AI That Talks to Your Oven.
Restaurant Results, Every Time.
Agix built the intelligence layer that translates Innit's recipe knowledge into real-time appliance commands, setting oven temperatures, adjusting induction zones, and monitoring thermometers automatically so home cooks achieve professional results without guesswork.
A connected kitchen platform at the intersection of recipe intelligence and smart appliance control.
Founded in 2014 and headquartered in San Jose, CA, Innit partners with major appliance manufacturers, Samsung, Whirlpool, GE Appliances, Bosch, LG, to create an end-to-end guided cooking experience. Their platform bridges the gap between knowing a recipe and executing it flawlessly, turning connected appliances from expensive hardware into active cooking partners.
With integrations across 50M+ connected appliances and partnerships with 15+ appliance brands, Innit's vision is making home cooking as reliable as professional kitchen execution. Agix's role: build the AI orchestration layer that makes that vision real in real time.
43% recipe abandonment. 12% appliance utilization. A $300 smart oven operating like a dumb one.
Home cooking failures aren't caused by bad cooks, they're caused by imprecise instructions, missing feedback, and zero coordination between appliances. Every stove's "medium-high heat" is different. Nobody knows when golden-brown is golden-brown.
43% of home recipes fail, not because of skill, but because instructions don't translate to real kitchens.
Recipe instructions are written for an ideal environment that doesn't exist in anyone's kitchen. "Bake at 350°F" ignores that your oven runs 30° hot. "Cook until golden brown" assumes you know exactly what that looks like. "Medium-high heat" means nothing without knowing your stovetop's output. The result: 43% of attempted recipes fail or produce significant deviations from the intended result. For Innit, every failed recipe is a lost user, someone who tried the app once, had a bad meal, and never came back.
Multi-appliance coordination is cognitively overwhelming, oven, stovetop, thermometer, timer all at once.
A moderately complex recipe requires tracking the oven temperature, managing two stovetop zones at different heat levels, monitoring a meat thermometer, and following step-by-step instructions, all simultaneously, with both hands occupied and a kitchen filling with smoke. This is not a recipe problem; it's a cognitive load problem. Most home cooks can handle one stream of information at a time. Cooking a full meal demands five. Innit's guidance-only app helped with the instructions, but still left all the monitoring and adjusting to the cook.
Smart appliances were being used as dumb ones, only 12% of their features ever activated by owners.
Consumers spent $1,200 on a smart oven and used it exactly like a $300 one. The connectivity features, remote preheating, precision temperature probes, cooking mode selection, were buried in companion apps that nobody opened while cooking with floury hands and a hot pan. The hardware was ready. The software bridge between "what the recipe needs" and "what the appliance can do" didn't exist. Innit's appliance manufacturer partners were losing the value proposition of smart features before customers ever discovered them.
Innit FoodLM: the domain-specific validator that ensures every AI response is grounded in food science.
General LLMs hallucinate on food and nutrition. Innit FoodLM validates every output through four specialized models, Nutrition & Diets, Health Conditions, Personalized Shopping, and Culinary & Cooking, before it ever reaches a user.
Validates macro/micro accuracy, dietary compliance (vegan, keto, diabetic), and ingredient substitution safety across 10,000+ food items.
Flags recipes or ingredients that conflict with declared health conditions, diabetes, celiac, hypertension, allergies, before any recommendation surfaces.
Cross-references recipe ingredient lists with retailer inventory, pricing, and the user's pantry, generating optimized shopping carts with one tap.
Validates technique guidance, timing parameters, and appliance-specific instructions, ensuring "sear for 3 minutes" reflects actual performance data, not generic assumptions.
Six layers that turn a recipe into a coordinated, real-time cooking session across every appliance in your kitchen.
Each layer operates independently and as part of a unified cooking state, passing live sensor readings, appliance status, recipe phase, and user context through every step from ingredient prep to plating.
Recipe Knowledge Graph
A structured recipe database encoding exact appliance settings for each cooking step, calibrated by brand and model. "Medium-high heat" is no longer an ambiguity, for a GE induction cooktop it becomes 1,400W, for a Samsung it becomes a precise zone percentage, for a Bosch it becomes a numbered setting. The knowledge graph covers 50+ appliance models and continuously updates from real cook outcome data, improving accuracy with every session completed.
Multi-Appliance Orchestration Engine
Coordinates oven preheating, cooktop zone adjustment, smart thermometer monitoring, and range hood control across multiple appliances simultaneously, based on recipe state, not a clock. The orchestration engine knows that the oven should start preheating 20 minutes before the step that needs it, that the stovetop zone should reduce when the thermometer approaches target temperature, and that the range hood should activate the moment high-heat searing begins. All of this happens without a single tap from the cook.
