We build the AI layer for real products.
FlooxAI is a product studio. We design, build and ship software powered by large language models and autonomous agents — from commerce automation and relationship coaching to creative tooling and LLM safety infrastructure. Below are the apps and platforms we run.
- 8
- products shipped
- 6
- live in production
- LLM-native
- from the ground up
What we do
One studio, the whole AI stack.
Everything FlooxAI builds runs on the same foundation — large language models, autonomous agents and the engineering to make them dependable in production.
AI applications
Full-stack products with intelligence built into the core — not bolted on. Web, mobile and API, end to end.
LLM integration
Claude and frontier models wired into your stack — tool calling, retrieval, structured output, streaming, and the guardrails around them.
Agentic chat
Conversational agents that take real actions — calling your services, running multi-step plans, and streaming their thinking back to the user.
AI workflows
Visual DAGs of typed nodes — reasoning, actions, conditions, human approval — executed by agents, streamed live, paused and resumed on demand.
Products & platforms
What we’ve shipped.
Eleven products across commerce, finance, real estate, healthcare, recruiting, relationships, creativity, sports and developer infrastructure — each one an AI idea taken all the way to a real, running app.
Retailers short-pay you. Netbalance gets it back.
A commercial claims-resolution desk for suppliers that sell through retailers and distributors. Claude Sonnet reads every remittance and supplier agreement into structured terms, then judges each deduction against a 491-code claims taxonomy — accept, dispute or ask for evidence. A streaming assistant (WebSocket) works across 12 toolchains with pgvector search over the customer’s documents, while Celery workflows import mailboxes, draft disputes and evidence packets, chase replies and match repayments on the bank statement. The customer steps in only for judgment calls.
- Claude remittance & contract reader
- Deduction judging · 491 claim codes
- Streaming assistant · 12 toolchains
- Auto-drafted disputes & follow-up
Every new development in Brazil, in one format.
An investor portal that puts Brazil’s new-construction market side by side — 10,900+ developments across 183 cities, with the price and status of nearly 900k units, maps, and a PDF per development with a rental estimate. A scraping pipeline keeps the catalog fresh and mirrors 290k+ photos and floor plans. On top sits a Claude Sonnet assistant, streamed over SSE, that picks from 38 tools in 6 toolchains to search, compare, shortlist and send leads by email or WhatsApp — and builds its own automations: graphs of agent, tool, condition, loop and human-approval nodes run by Celery, each agent tiered Haiku → Sonnet → Opus.
- Claude assistant · 38 tools
- Agent-built automations
- 10.9k developments · 898k units
- Map search & investor PDFs
From policy review to the point of care, in one governed workflow.
Built for regulated clinical settings, Karibu turns a facility’s own policies into guidance staff can trust. An LLM builds a knowledge tree from uploaded handbooks that admins approve value by value, and an alignment engine surfaces conflicting or outdated policies for human review. Approved knowledge reaches the floor as five-minute lessons taught by voice avatars (GPT-5.1 with Deepgram speech streamed in and out, in English and Spanish) and an ask-anything assistant whose answers are source-bounded — retrieved only from approved policies, with the source behind every reply.
- Policy alignment & conflict detection
- Source-bounded RAG assistant
- Voice-avatar microlearning
- Human-reviewed knowledge
Run your Shopify store in plain language.
A Claude-powered agent that operates a Shopify store through 20+ Admin-API tools — products, inventory, orders, fulfillment, shipping and email. Merchants chat in plain English and watch the agent reason and act in real time over a WebSocket. A visual builder turns repeatable jobs into DAGs of typed nodes (reasoning, action, condition, human-approval, loop) that run on a schedule, each node sized to the right Claude tier (Haiku → Sonnet → Opus).
- Claude agent · 20+ store tools
- Streaming reasoning chat
- Visual workflow DAGs
- Scheduled & human-approval runs
An AI coach that helps couples connect deeper.
Cora is a LangGraph ReAct agent (GPT-4o) that coaches each partner privately. She streams warm replies over WebSocket and calls scoped read/write tools to see the couple’s moods, goals and memories — then acts: logging moods, saving memories and generating mood-aware care and intimacy ideas, each scored across friendship, love, desire and passion. Proactive Celery flows nudge each partner with fresh, personalized ideas day to day.
- Cora · LangGraph ReAct agent
- Streaming coaching chat
- Scoped read/write tools
- Proactive mood-aware nudges
A real-time circuit breaker for rogue LLM sessions.
The safety layer for everything agentic. Turnstile is a transparent proxy in front of any LLM provider that meters every agent session in real time and trips a circuit breaker the instant one goes rogue — a runaway tool-calling loop or a blown budget — without taking down the rest of your app. Swap one base URL; keep your key.
- Per-session kill switch
- Runaway agent-loop detection
- Pre-execution budget caps
- Drop-in proxy, no SDK
Apply to 10 jobs in the time it takes to apply to one.
Upload a resume and a fleet of agents takes over. A streaming chat assistant (OpenAI GPT-5.2) wields 21 tools to search jobs and companies, find hiring-manager emails and apply on your behalf, while background agents tailor your resume and cover letter to each role, extract required skills to score fit, and write personalized outreach campaigns sent straight through Gmail.
- 6-agent fleet · 21 tools
- Streaming job-search chat
- Auto-apply & resume tailoring
- AI email outreach
Local groceries, sold from your neighbors’ homes.
A community marketplace where neighbors run grocery stores from home — with a dedicated AI service running through it end to end. Kitchi, a streaming shopping assistant (GPT-4o-mini), searches the catalog and fills carts through 13 tools; a Claude-Opus founder assistant answers analytics questions; and generative models produce store promos — images (Gemini) and video (Seedance) — plus invoice-to-catalog matching for sellers.
- Kitchi AI shopping assistant
- AI promo image & video gen
- Invoice→catalog matching
- Claude analytics assistant
Safe secondhand selling with local pickup points.
A marketplace that fixes the logistics and safety of buying secondhand: a network of neighborhood “warehouses of the people” where sellers drop off inventory and buyers inspect items in person. The AI roadmap turns listing into a single photo — vision models draft the title, condition and category and suggest a fair price — while semantic search and a chat-safety agent flag scams and keep the marketplace clean.
- Photo → listing vision assistant
- AI fair-price suggestions
- Semantic search
- Chat-safety / scam detection
From the first idea to a finished film — one platform.
An agentic writing co-pilot (GPT-4o) that carries one story from idea to finished film. It loads capability packs on demand to generate characters and locations, draft chapters and scenes, adapt them into screenplays, and build pre-production breakdowns and storyboards — streaming every tool call live over WebSocket. It even renders storyboard key-frames (Gemini) and animates them into clips (Seedance), with approval gates before anything commits.
- GPT-4o agent · capability packs
- Idea → manuscript → screenplay
- Storyboard image generation
- Image-to-video clips
Predict the World Cup. Climb the leaderboard.
A competitive World Cup prediction game: players guess scores, goal-scorers and squads across every round and race up a live leaderboard. The AI layer adds a prediction copilot — an agent that reads form, head-to-head history and squad news to explain the odds and suggest picks — and turns each result into an instant natural-language recap.
- Prediction copilot agent
- Form & head-to-head reasoning
- Natural-language recaps
- Live leaderboard
Have an idea that needs an AI brain?
We take AI products from concept to a live, dependable app — models, agents, workflows and the engineering around them. Tell us what you’re building.