Gabriel AI Orchestrator

Gabriel is FOTOhub's intelligent platform orchestrator — a proprietary AI layer that classifies user intent, routes requests to optimal features, selects the best available model, builds prompts, and orchestrates multi-step workflows.

Gabriel AI is free to use for authenticated users and provides real-time suggestions, streaming responses, and proactive recommendations.

All authenticated Gabriel endpoints accept your standard API key (fh_live_*) via the Authorization: Bearer header — the same key you use for every other endpoint. A Supabase session JWT is also accepted (used by the web dashboard), but no separate credential is required for API integrations.

Endpoints

Method	Path	Auth	Description
POST	/v1/ai/gabriel	API key or JWT	Classify intent and route (single-shot)
POST	/v1/ai/gabriel/stream	API key or JWT	Streaming orchestration (SSE)
POST	/v1/ai/gabriel/suggest	None	Lightweight autocomplete suggestions
POST	/v1/ai/gabriel/recommend	None	Proactive context-aware recommendations
POST	/v1/ai/translate	API key or JWT	Translate text

---

POST /v1/ai/gabriel

Classify user intent and return a routing decision with optimal model selection, credit estimation, and contextual tips.

Rate limit: 30 requests/minute per user

Authentication: API key (fh_live_*) or Supabase session JWT

Request Body

Field	Type	Required	Description
prompt	string	Yes	User's natural language request (max 1000 chars)
language	string	No	Language code (default: "pl")
context	object	No	Additional context for better classification
enhance_prompt	boolean	No	When true, Gabriel enriches the prompt using model-specific architecture knowledge

Context Object

Field	Type	Description
user_tier	string	User's subscription tier ("free", "pro", "business")
wallet_balance_usd	float	Current prepaid USD wallet balance
recent_tools	string[]	Last 5 features the user used
brand_id	string	Active brand kit ID (for brand-aware suggestions)

Response

Field	Type	Description
action	string	"route" | "answer" | "workflow" | "error"
target	string	Feature path to navigate to (e.g., /generate/image)
params	object	Pre-configured parameters for the target feature
model_selected	string	Optimal available model for the request
suggested_actions	array	Alternative actions the user might want
confidence	float	Classification confidence (0.0 - 1.0)
credits_estimated	integer	Estimated credit cost
tips	string[]	Contextual prompt engineering tips

Actions

Action	Description
route	Navigate user to a FOTOhub feature with pre-filled parameters
answer	Direct text response (FAQ, help, platform guidance)
workflow	Multi-step task requiring sequential operations
error	Classification failed or timed out

Examples

Python

from fotohub import FotoHub

client = FotoHub(api_key="your-api-key")

result = client.gabriel.classify(
    prompt="Generate a cinematic photo of a sunset over mountains",
    language="en",
    enhance_prompt=True
)

print(result)
# {
#   "action": "route",
#   "target": "/generate/image",
#   "params": {
#     "model": "seedream-5-0-260128",
#     "prompt": "a sunset over mountains, cinematic color grading, golden hour, dramatic volumetric clouds, 8K, ultra detailed, shallow depth of field"
#   },
#   "model_selected": "seedream-5-0-260128",
#   "confidence": 0.95,
#   "credits_estimated": 1,
#   "tips": ["Seedream excels at photorealism — add '8K, sharp detail' for best results"]
# }

TypeScript

import { FotoHub } from 'fotohub';

const client = new FotoHub({ apiKey: 'your-api-key' });

const result = await client.gabriel.classify({
  prompt: 'Make a 5-second video of a cat walking through a garden',
  language: 'en',
  context: { wallet_balance_usd: 50 }
});

// {
//   "action": "route",
//   "target": "/generate/video",
//   "params": { "model": "seedance-2-0-pro", "duration": 5, "prompt": "..." },
//   "model_selected": "seedance-2-0-pro",
//   "confidence": 0.92,
//   "credits_estimated": 10,
//   "tips": ["Seedance: describe camera movement for dynamic videos"]
// }

cURL

curl -X POST https://apis.fotohub.app/v1/ai/gabriel \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer fh_live_YOUR_API_KEY" \
  -d '{
    "prompt": "Create a brand identity for my coffee shop",
    "language": "en",
    "enhance_prompt": true,
    "context": {
      "user_tier": "pro",
      "wallet_balance_usd": 100
    }
  }'

---

POST /v1/ai/gabriel/stream

Streaming orchestration via Server-Sent Events (SSE). Provides progressive feedback as Gabriel processes the request — reduces perceived latency and enables real-time UX.

