YC Winter 2026 (W26) companies

10x Science builds an AI-native platform that automates protein characterization for drug developers. As AI-driven discovery floods pipelines with new therapeutic candidates, scientists face months of manual analysis to evaluate whether each candidate is viable before it can advance to clinical trials. 10x Science replaces that bottleneck, letting teams identify which candidates to advance and which to cut, earlier and with greater confidence.

21st builds a registry of React UI components, agent templates, and an SDK for developers building AI-powered applications. The library ships over 12,000 components, covering marketing blocks, UI primitives, and shadcn themes, each delivered as a prompt that any AI coding tool can consume directly. It replaces the work of hand-coding common interface patterns and wiring up agent infrastructure from scratch.

Aemon builds an autonomous AI research engineer for B2B technical teams. It replaces human-led R&D cycles by generating, testing, and iterating on thousands of solution approaches at machine speed, targeting hard engineering and optimization problems. The system defines success criteria, studies current research literature, runs experiments inside a customer's codebase, and delivers production-ready technical breakthroughs.

Agentic Fabriq builds an identity, permissions, and audit platform for AI agents. It gives every agent a defined identity, scopes its access to only the tools and data an admin has authorized, and logs every action it takes, replacing ad-hoc credential sharing and manual access controls that break down as organizations deploy multiple agents. Companies use it both internally, to govern agents acting on behalf of employees, and externally, to let customers connect their own accounts without the product ever holding a token.

Alt-X builds an AI underwriting layer inside Excel for private-market finance. It ingests raw deal materials, PDFs, CIMs, and data room files, and outputs clean, auditable financial models that analysts can review and edit immediately. The product replaces the manual work of copying figures into spreadsheets and rebuilding underwriting models from scratch for real estate and private-equity deals.

Approxima builds autonomous AI agents that handle software maintenance work: scoping tasks, writing code, validating fixes in sandbox environments, and merging changes to production. It targets engineering teams who currently spend developer time on routine bug fixes, maintenance, and incremental updates rather than new product work. The agents run end-to-end on each pull request, replacing the manual triage and review cycle that slows teams down.

ARC Prize Foundation builds open-source AI benchmarks that measure progress toward artificial general intelligence, serving researchers, AI labs, and regulators who need rigorous tools to evaluate whether AI systems can match human learning efficiency. Its ARC-AGI benchmark series exposes the gap between tasks that are trivial for humans and hard for current AI, giving the field a shared standard for tracking genuine intelligence rather than narrow capability. The foundation also runs global competitions and awards research grants to keep AGI progress open and accessible.

Ashr builds a testing and evaluation platform for AI agents, serving B2B companies that ship agent-powered products. It generates large sets of realistic user journeys across an agent's tool calls, outputs, and conversation flows, catching errors, inconsistencies, and regressions before they reach end users. The platform replaces hours of manual QA and supports voice, text, image, file, and multimodal inputs via a Python and TypeScript SDK.

Asimov collects real-world human movement data, at scale, from kitchens, offices, hotels, and homes across multiple continents to train humanoid robots. It serves frontier robotics teams that need diverse, annotated datasets covering natural hand-object interactions, 3D body pose, depth maps, and activity segmentation. The product replaces teleoperation data gathered in controlled lab settings, which produces limited variety and volume.

Aurorin CAD builds mechanical CAD software for engineers, replacing legacy tools like SolidWorks, Fusion 360, and Onshape. It runs on a custom parametric and B-Rep kernel built for modern CPUs and GPUs, so files that take hours to open in existing tools load instantly. AI is embedded in the core, letting engineers design complex parts through a chat interface instead of navigating menus step by step.

AutoSitu builds an AI operating system for real estate development projects, serving city planning departments and design firms. It runs coordinated AI agents through cross-department review workflows, checking zoning, code compliance, QA/QC, quantity takeoffs, and scope gaps in one connected system. The product replaces the manual, siloed coordination that planners, architects, and developers currently do across agencies and disciplines from site feasibility through construction.

Autumn AI builds a real-time signal intelligence platform for go-to-market teams. It continuously monitors public sources, including social posts, code commits, blogs, and announcements, and surfaces buying signals the moment they appear, replacing manual prospect research and scattered database lookups. Sales and GTM teams define their ideal customer profile and the signals they care about, then receive a filtered, real-time feed of relevant intent data.

Avoice builds an AI workspace for architecture firms, handling the administrative and technical overhead that pulls architects away from design work. The platform covers specifications, quality control, submittals, RFIs, timesheets, and fee management, replacing a fragmented stack of legacy tools with a single environment that understands drawings, codes, and materials. Small and mid-sized practices use it to run projects end to end without heavy manual customization.

