write to RTDB

Listen for results

How it works

Client writes to RTDB

AI analyzes & selects model

Client listens for results

Why teams choose FireGen

AI Request Analyzer

Firebase-native

LROs handled

Big-file friendly

Universal schema

Secure by default

Drop-in Extension

Supported models (auto‑routed)

Why not DIY?

Pricing

Starter

Pro

FAQ

Get early access

Prompt in. The right model out.

Semantic prompt‑to‑model routing

Install Extension

Write to RTDB

Listen for results

How it works

Client writes to RTDB

AI analyzes & selects model

Client listens for results

Why teams choose FireGen

AI Request Analyzer

Firebase-native

LROs handled

Big-file friendly

Universal schema

Secure by default

Drop-in Extension

Supported models (auto‑routed)

Why not DIY?

Pricing

Starter

Pro

FAQ

Can I just send a prompt and let FireGen decide the model?

Does FireGen work with Veo 3 long-running operations?

What about synchronous models like Lyria or TTS?

Can I host outputs in Firebase Storage instead of GCS?

Is there a codename?

Get early access