Rate limit: 30 requests/minute per user

Authentication: API key (fh_live_*) or Supabase session JWT

Request Body

Same as /v1/ai/gabriel (prompt, language, context).

SSE Event Types

Events are sent as data: {json}\n\n lines. The stream ends with data: [DONE]\n\n.

Event Type	Description	Fields
thinking	Gabriel is processing	content: status message
routing	Intent classified, routing in progress	tool: tool being called, content: description
result	Final routing decision	Same fields as single-shot response
error	Processing failed	message: error description

Example

const response = await fetch('https://apis.fotohub.app/v1/ai/gabriel/stream', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'Authorization': `Bearer ${token}`
  },
  body: JSON.stringify({
    prompt: 'Generate a portrait photo with dramatic lighting',
    language: 'en'
  })
});

const reader = response.body.getReader();
const decoder = new TextDecoder();

while (true) {
  const { done, value } = await reader.read();
  if (done) break;

  const text = decoder.decode(value);
  for (const line of text.split('\n')) {
    if (line.startsWith('data: ')) {
      const data = line.slice(6);
      if (data === '[DONE]') break;

      const event = JSON.parse(data);
      switch (event.type) {
        case 'thinking':
          console.log('Processing:', event.content);
          break;
        case 'routing':
          console.log('Using tool:', event.tool);
          break;
        case 'result':
          console.log('Route:', event.target, event.model_selected);
          console.log('Tips:', event.tips);
          break;
      }
    }
  }
}

Stream Timeline

Client                    Gabriel
  |--- POST /stream ------->|
  |<-- thinking: "..." ------|  (immediate, <100ms)
  |<-- routing: {...} -------|  (after intent classification)
  |<-- result: {...} --------|  (final decision + tips)
  |<-- data: [DONE] ---------|

---

POST /v1/ai/gabriel/suggest

Lightweight autocomplete suggestions. No LLM call — pure in-memory fuzzy matching for sub-50ms response times. Use this for keystroke-speed suggestions as the user types.

Rate limit: 60 requests/minute per IP

Authentication: None required

Request Body

Field	Type	Required	Description
partial	string	Yes	Partial user input (min 2 chars, max 200)
tab	string	No	Current tab context: "all", "image", "video", "audio"
page	string	No	Current page path (e.g., /generate/new)

Response

Field	Type	Description
suggestions	array	Up to 5 ranked suggestions

Suggestion Object

Field	Type	Description
text	string	Suggestion text
category	string	"prompt" | "tip" | "model" | "feature"
target	string	Optional navigation target
icon	string	Optional icon identifier

Example

TypeScript

// Debounce to 300ms for optimal UX
const suggestions = await fetch('https://apis.fotohub.app/v1/ai/gabriel/suggest', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    partial: 'portrait photo',
    tab: 'image',
    page: '/generate/new'
  })
}).then(r => r.json());

// {
//   "suggestions": [
//     { "text": "portrait photo, studio lighting, shallow depth of field, 8K", "category": "prompt", "icon": "camera" },
//     { "text": "Try Seedream for best portrait quality (1 credit)", "category": "tip", "icon": "sparkles" },
//     { "text": "Use face enhancement for portrait shots", "category": "feature", "target": "/tools/enhance", "icon": "wand" }
//   ]
// }

cURL

curl -X POST https://apis.fotohub.app/v1/ai/gabriel/suggest \
  -H "Content-Type: application/json" \
  -d '{"partial": "sunset", "tab": "image"}'

Integration Pattern

// Recommended: debounced suggestions with local fallback
let debounceTimer: number;

function onInputChange(value: string) {
  // Show local suggestions immediately
  showLocalSuggestions(value);

  // Fetch API suggestions after 300ms pause
  clearTimeout(debounceTimer);
  debounceTimer = setTimeout(async () => {
    if (value.length >= 2) {
      const { suggestions } = await fetchSuggestions(value);
      if (suggestions.length > 0) {
        replaceSuggestions(suggestions); // API results take priority
      }
    }
  }, 300);
}

---

POST /v1/ai/gabriel/recommend

Proactive context-aware recommendations. Returns 1-3 relevant tips based on the user's current state. Template-based (no LLM call), responds in <100ms.