AxionOrbital Space converts raw Synthetic Aperture Radar (SAR) satellite data into photorealistic, analysis-ready optical imagery in real time, using a proprietary one-step diffusion architecture. Defense teams, commodities traders, and disaster response operators rely on it for continuous Earth observation that works through clouds and at night, conditions that make conventional optical satellites unusable. The product replaces manual SAR interpretation workflows and restores compatibility with standard computer vision pipelines.

Axis builds an AI-native analytical system for physical commodities trading desks. Traders use it to deploy models that continuously monitor markets, derivatives positioning, refinery data, storage inventories, trade flows, and geopolitical signals across metals, agriculture, and energy. It replaces the manual process of aggregating and analyzing the sprawling data sets that underpin both exchange-traded and bilateral OTC decisions.

Balance runs an accounting firm for SMBs in the UK and Denmark, pairing its own AI agents with human accountants to keep books reconciled continuously through the month. The agents handle transaction matching, categorisation, and document reconciliation as activity happens, so clients avoid the traditional month-end rebuild. The service replaces an in-house finance hire across bookkeeping, VAT, statutory accounts, payroll, and management reporting.

BaseFrame builds a desktop app that scans each employee's existing work history, across email, documents, Slack, calendar, browser, and files, to pinpoint the highest-value tasks to automate. It replaces the manual audit of workflows and the guesswork of generic AI demos by using local models to analyze data on-device, keeping sensitive information off external servers. Teams export the identified automations to their tool of choice and share them across the organization.

Beacon Health builds AI agents that operate directly inside EHR systems, handling the administrative back-office work that primary care physicians lack time to complete. The agents execute end-to-end workflows, including preventative screenings, prior authorizations, referrals, and risk adjustment coding, navigating EHR interfaces the way a human staff member would. This replaces manual, time-intensive tasks that otherwise fall through the cracks across a physician's patient panel.

BeeSafe AI builds an anti-fraud platform that deploys AI agents to intercept scammers directly, before real victims are reached. The agents pose as targets across SMS, voice, email, and social channels, extract the mule accounts and payment infrastructure fraudsters rely on, and produce cross-channel intelligence on active scam campaigns. Banks and financial institutions use that intelligence to catch authorized-push-payment fraud and money laundering that conventional detection systems miss because the transactions are victim-authorized.

Beyond Reach Labs builds deployable solar arrays for satellites and orbital infrastructure, replacing rigid fixed-panel designs that cap out on power as launch volume grows. The flagship product, the Flarewing, stows compactly for launch and expands in orbit to deployed areas from 30 to 625 square meters, delivering up to 200 kW per wing. It targets spacecraft operators facing power ceilings that today block orbital datacenters, space stations, and lunar outposts from operating at the scale they require.

Bidflow builds AI estimating software for electrical contractors bidding on commercial construction projects. Its first product, LuxTakeoff, reads uploaded lighting drawings, extracts fixture schedules, counts fixtures, and measures linear quantities, replacing the sheet-by-sheet manual counting that makes takeoffs the slowest part of preparing a bid. Every AI-generated result is reviewable and editable before the contractor exports quantities to a quote.

Booko builds dynamic pricing software for businesses that sell bookable time slots, such as fitness studios, medspas, golf simulators, and escape rooms. It connects to existing booking platforms, analyzes historical booking data to forecast which slots will go unfilled, and automatically adjusts prices and incentives before those slots become unsellable. This replaces manual, static pricing and the revenue lost when time slots expire empty.

Bubble Lab builds an operations platform for community operators and event organizers, replacing a patchwork of software tools with a single system that handles member management, event logistics, and administrative workflows. Teams describe it as consolidating tasks that previously required ten or more separate tools. The product automates work from speaker and sponsor sourcing to membership admin, freeing organizers to focus on member relationships rather than back-office coordination.

Burt builds AI teammates for freight brokers and forwarders, handling the repetitive coordination work, quoting, load building, carrier sourcing, appointment scheduling, and track-and-trace, that otherwise consumes most of an operations team's day. The product works on top of existing systems and processes, learns from corrections over time, and escalates to a human when it encounters something outside its knowledge. It replaces manual, high-volume back-office tasks so operators can focus on customer and carrier relationships.

Button Computer builds a small clip-on hardware device that gives users instant voice access to AI, without relying on a smartphone or laptop. Users press the button to speak, and the device responds in under a second through a built-in speaker or Bluetooth headphones. It replaces the friction of pulling out a phone or typing to query AI, targeting individuals who want hands-free AI throughout their day.

Byteport builds software for accelerating large file transfers, from 1 GB to 100 TB, across constrained and high-stakes networks. Its proprietary DART protocol replaces TCP for customers in AI infrastructure, robotics, satellite operations, and enterprise IT, where slow or failed transfers carry real operational cost. It deploys across on-premises, air-gapped, hybrid, and cloud environments, replacing the default assumption that TCP is the only option.