Real-Time Sensor Integration
Live sensor data from connected meat thermometers, oven internal probes, and induction zone readings creates a continuous picture of cooking progress versus planned progress. When a chicken breast reaches 155°F eighteen minutes early because the oven runs hot, the system detects it, adjusts the oven setpoint to prevent overcooking, and updates every subsequent step's timing to reflect the new reality. No timer-based cooking system can do this, only one with real sensor input can respond to what's actually happening in the pan.
Adaptive Cooking Adjustments
When sensor data indicates cooking is ahead or behind schedule, the system automatically adjusts subsequent step timing and appliance settings, not just the current step. If the oven is slow to preheat, every downstream timing window shifts accordingly. If a sear is progressing faster than expected, the next step's rest period extends to compensate. This closed-loop adjustment is what separates the Agix-powered Innit experience from every timer-based recipe app on the market: the AI is responding to reality, not assumptions about how long things should take.
Voice-Guided Instruction System
Step-by-step voice guidance synchronized to actual cooking progress, not a fixed timer, designed for hands-busy cooking where a touch screen is unusable. The voice layer knows when the oven has finished preheating before telling the cook to put the dish in. It knows when the thermometer signals the correct internal temperature before saying "remove from heat." It can answer in-context cooking questions ("how do I know when the oil is ready?") with technique-specific guidance drawn from the culinary knowledge graph, not a generic search result.
Cooking Technique Teaching Engine
Beyond executing recipes, the system teaches. Why you shouldn't move searing meat for the first two minutes. How to tell by sound when oil reaches the right temperature. What "reduce by half" looks like visually after 8 minutes on medium. What the difference between a simmer and a boil means for a cream sauce. These micro-lessons are embedded in context, delivered at the moment the technique applies, tied to the specific recipe and appliance in use, building competence that carries over to the next session, even without the app.
Innit iQ: a modular food and wellness intelligence platform for brands, retailers, and pharma.
The intelligence layer Agix built doesn't just power Innit's consumer app, it's the foundation of Innit iQ, a B2B platform that lets CPG brands, retailers, pharmaceutical companies, and insurers embed the same food intelligence into their own products.
Innit Integrator™
Plug-and-play food and wellness modules for existing websites and apps (CPG, retail, pharma, insurance).
Innit App Builder
Full-featured app development using the iQ CMS and design team, delivering production apps in weeks.
Innit iQ CMS
A content management layer with built-in ChatAI, analytics, module management, and user growth tracking.
AI-generated recipes that are validated, nutritionally accurate, and calibrated to the cook's health profile.
The recipe intelligence layer doesn't just suggest dishes, it generates them from scratch using multi-model AI (GPT-4, Gemini, Imagen) and then validates every output through FoodLM before surfacing it. Every generated recipe includes precise nutritional data, ingredient quantities per serving, dietary flags, and appliance-specific cook parameters.
From recipe to grocery cart to kitchen, the full food journey, closed in one platform.
The retailer integration layer converts Innit's recipe intelligence into a complete shopping experience, ingredient-level cart building, ingredient swaps, meal planning, and one-click checkout, embedded directly in retailer apps and websites.
All 24 recipe ingredients mapped to specific retailer SKUs, with current pricing and quantity calculation, added to cart in a single action.
FoodLM-validated substitutions for unavailable or disliked ingredients, with adjusted cooking parameters for the swap, not just a different product.
Weekly meal planning with calendar sync, per-serving cost calculation, and consolidated grocery lists across the full week's recipes.
Saved recipes with pantry inventory tracking, so the next week's cart only orders what you're actually out of, eliminating waste and duplicate purchases.
Cooking success metrics across connected households.
Every number is measured against the same recipes without AI guidance, same dishes, same difficulty level, same user cohort.
vs 57% for the same recipes without AI guidance, a 35-point improvement in successful cook outcomes
Weekly cooking sessions per user increased 2.65× when appliance control activated vs guidance-only mode
Smart appliance feature usage jumped from 12% to near-100% when Innit orchestrated the cooking session
Post-cook satisfaction score vs 3.9/5 for manual cooking from the same recipes, nearly a full point improvement
I've tried to make beef Wellington three times in my life and failed every time. With Innit, I made it for my parents' anniversary and it was perfect. The oven literally changed temperature by itself at the right moment. I felt like a professional chef.
Six design decisions that separated the outcome from every recipe app before it.
Closing the Feedback Loop With Real Sensors
Without real temperature data from a connected thermometer, guidance is just a timer with extra steps. With actual sensor data from smart thermometers and oven probes, the system can respond to reality, not assumptions about how long things take in a generic kitchen.