Rate limit: 30 requests/minute per IP

Authentication: None required

Request Body

Field	Type	Required	Description
page	string	No	Current page path
recent_actions	string[]	No	Last few actions taken
wallet_balance_usd	float	No	User's prepaid USD wallet balance
has_brand	boolean	No	Whether user has a brand kit

Response

Field	Type	Description
recommendations	array	Up to 3 contextual recommendations

Recommendation Object

Field	Type	Description
text	string	Recommendation text
target	string	Navigation target
icon	string	Icon identifier

Example

curl -X POST https://apis.fotohub.app/v1/ai/gabriel/recommend \
  -H "Content-Type: application/json" \
  -d '{
    "page": "/generate/new",
    "wallet_balance_usd": 5,
    "has_brand": false
  }'

# Response:
# {
#   "recommendations": [
#     { "text": "Low credits — Seedream gives best quality at 1 credit", "target": "/generate/new", "icon": "coins" },
#     { "text": "Create a Brand Kit for consistent style across generations", "target": "/brand", "icon": "palette" }
#   ]
# }

---

POST /v1/ai/translate

Translate text between languages using FOTOhub's built-in translation engine.

Rate limit: 30 requests/minute per user

Authentication: API key (fh_live_*) or Supabase session JWT

Request Body

Field	Type	Required	Description
text	string	Yes	Text to translate (max 10,000 chars)
target_language	string	Yes	Target language code (e.g., "en", "pl", "de")
source_language	string	No	Source language (auto-detected if omitted)

Response

Field	Type	Description
translated_text	string	The translation
source_language	string	Detected/specified source language
target_language	string	Target language
character_count	integer	Character count of translated text

Examples

Python

from fotohub import FotoHub

client = FotoHub(api_key="your-api-key")

result = client.translate(
    text="The quick brown fox jumps over the lazy dog",
    target_language="pl"
)
# { "translated_text": "Szybki brązowy lis przeskakuje nad leniwym psem", ... }

TypeScript

const response = await fetch('https://apis.fotohub.app/v1/ai/translate', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    text: 'Bonjour le monde',
    target_language: 'en',
    source_language: 'fr'
  })
});

const { translated_text } = await response.json();
// "Hello world"

cURL

curl -X POST https://apis.fotohub.app/v1/ai/translate \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Hallo Welt",
    "target_language": "en"
  }'

---

Orchestrator Capabilities

Gabriel provides 10 function-calling tools for intelligent platform routing:

Tool	Description
route_to_image_generation	Route to image generation with optimal model + params
route_to_video_generation	Route to video generation with model + duration
route_to_chat	Route to LLM chat with selected model
route_to_music_generation	Route to music/audio generation
route_to_image_editing	Route to image editing tools (upscale, bg remove, etc.)
route_to_3d_generation	Route to 3D model generation
answer_question	Direct text answer for FAQ/help
route_to_brand	Route to brand kit management
route_to_tools	Route to platform tools (face swap, lip sync, etc.)
create_workflow	Multi-step workflow for complex tasks

Dynamic Model Awareness

Gabriel only recommends models that are currently available. Model availability is checked against the platform's model status database (refreshed every 60 seconds). If a model is disabled or experiencing issues, Gabriel automatically selects the next best alternative.

Prompt Enhancement

When enhance_prompt: true is set, Gabriel applies model-specific prompt engineering:

Model Family	Enhancement Strategy
Seedream	Adds photorealistic quality terms, resolution hints, optimized structure
Seedance	Adds camera movement, temporal progression, cinematic terms
FLUX	Adds detailed composition, style-specific tokens
Imagen	Adds quality terms, structured scene description
Wan	Adds motion description, scene progression
Grok Imagine	Adds studio lighting terms, product photography markers, commercial quality

Multi-Step Workflows

When a request requires multiple operations, Gabriel returns action: "workflow" with a steps array:

{
  "action": "workflow",
  "params": {
    "steps": [
      { "action": "route_to_image_generation", "target": "/generate/new", "params": {"model": "seedream-5-0-260128", "prompt": "..."} },
      { "action": "route_to_image_editing", "target": "/tools/remove-bg", "params": {"tool": "background_removal"} },
      { "action": "route_to_brand", "target": "/brand", "params": {"action": "save_asset"} }
    ],
    "description": "Generate logo, remove background, save to brand kit"
  }
}

---

Credit Estimation

Gabriel estimates costs for 50+ models across image, video, and chat:

Image Models

Model	Credits/image
flux-2-klein-4b	1.0
minimax-image-01	1.0
grok-imagine-image	1.0
seedream-5-0-260128	2.0
seedream-4-5-251128	2.0
flux-1.1-pro	2.0
flux-kontext-pro	2.0
flux-2-pro	2.0
imagen-4-standard	3.0
grok-imagine-image-pro	3.0
imagen-4-ultra	5.0

Video Models

Most models bill per second (see the full catalog); MiniMax Hailuo bills a flat per-video amount.