Cajal builds formal verification infrastructure for AI-generated software, using the Lean proof framework to prove code correct down to the binary level. Its tools replace human code review and testing in environments where machine-written code is produced faster than humans can audit and where AI-driven exploits are a growing threat. The company targets quantum computing and finance as its first applied domains, with broader infrastructure, including an open-source binary lifter called Talos, in active development.

Canary builds an AI QA engineer for software teams shipping code with AI coding tools. It reads source code directly to understand developer intent, then runs the application and attempts to break every change before it reaches production, replacing manual test writing and flaky DOM-scraping approaches. Engineering teams use it to catch broken user flows that AI-generated code introduces faster than traditional QA can handle.

Captain builds a file search platform for AI agents, replacing hand-wired RAG pipelines. It indexes unstructured content from cloud storage like S3, Google Drive, and SharePoint, handling scans, handwriting, tables, and recordings that standard pipelines miss. Developers query across petabyte-scale, multimodal data through a single API that self-tunes retrieval accuracy against their own files.

Cardboard builds an AI-native video editor that runs in the browser and on Mac, targeting growth and marketing teams as well as independent creators who need to produce videos at high volume. Teams feed it raw footage and a brief, and the editor produces an editable first cut in minutes, replacing the cycle of external agencies, slow turnarounds, and back-and-forth reviews. From that cut, users refine edits through chat or direct timeline manipulation, generate B-roll, music, voice-overs, and motion graphics, and share or export without leaving the tool.

Cardinal builds an AI platform that runs precision outbound sales for B2B companies. Its agents handle prospect research, outreach, follow-up, scheduling, and CRM syncing, replacing the stack of tools and the GTM engineering work those motions typically require. The target customer is a founder or sales leader who wants a coordinated outbound operation without assembling it manually.

Caretta builds a real-time AI copilot for B2B sales reps on live calls, surfacing answers to hard questions, objection handlers, and account context the moment they are needed. It replaces manual pre-call research, in-call scrambling, and post-call note-taking by automatically generating transcripts, CRM updates, and next steps. Each call feeds back into a shared knowledge base, so the team's collective experience becomes available to every rep on every future call.

Carrot Labs builds specialized large language models tuned to a business's specific workflows, then continuously refines them against that business's own success metrics. The result is a model that encodes proprietary operational knowledge, growing more accurate and harder to replicate over time. This replaces the manual work of prompt engineering, generic model selection, and ad-hoc fine-tuning that B2B teams do to coax general-purpose AI into fitting their processes.

Carson builds a desktop AI app that replaces generic chat interfaces with purpose-built, shareable workspaces generated on demand. Sales, recruiting, and operations teams use it to run workflows like prospect research, personalized outreach, and deck creation, work that would otherwise require manually coordinating across multiple tools. Each workspace pulls in relevant data, applies company branding, and surfaces drafts and summaries in a single pane.

Cascade builds tooling that lets B2B companies fine-tune and align AI models against their own proprietary data and internal workflows. The result is task-specific models that outperform generic foundation models on the company's particular use cases while cutting compute cost and latency. It replaces the manual, ad-hoc work of prompt engineering and model customization that data and ML teams currently do by hand.

CatchBack Cards builds a mobile and web platform for Pokémon and sports card collectors to create and open custom digital mystery packs with personalized odds, verified by cryptographic tooling. Collectors pull cards digitally and receive the physical cards by mail, buyback offers paid out via Venmo or PayPal, or access to a $1 card marketplace. The platform replaces opaque, static retail pack formats with a configurable, collector-driven experience built around the chase for specific cards.

CellType builds AI agents that run the full drug discovery pipeline, using biological foundation models to simulate human biology. The technology, developed with Google DeepMind, replaces wet-lab-first discovery workflows by computationally identifying and validating treatment candidates. CellType targets large pharmaceutical companies, with Top 10 pharma already among its partners.

Chamber builds AIOps agents for machine learning teams running GPU fleets on cloud infrastructure. The agents monitor training jobs around the clock, diagnose failures, reclaim idle GPUs, and rerun workloads from checkpoints, handling the operational work that pulls researchers away from model development. It replaces the manual cluster-watching and failure-triaging that ML engineers and AI scientists otherwise do themselves.

Chasi builds AI agents for equipment dealers, handling sales inquiries, rental requests, and service communications around the clock. The agents plug into a dealer's existing systems and replace the manual work of chasing emails, logging data, and following up on quotes, work that otherwise consumes rep time and lets deals go cold. Dealers get faster response times and higher fleet utilization without adding headcount.

