RE::AI FLOW

-

Queries

-

Tokens

-

Tok/Q

-

Sessions

-

Avg Lat

-

Max Lat

-

Good

-

Bad

Mode	Queries	Tokens	Tok/Q	Sessions	Avg Lat	Good	Bad

Loading...

0

Queries

0

Tokens

0

Tok/Q

-

Avg Lat

-

Min Lat

-

Max Lat

0

Good

0

Bad

$0

Avg Cost

$0

Cost/1M

100%

Logs

[Ready]

Loading...

User Tiers Management

🤖 Agent API Keys

API keys for LLM providers used by agents

Vertex AI (Google) Not configured

OpenAI Not configured

Anthropic Not configured

Mistral Not configured

🔍 Search & Grounding

API keys for web search and grounding features

Google Search API Not configured

Search Engine ID Not configured

💾 Storage & Logging

Configure where logs and data are stored

Storage Type BigQuery Active

BigQuery Dataset Connected

PostgreSQL Connection Not configured

🔗 Deployed Agent Endpoints

API endpoints for deployed flows - use these in your applications

re::dact

POST /api/v1/agents/redact

re::check

POST /api/v1/agents/recheck

re::search

POST /api/v1/agents/research

re::doc

POST /api/v1/agents/redoc

🔑 API Access Keys

Generate API keys for external access to your agents

Production Key

rk_live_****************************

Development Key

rk_test_****************************

🔐 Custom Secrets / Environment Variables

Define custom secrets for use in flows and agents

-

Today Total

-

Accepted

-

Rejected

-

Errors

-

Avg Quality

-

Sources

-

Nano Cost

Languages

📁All Sources0

All Sources

Loading...

Recent Articles 0

Loading...

-

Total Archived

-

Today

-

Errors

-

Avg Quality

-

Running

Languages

📁All Sources0

All Sources

Loading...

Recent Archived 0

Loading...

-

Live

-

Archive

-

App Logs

-

Uploaded

-

Total

-

Tokens

BigQuery Datasets

Loading...

Select a dataset

Select a dataset to view samples

Page 1

🧪 Dataset Recipe Builder

Recipe Name

Source Proportions (100%)

📰 Live: 30%

📦 Archive: 30%

💬 Logs: 20%

📤 Upload: 20%

Train

Val

Test

📋 Recipes

No recipes yet

📤 Uploaded Files

No files uploaded

🧠 Training OFF

0

MODELS

0

JOBS

$0

TODAY

Current Production Model

No production model set

Trained Models

No models trained yet

Training Jobs

No active jobs

✨ Distillation OFF

0

TOPICS

0

GENERATED

0

APPROVED

Topics

No topics. Add one to start generating.

Generation Config

Languages

Styles

Per Combo

Select topics to see estimate

Recent Texts

No synthetic texts yet

⚡ Spark Catcher v4.0

Cross-encoder local → Gemini quote extraction

🏠 LOCAL Cross-Encoder (FREE)

Xenova/ms-marco-MiniLM-L-6-v2 BROWSER

Cross-encoder reranking (~23MB) — same model as server Alert Generator

Scores (project, page) pairs locally → only matched projects go to Gemini

Threshold: -1.0 (logit score)

☁️ Phase 1: Cloud Retrieval

Embedding Model

Similarity Threshold

0.45

Max Chunks

Chunk Type

Chunk Size (chars)

🎯 Phase 2: Verification

Provider

Model

Temperature

0.1

Attention Threshold

0.70

Reasoning Mode Chain-of-thought (wolniejsze, dokładniejsze)

Pipeline v4.0

1. Cross-Encoder (browser, FREE)
Xenova/ms-marco-MiniLM-L-6-v2 (~23MB)
Scores (project, page) pairs locally

2. Gemini Phase 2 (cloud)
gemini-2.5-flash-lite via Vertex AI
Extracts precise quotes with positions

Flow: Cross-encoder filters → matched projects → Gemini quotes → highlights

📡 Page Monitor

Scan Interval

📋 Active Config v4.0

Loading...

🎤 Whisper Voice-to-Text

Enable voice transcription in RE::DACT and Sparks editors. Uses OpenAI Whisper API.

🤖 Model Selection

Whisper Model

Default Language

Output Format

⚙️ Recording Settings

Max Recording Duration

Audio Quality

Auto-punctuation

📊 API Status

Not checked

Click "Test API" to verify connection

🎙️ Test Recording

Click to start test recording

📈 Usage Statistics

0

Today

0:00

Total Time

--

Avg Accuracy

$0.00

Est. Cost

⌨️ Keyboard Shortcuts

Start/Stop Recording Ctrl+Shift+R

Cancel Recording Escape

Insert at Cursor Enter

—

Total Visits

—

Unique Visitors

—

Avg Time on Page

—

Bot Requests

Pages

Page	Visits	24h	7d	30d	Unique	Avg Duration	Max Duration
Loading...