Appliance-Specific Calibration
Translating generic recipe instructions into model-specific settings was the key insight. "Medium-high heat" on a GE induction cooktop is 1,400W. The system knows this for 50+ appliance models, and that knowledge is what converts a failed recipe into a successful one.
Pre-Activation Timing
Starting oven preheating 20 minutes before the step that requires it, rather than telling the user to preheat at the moment of need, removed one of the most common sources of timing failures in home cooking. The system knows the full recipe timeline and acts ahead of it.
Voice Interface for Hands-Busy Cooking
Touch interfaces fail when your hands are covered in flour. Voice guidance synchronized to actual cooking state, not a fixed timer, matched the real workflow of cooking for the first time. Users could ask questions, request repeats, and navigate steps without touching a screen.
Success Rate as the North Star Metric
Defining success as "dish came out as intended" rather than "user completed the session" forced every product decision toward actual cooking outcomes. Engagement, session length, and DAU are easy to optimize for. "Did the food taste right?" is harder, and far more meaningful.
FoodLM Domain Validation Layer
General LLMs are unreliable on food science. FoodLM, Innit's domain-specific validation layer, ensures that every AI-generated recipe and guidance step is reviewed by a model trained specifically on nutrition, health conditions, culinary technique, and personalized shopping, before it reaches the user.
What powers this system.
Agentic AI Systems
Multi-step cooking orchestration across connected appliances, autonomous action execution triggered by real sensor data, not fixed timers or manual input.
Computer Vision
Visual cooking state detection, identifying browning, doneness, and consistency from camera feeds for cues that sensors alone can't capture.
Conversational AI
Voice-guided cooking instruction with context awareness, answers cooking questions, navigates steps, and adapts guidance to the actual state of the food.
Retail AI Solutions
Recipe-to-cart intelligence for grocery retailers, ingredient mapping, real-time inventory, personalized substitutions, and one-tap shopping cart generation.
Operational AI
Real-time multi-device orchestration and process automation, the core intelligence type that drives Innit's appliance coordination engine.
Custom AI Product Development
End-to-end AI product builds for consumer platforms, from architecture to deployment, including domain-specific model fine-tuning and IoT integration.
Common questions about building connected kitchen AI systems like the one deployed at Innit.
Innit has certified integrations with 15+ major appliance brands including Samsung, Whirlpool, GE Appliances, Bosch, LG, Dacor, and others. The platform supports smart ovens, induction cooktops, range hoods, and Bluetooth meat thermometers. New appliance integrations are added as manufacturers open their APIs. The integration layer is abstracted above individual manufacturer APIs, meaning a firmware update from one brand doesn't break the broader orchestration, only that brand's connector requires updating.
Yes. Without connected appliances, Innit operates as an intelligent recipe guide with step-by-step voice instructions, precise timing, and technique guidance. Users without smart appliances still see significantly better results, a 78% success rate versus 57% without any guidance, because the instruction quality and contextual coaching alone eliminates most failure modes. The 92% success rate and full appliance automation require connected appliances, but the guidance-only experience is already a substantial improvement over following a printed recipe.
Recipe scaling is deeply integrated into the cooking intelligence layer. When a recipe is scaled from 4 to 8 servings, the system recalculates not just ingredient quantities but also cooking times, temperatures, and vessel sizes, a doubled batch in the same pan may require different heat settings and a longer cooking time than simply doubling the timer. The knowledge graph encodes these non-linear scaling relationships for each recipe category, ensuring that scaling a dish doesn't produce predictably worse results than the original serving size.
Yes. The recipe intelligence layer includes a substitution graph that maps dietary substitutions, vegan butter for regular butter, gluten-free flour for wheat flour, and adjusts cooking parameters accordingly. Gluten-free baked goods require different temperatures and times than their wheat equivalents; the system knows these differences and updates appliance commands accordingly. All substitutions are validated by FoodLM's health conditions model before surfacing, ensuring a swap that's appropriate for the user's declared dietary needs and health profile.
Innit iQ exposes the same food intelligence infrastructure that powers the consumer app through three B2B deployment models. Innit Integrator™ is a set of plug-and-play modules (recipe discovery, glycemia tracking, dietary coaching, shopping) that embed into an existing website or app via API. Innit App Builder uses the iQ CMS and dedicated design team to deliver a fully custom-branded food and wellness app in weeks rather than years of internal development. The iQ CMS gives enterprise clients a management interface for their content, users, analytics, and AI modules with no engineering required. CPG brands use it to create recipe experiences tied to their products; pharmaceutical companies use it to deliver nutrition-based health interventions; retailers use it to increase basket size and reduce cart abandonment.
Ready to build AI that doesn't just guide users, but acts on their behalf?
From smart kitchen orchestration to agentic workflows across any connected device ecosystem, most projects go from kickoff to deployed AI in 8–16 weeks.