ClaimGlide automates the end-to-end prior authorization process for private medical practices, covering submission, payer follow-up, denial appeals, and approval documentation. It targets specialty practices such as orthopedics, physiatry, pain clinics, and behavioral health. The product replaces manual prior-auth work done by dedicated administrative staff, pulling data directly from the practice's EMR and handling payer phone calls on the practice's behalf.

CodeWisp lets anyone create and publish playable web games by typing plain-English descriptions, with no coding or prior game-development experience required. The platform generates code, structure, and assets from user prompts and accepts follow-up prompts to refine the result. It replaces the traditional path of learning game engines, writing code, and managing assets manually.

Cofia builds an AI platform that learns how a team runs its internal operations and then implements workflow automations on its own. It targets B2B companies whose operations staff spend time manually executing repeatable internal processes. Cofia replaces the need for those teams to identify, design, and configure automations themselves.

Compresr builds a context compression API for teams running LLM-powered agents and retrieval-augmented generation pipelines. It strips tokens from input context that the model would not meaningfully use, cutting token volume by up to 90% while matching or beating full-context accuracy on benchmarks. The product replaces manual prompt trimming and brute-force full-context calls that inflate cost and latency.

Condor Energy builds energy procurement software for large commercial, industrial, and data center electricity buyers. It aggregates consumption data, power purchase agreements, solar, storage, and market prices into real-time dashboards, then applies AI-driven optimization to hedging and procurement decisions. The product replaces spreadsheet-based energy management and the manual work of tracking exposure, costs, and contracts across multiple sites.

Confluence Labs builds AI that designs experiments and extracts maximum signal from minimal data, targeting research domains where each data point is expensive to generate. The focus is drug discovery and materials science, where the number of experiments required to reach a breakthrough is the binding constraint. The product replaces brute-force experimental search with adaptive, sample-efficient AI that tells researchers which experiments to run next.

Congruent builds AI-native radar hardware and a paired generative simulator for end-to-end autonomous vehicle training pipelines. Current automotive radars output processed point clouds and lock out the raw sensor data that modern single-network autonomy stacks require. Congruent's radar exposes raw data directly and ships with a world-model-integrated digital twin that generates synthetic raw radar scenes, replacing the manual, simulation-free sensor integration work that blocks mass-market autonomy development.

Constellation Space builds an ML-native operating system for large satellite fleets, replacing manual ground-based orchestration and reactive monitoring tools. Satellite operators ingest fleet telemetry through a single API, then get automated predictions on signal quality, traffic, weather, and collision risk hours before problems surface. The system handles data routing, stream isolation, and network handoffs autonomously, under operator-defined policy and with a full audit trail.

Copperlane builds an AI-native mortgage origination platform for lenders. Its agents collect and verify applicant documents, auto-populate form fields from files like W-2s and bank statements, and complete initial underwriting tasks, replacing the manual back-and-forth between loan officers and borrowers. The platform also surfaces a borrower-facing assistant that answers compliance questions and adapts application forms to each applicant's situation.

Corelayer builds an AI on-call engineer for engineering teams in finance, healthcare, and insurance. It continuously monitors logs, metrics, and data across production environments, then runs agents to diagnose incidents and propose fixes, replacing the manual debugging work engineers do during outages. Because production data in regulated industries is sensitive, Corelayer supports on-premises deployments and hardware-backed secure inference so agents can use live data as debugging context.

Corvera builds a sales intelligence platform for consumer packaged goods brands, consolidating distributor depletions, retailer portal data, and ERP feeds into a single governed data layer. It ranks every account by dollar opportunity and surfaces a prioritized call list each morning, replacing the manual work of combing through fragmented spreadsheets and portals. Sales reps open one list and work the accounts that matter most.

Crosslayer Labs builds an outside-in security platform that detects impersonation attacks targeting websites and APIs. It continuously discovers and monitors an organization's full internet footprint, spanning DNS, BGP, TLS certificates, and JavaScript, then correlates signals across those layers to identify attack patterns and deliver root-cause remediation guidance. The product replaces manual, siloed network monitoring for companies that depend on the web for secure operations.

Crow builds a chat-based control layer that sits on top of existing B2B software products. Developers connect Crow's AI agent to their product's APIs and data sources, letting end users type natural-language commands instead of navigating menus and clicking through interfaces. Crow executes real actions inside the host product, replacing the need to redesign or rebuild the UI to support conversational interaction.

Cumulus Labs builds a full-stack AI inference platform for teams running fine-tuned and open-source models. Its three products cover GPU orchestration (Paladin), multimodal inference for text, vision, and audio (Ion), and agent fine-tuning and evaluation (Talos). The platform replaces the choice between self-managed GPU infrastructure and expensive hosted providers by handling orchestration, kernel optimization, and deployment in one place.

DAIVIN! builds tankless breathing equipment for extreme environments, starting with underwater diving. Its SPRING system generates breathable air on demand from water, replacing the heavy, pressurized oxygen tanks that divers, and eventually astronauts and land-based operators, currently depend on. The product targets any professional or explorer who must work where conventional tank supply is dangerous or impractical.