Model	Credits/s
wan2.2-t2v-plus	1.2
hailuo-o2	— (6 flat)
kling-v3	5
seedance-2-0-pro	9.4
sora-2	8
gemini-omni-flash	6
veo-3.1-generate-001	12
grok-imagine-video-1.5	9

Chat / LLM Models

Model	Credits/message
gemini-flash	1.0
gemini-pro	2.0
gpt-4o	2.0
claude-sonnet	2.0

Best Value

seedream-5-0-260128 offers the best quality-to-credit ratio for image generation, and wan2.2-t2v-plus / hailuo-o2 are the most cost-effective for video. For chat, gemini-flash (1 cr) is the cheapest option.

---

Error Handling

All endpoints gracefully degrade on failure:

Scenario	Behavior
Timeout (>5s for Gabriel, >10s for Translate)	Returns error action / original text
Model error	Auto-failover to backup model (transparent to caller)
Rate limit exceeded	HTTP 429 with retry-after header
Invalid JWT	HTTP 401 (for authenticated endpoints)
Missing prompt	HTTP 422 with validation details

Model Failover

Gabriel uses a multi-model failover chain. If the primary model is unavailable, it automatically falls back without returning an error:

Primary → Backup 1 → Backup 2

The response always succeeds unless all models in the chain are down.

---

Integration Patterns

Building a Conversational Assistant

from fotohub import FotoHub

client = FotoHub(api_key="your-api-key")

def handle_user_message(message: str, user_context: dict):
    result = client.gabriel.classify(
        prompt=message,
        language="en",
        context=user_context,
        enhance_prompt=True
    )
    
    if result["action"] == "route":
        return {
            "message": f"I'll use {result['model_selected']} (~{result['credits_estimated']} credits)",
            "action": result["target"],
            "params": result["params"],
            "tips": result.get("tips", [])
        }
    elif result["action"] == "workflow":
        steps = result["params"]["steps"]
        return {
            "message": f"This requires {len(steps)} steps: {result['params']['description']}",
            "workflow": steps
        }
    elif result["action"] == "answer":
        return {"message": result["answer"]}
    else:
        return {"message": "I couldn't understand that request."}

Real-time Suggestions with Streaming

import { FotoHub } from 'fotohub';

const client = new FotoHub({ apiKey: 'your-api-key' });

// 1. Autocomplete as user types (debounced 300ms)
const suggestions = await client.gabriel.suggest('portrait photo', {
  tab: 'image',
  page: '/generate/new'
});

// 2. Stream the final classification
const stream = await client.gabriel.stream({
  prompt: 'Generate a cinematic portrait with dramatic lighting',
  language: 'en'
});

for await (const event of stream) {
  if (event.type === 'thinking') updateUI('Processing...');
  if (event.type === 'routing') updateUI(`Using ${event.tool}...`);
  if (event.type === 'result') navigateTo(event.target, event.params);
}

Proactive Onboarding Tips

// Show contextual tips when user lands on a page
const { recommendations } = await fetch(
  'https://apis.fotohub.app/v1/ai/gabriel/recommend',
  {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      page: '/generate/new',
      wallet_balance_usd: user.credits,
      has_brand: user.brands.length > 0,
      recent_actions: ['image_generation', 'image_generation']
    })
  }
).then(r => r.json());

// Display as floating chips:
// "Try video generation — Seedance creates 5s clips for 1 credit"
// "Create a Brand Kit for consistent style"

Multi-language App with Auto-translation

async function translateUI(texts: string[], targetLang: string) {
  const results = await Promise.all(
    texts.map(text =>
      fetch('https://apis.fotohub.app/v1/ai/translate', {
        method: 'POST',
        headers: { 'Content-Type': 'application/json' },
        body: JSON.stringify({ text, target_language: targetLang })
      }).then(r => r.json())
    )
  );
  return results.map(r => r.translated_text);
}