Countries

Country	Visits	%
Loading...

Power Visitors

IP	Country	Pages	Total Time	Last Visit	Browser/OS	Pages Visited
Loading...

Bot Traffic

Bot Name	Requests	Last Seen	Pages Crawled
Loading...

Loading reports...

LLM Analysis 0 selected

1 Mode

2 Platform

3 Model

4 Data

5 Params

6 Review

Select Training Approach

Choose how you want to create or adapt your model

🎯 LoRA / PEFT Adapter RECOMMENDED

Train a small adapter on top of any HuggingFace model. Memory efficient, fast, keeps base model frozen.

Time: 1-4 hours

Cost: $1-10

Data: 100-10k samples

Memory: ~8GB VRAM

Best for: Chat fine-tuning, style transfer, domain adaptation

☁️ Vertex AI Native Tuning GOOGLE MANAGED

Use Google's managed supervised fine-tuning for Gemini models. No GPU management, automatic scaling.

Time: 2-8 hours

Cost: $5-50

Data: 100-10k samples

Memory: Managed

Best for: Gemini fine-tuning, production deployments

Supported: gemini-2.5-flash, gemini-2.5-pro, gemini-2.0-flash

🔧 Full Fine-tune Open Source FULL CONTROL

Full parameter fine-tuning of open source models (Mistral, Llama, Phi). Updates ALL weights.

Time: 4-48 hours

Cost: $20-500

Data: 1k-100k samples

Memory: A100 40/80GB required

Best for: Major domain shift, own hosting

Supported: Mistral 7B, Llama 3 8B, Phi-3, Qwen2, DeepSeek

📊 Embedding Model Fine-tuning SEMANTIC SEARCH

Fine-tune sentence-transformers for semantic search, similarity, and retrieval tasks.

Time: 1-8 hours

Cost: $2-20

Data: 1k-100k pairs

Memory: ~8-16GB VRAM

Best for: RE::SEARCH, article similarity, RAG

Supported: MiniLM, E5, BGE, MPNet, multilingual-e5

🏗️ Train from Scratch EXPERT ONLY

Build a completely new model architecture from random weights. Requires massive data and compute.

Time: 24-168+ hours

Cost: $100-10,000+

Data: 100k-10M+ samples

Memory: Multi-A100/H100

Best for: Research, proprietary architecture

Consider fine-tuning existing model first - this is rarely needed

Quick Comparison

Approach	Speed	Cost	Data Needed	Best For
🎯 LoRA	⚡⚡⚡	💰	100-10k	Quick adaptation, limited resources
☁️ Vertex Native	⚡⚡	💰💰	100-10k	Gemini production, managed infra
🔧 Full FT OSS	⚡	💰💰💰	1k-100k	Major domain shift, own hosting
📊 Embeddings	⚡⚡	💰	1k-100k pairs	Semantic search, similarity
🏗️ Scratch	🐢	💰💰💰💰	100k+	Research, proprietary arch

Configure Infrastructure

Set up your training environment

GCP Project ID *

Region

Machine Type

GPU

Staging Bucket *

gs://

Service Account (optional)

Estimated: $0.73/h

Select Training Data

Choose or prepare your training dataset

Loading datasets...

Selected: 0 Samples: 0 Languages: - Size: 0 MB

Data Format

Configure Training Parameters

🎯 LoRA Configuration

LoRA Rank (r) Higher = more capacity, more memory

LoRA Alpha Usually 2x-4x rank

LoRA Dropout

Target Modules

Use QLoRA (4-bit quantization) Reduces memory by ~75%

☁️ Vertex AI Tuning

Epochs Recommended: 1-5

Adapter Size

LR Multiplier

📊 Training Configuration

Epochs

Batch Size

Learning Rate

Warmup Steps

Grad Accumulation Effective batch = batch × accum

Max Seq Length

Review & Start Training

🔍 Pre-flight Checks

⏳Platform Connection-

⏳Storage Access-

⏳Model Availability-

⏳Dataset Format-

⏳Quota & Limits-

📋 Configuration Summary

ModeLoRA

PlatformVertex AI

Model-

Dataset-

📤 Output

Model Name *

Description

Push to HuggingFace Hub Run evaluation Email notification

💰 Estimate

Compute$0.00

Storage$0.10

Total$0.00

Duration~2 hours