Didit builds a single API that handles identity verification, fraud detection, and compliance checks, covering KYC, KYB, AML, biometrics, and real-time transaction monitoring. It serves businesses that would otherwise stitch together multiple point solutions or build these workflows in-house. One integration replaces separate vendors for document verification, liveness checks, watchlist screening, and ongoing risk monitoring across users, businesses, and transactions.

Ditto Biosciences discovers drugs for autoimmune diseases by mining proteins from parasites, viruses, ticks, and worms, organisms that have evolved over millions of years to modulate the human immune system. The team uses AI and computational protein structure prediction to identify those immunomodulatory proteins and engineer them into therapeutic candidates. This approach replaces conventional small-molecule and antibody drug discovery with biology that nature has already stress-tested across evolutionary time.

Docura Health builds a workers' compensation platform that organizes claim records, generates chronologies, and structures med-legal reports for physicians, claims adjusters, carriers, and employers. It replaces manual medical record review and report drafting across the full claims lifecycle, from intake through settlement. All parties work from the same claim data, with workflows tailored to each role.

Doomersion turns the time Gen-Z already spends doomscrolling into language learning, replacing structured study apps and classroom drills. The mobile app serves short-form videos in the learner's target language, continuously calibrated to their current level so each scroll is slightly more challenging than the last. It targets everyday consumers who want to pick up a new language without a separate study habit.

EigenPal builds an agentic document-processing platform for enterprise teams handling high volumes of structured and unstructured paperwork, including KYC packets, invoices, insurance claims, shipping forms, and loan documents. It extracts data, routes decisions, and logs every action with human checkpoints, replacing manual document review and stitched-together OCR pipelines. The platform runs self-hosted or in the cloud and handles messy scans, handwritten forms, and third-party documents.

Emdash builds an open-source, provider-agnostic desktop application for engineering teams that want to run multiple coding agents in parallel. Teams can execute concurrent tasks across isolated Git worktrees, pull issues directly from Linear, Jira, GitHub, and other trackers, and manage prompts, skills, and tools from a single interface. It replaces the manual, sequential workflow of switching between agents, terminals, and context windows.

End Close builds automated reconciliation software for high-volume payments companies: fintechs, marketplaces, banks, and payroll providers. It replaces manual exception handling and end-of-month close work by continuously matching transactions across exact, fuzzy, and many-to-many patterns, then deploying AI agents to investigate and resolve discrepancies without human intervention. Finance and payment operations teams get an always-current audit trail and real-time position data they can expose directly inside their own products.

Envariant builds an interpretability SDK for foundation model teams, giving them tools to detect, trace, and steer model behaviors directly in the latent space. It targets problems like hallucinations, output drift, and brittle responses in high-stakes domains where traditional verification is too slow or expensive. The SDK replaces iterative data patching with programmatic control over model reasoning, exposing primitives for causal tracing, inductive steering, principle extraction, and edge-case synthesis.

Eos AI builds an AI operating system for healthcare that runs revenue-cycle and administrative workflows, including insurance eligibility, billing, referrals, scheduling, and patient follow-up, inside the tools clinics already use. It harmonizes patient data across disparate systems into a centralized index, then reasons across that data to surface eligible patients, flag early care interventions, and trigger required approvals, replacing manual back-office work done today by clinical and administrative staff.

Fed10 builds AI agents that perform legislative monitoring and policy analysis for businesses. The agents track every bill moving through U.S. legislatures, identify regulatory threats, and deliver actionable guidance, replacing the work that human policy consultants and lobbyists handle at hourly rates over days. It is built by three former lobbyists with direct bill-passage experience.

Fenrock AI builds specialized AI agents for the back office of small- to mid-sized US banks, starting with compliance. The agents handle regulatory compliance work that bank staff currently do manually, helping institutions pass regulatory exams and improve efficiency ratios.

Fern builds high-fidelity reinforcement learning and evaluation environments for robotics teams. It gives robotics developers the simulation infrastructure needed to train and test physical AI systems at scale, replacing the need to build custom evaluation tooling in-house.

Fixture builds a CRM for startups that consolidates customer interactions, marketing signals, and communication tool data into a single go-to-market context layer. AI agents surface the next actions on open deals and flag contacts that need attention, replacing the manual work of logging activity and deciding what to follow up on. The system is fully programmable via API and MCP, so teams can operate it from a web UI or from external AI tools like Claude.

Foreman builds a construction project management platform for contractors and trades, replacing the spreadsheets, email chains, and disconnected apps that run most job sites today. Contractors upload plans or a project description and Foreman's AI generates takeoffs, estimates, and proposals; every document, cost, change order, photo, and field log then lives in a single record tied to that job. The platform covers the full project lifecycle from first bid to final invoice.

Fort builds a wearable device for people focused on strength training and longevity. It automatically tracks reps, sets, and exercises using heart rate and motion data, replacing the manual logging that competing wearables require. It also monitors sleep and recovery, and pairs with an expert-designed training program to deliver actionable workout feedback.

Forum builds a regulated exchange where traders can go long or short on cultural attention. It constructs indices from search, social, and streaming data to quantify what audiences care about at any moment, turning that signal into a tradable asset class. The product replaces informal bets on cultural relevance with a structured market for managing demand and cultural risk.

FullSeam builds AI agents for corporate accounting teams, handling accounts payable, accounts receivable, cash application, and reconciliation. The agents take over manual, document-heavy work that accountants typically do outside their core accounting software, such as scanning invoices, coding GL accounts, and posting to ERPs. The target customer is the finance and accounting function at B2B companies.

Galactic Resource Utilization Space, Inc. (GRU Space) builds permanent surface infrastructure on the Moon, starting with the first lunar hotel targeted to open in 2032. Its hardware converts lunar regolith into structural bricks and deploys modular pressurized inflatable habitats engineered to survive the Moon's thermal and pressure extremes, replacing the need to ship finished construction materials from Earth. A phased mission plan runs from a systems test landing through cave-base deployment to hotel operations, with long-term buildout of roads, warehouses, and mass drivers ahead of equivalent development on Mars.

General Astronautics builds robots designed to operate in microgravity environments, serving space research and manufacturing applications. Its hardware targets the physical tasks that humans cannot easily or safely perform in orbit, replacing manual or ground-dependent methods for on-site experimentation and production in space.

General Legal runs an AI-native law firm serving growth-stage founders on commercial legal work: contract drafting, review, and negotiation across technology transactions, employment, data privacy, fintech, crypto, and government contracts. It replaces the traditional outside-counsel model, where founders wait days and pay variable hourly rates, with staff attorneys backed by AI tooling that turns contracts in hours. The target customer is a B2B founder who needs deals closed fast without retaining a full-service firm.

Glue builds an open-source design editor purpose-built for AI coding agents. On an infinite canvas, agents create new UI components, iterate on existing interfaces, and output production-ready code in a single workflow. It replaces the split between static mockup tools and separate implementation environments, collapsing design and engineering into one loop for builders working alongside agents.

Grade builds a payroll and payments API for companies that pay people based on work delivered rather than hours logged. Brands, agencies, and marketplaces use it to pay creators, influencers, and contractors once tasks or targets are met. Grade handles tax form collection, identity verification, compliance reporting, and global payouts, replacing manual payment workflows and in-house compliance processes.

GrazeMate builds autonomous drone systems that herd and monitor cattle across large ranches, replacing mustering work done by helicopters, motorbikes, and horses. Ranchers pick a destination and the drone plans the route, moves the herd, and repeats on a schedule, while simultaneously locating and counting animals, estimating weights, measuring pasture biomass, and flagging water, fence, and health issues. The result is a single operator managing thousands of head across millions of acres from a phone.

Haladir builds an AI platform that handles planning, optimization, and workflow automation for logistics operators, including 3PLs, distributors, and brands. Its Nomos product runs solver-grade optimization across inventory allocation, wave planning, vehicle routing, dock scheduling, and labor, replacing work that operations managers and OR analysts would otherwise spend days or weeks on. Its Talon agents automate external and internal workflows: consolidating carrier and inventory-owner communications, handling customer calls and quotes, and executing plans directly inside warehouse, transport, and order management systems.

Hex Security deploys AI agents that probe web apps, APIs, and cloud infrastructure around the clock for exploitable vulnerabilities, replacing the traditional annual penetration test, a weeks-long, six-figure engagement that delivers a static PDF. Security and engineering teams get continuously verified, proof-of-concept findings instead of a once-a-year snapshot that is outdated before it is acted on.

Hlabs manufactures US-made electronic and mechanical components for robotics teams, covering actuators and other core building blocks that builders currently source overseas or fabricate themselves. It sells plug-and-play robotic primitives for early-stage development and works with production-scale teams on custom designs. Customers build across household robots, humanoids, arms, quadrupeds, and defense platforms.

Human Archive collects, labels, and delivers large-scale multimodal datasets, covering RGB, RGB-D, audio, tactile, motion capture, and robot data, to frontier labs and general-purpose robotics companies building foundation models. The company designs and deploys custom hardware across residential and manufacturing environments, then runs the captured data through internal quality assurance, anonymization, and annotation pipelines. It replaces the fragmented, low-volume data collection efforts that robotics teams would otherwise run themselves.

IncidentFox builds an AI site reliability engineer that investigates and resolves production incidents autonomously. It ingests a team's codebase, past incidents, logs, metrics, and deployment history to understand their specific stack, then queries live systems like Datadog, Coralogix, and CloudWatch to identify root cause and generate ready-to-run fix scripts. Engineering teams get a diagnosis and fix waiting for review instead of waking up to a raw alert at 2am.

InventoryQuant automates contents inventory for insurance claims, replacing the manual work done by public adjusters, restoration contents firms, and insurance carriers. Adjusters record a video or audio walkthrough of a loss site; the platform identifies every item on camera, transcribes spoken descriptions, sources replacement prices from retailers, and produces a submission-ready report. Work that previously took days of hand-typing, tab-by-tab price hunting, and report formatting is handled in minutes.

Inviscid AI builds an agentic physics simulation engine for industrial facilities, data centers, and energy systems. It combines live IoT sensor data with computational fluid dynamics solvers to run continuous thermal and airflow simulations, replacing work that previously required a specialist to manually set up each case and ran only periodically. Facilities engineers and building operators get real-time answers on heat distribution, ventilation, and equipment behavior without the lag of quarterly simulation cycles.

Jinba lets enterprise business teams build and deploy AI workflows by describing them in plain language, replacing the drag-and-drop builders and IT backlogs that traditionally gate automation projects. It targets regulated industries, including banking, insurance, and manufacturing, where compliance controls, audit logging, role-based access, and on-premise deployment are requirements. Non-technical staff can create and run workflows directly, cutting engineers out of routine automation work.

Kita builds an AI platform that automates credit underwriting for lenders in emerging markets and community-focused institutions in the US. It ingests unstructured borrower documents, such as e-wallet records, bank statements, and utility bills, extracts fraud-checked data, and produces decision-ready credit files, replacing the manual review work that credit and risk teams currently do file by file. Its models learn from document signals linked to repayment outcomes, continuously refining fraud detection and risk scoring over time.

Klaus AI offered hosted, cloud-based instances of OpenClaw, an open-source AI assistant, connected to messaging platforms such as Slack in minutes. The service targeted individuals and small teams who wanted a personal AI assistant without managing their own server. Klaus shut down on September 1, 2026, and all customer assistants are now offline.

Kyten Technologies manufactures custom aerospace-grade battery packs for drones, submarines, and satellites. The company is building a factory to serve US defense and aerospace primes whose programs are outpacing what the legacy battery supply chain can deliver. The founders bring direct experience developing and producing battery packs for orbital spacecraft, and now offer that same rapid development and volume production capability to the broader industry.

Lance builds AI agents that run day-to-day hotel operations, handling inbound calls, reservation management, housekeeping coordination, and staff scheduling. The product targets hotel owners and operators, replacing manual work that front-desk associates, housekeeping supervisors, and scheduling managers currently do by hand. Lance agents plug into existing hotel software and execute multi-step tasks autonomously, freeing on-property teams to spend time with guests.

Laurence runs Amazon advertising campaigns autonomously for brands, replacing the manual work of human ad managers who adjust bids and keywords by hand. It re-evaluates every keyword 24 times a day, reallocates budget to winning search terms in real time, and optimizes toward profit margins, customer acquisition cost, or lifetime value rather than surface metrics like ROAS. The system uses reinforcement learning and proprietary LLMs trained on the brand's existing ad copy.

LegalOS runs an AI-native immigration law firm that handles U.S. work visa petitions for companies hiring or transferring foreign talent. Specialized AI agents draft petition narratives, gather supporting evidence, and model regulator objections, cutting a process that traditionally consumes weeks of attorney time down to days. Licensed immigration attorneys review and sign every filing, covering visa categories including H-1B, O-1, L-1A, L-1B, TN, and employment-based green cards.

Lexius turns existing retail security cameras into an AI loss-prevention system for grocery and convenience stores. It watches every item cross the scan belt or self-checkout window, compares what it sees against the transaction log in real time, and flags skip-scans, sweethearting, and banana tricks the register never recorded. The product replaces manual lane monitoring and after-the-fact receipt audits without requiring new hardware.

Librar Labs builds an AI-native integrated library system for school librarians. Its core tool photographs physical shelves, automatically identifies and catalogs each book, and generates inventory and weeding reports, replacing manual shelf-scanning and legacy cataloging software. Patron management, mobile check-out, and catalog maintenance run from a single platform on a standard phone, with no dedicated hardware required.

Lucent watches every user session replay and automatically surfaces bugs, UX friction, and product insights for software teams. It replaces the manual work of scrubbing recordings and writing bug reports by delivering ready-to-act alerts with steps to reproduce, console logs, network requests, and affected-user counts. Product and engineering teams get a continuous feed of what is breaking and what is working, without reviewing sessions themselves.

Luel sources and licenses rights-cleared multimodal training data for frontier AI teams. It converts everyday human speech and actions into structured, high-quality datasets, replacing the ad-hoc data collection work those teams would otherwise do themselves. Customers can commission bespoke data collections or purchase off-the-shelf datasets through Luel's marketplace.

Mango Medical builds AI that generates personalized surgical plans for orthopedic procedures, delivering completed plans to surgeons in minutes rather than days. The product replaces a manual, time-intensive planning process that today stretches across days of specialist work. It targets orthopedic surgeons and the companies that supply them, and is pursuing FDA 510(k) clearance.

Mantis builds behavioral forecasting infrastructure by combining large language models with physics-based simulations to construct digital twins of individual people. The models ingest continuous physiological and behavioral data, such as smartwatch readings and digital interactions, to produce predictive forecasts of a person's physiology and behavior over clinically relevant time horizons. The company is starting in medicine, where its twins replace manual, intuition-driven patient assessment with model-driven trajectory forecasts.

Martini builds a collaborative, AI-native production platform for filmmakers, studios, and agencies creating professional AI-generated video. It replaces a fragmented stack of disconnected generative tools with a single environment that combines image generation, video generation, and world models on a built-in timeline, plus director-level camera control over position, lenses, and movement. The result exports directly to Premiere or Resolve, fitting into existing post-production pipelines.

Maven builds payment infrastructure for AI voice agents and chatbots, letting them collect card payments mid-call or mid-chat without interrupting the conversation. Companies building conversational agents use a single API call to hand off to Maven, which captures card data, processes the charge, and returns the caller, keeping card data off the agent company's servers entirely. Maven replaces the broken pattern of dropping a checkout link or forcing a human handoff whenever a customer needs to pay.

Maywood automates deal execution for investment banks, replacing the manual labor analysts and associates spend producing confidential information memoranda, financial models, and diligence responses. The platform generates finished, on-brand outputs while keeping every document editable and auditable inside a single deal workspace. It handles the full workflow from pitch materials through due diligence, work that banks today do by hand across disconnected tools.

Mendral builds an AI DevOps engineer for software engineering teams, handling the operational work that sits between writing code and shipping it: CI failures, flaky tests, slow builds, broken releases, supply chain security, and code review. It monitors, diagnoses, and acts on these issues autonomously, replacing the platform engineering work that teams would otherwise staff or build in-house. Over time it accumulates context from each build and incident to handle new problems faster without human intervention.

Menza builds an AI data analyst for consumer brands. Brands connect their data sources and the system continuously monitors for revenue opportunities, cost leaks, trends, and anomalies. It replaces the work a human analyst would do combing through disparate data to surface actionable signals around the clock.

Milliray builds millimeter-wave radar systems that detect, track, and classify small drones that conventional counter-UAS sensors miss. Its DroneTector product fuses mm-wave, camera, and acoustic sensors to cover targets from cheap consumer quadcopters to autonomous swarms, replacing legacy radar stacks that were designed before low-signature micro-UAS became a practical threat. Airports, critical infrastructure operators, and front-line defense teams are the primary customers.

MochaCare builds an agentic back-office platform for home care agencies, combining AI automation with human support to run hiring, scheduling, and client intake end-to-end. The platform handles caregiver recruiting and onboarding, credential tracking, and communications across calls, texts, and email, replacing the manual administrative work that agency coordinators otherwise do themselves.

Moda builds a monitoring and learning layer for AI agents. It ingests production traces, diagnoses failures by source (prompt, tool, memory, model, or workflow), and automatically generates verified fixes, from patched prompts to retrained tool adapters, that teams can review and ship. It replaces the manual work of combing through agent logs, reproducing failure patterns, and hand-authoring corrections.

Moritz runs a global AI-native law firm that delivers legal work the same day. AI handles the bulk of each matter, and attorneys review and finalize the output, cutting the cost to a fraction of traditional firms. The firm targets businesses that need complex deals closed fast, replacing slow, expensive conventional legal engagements.

MouseCat builds an AI platform that detects, investigates, and acts on financial crime for banks, fintechs, crypto exchanges, and merchants. The system learns a company's own SOPs, historical data, and case notes, then autonomously reviews suspicious transactions, accounts, and counterparties end-to-end. It replaces the manual case review work done by human fraud and compliance agents, cutting investigation time and false positives across fraud, scams, AML, and abuse.

Ndea builds frontier AI systems that combine deep learning's pattern recognition with formal reasoning and program synthesis into a unified architecture, targeting the limits of current AI that can only reflect its training data. The lab's focus is AGI capable of independent invention and discovery, replacing the human researcher as the bottleneck in scientific progress across fields like drug discovery, energy, and robotics.

o11 embeds AI agents directly inside Microsoft Office and Google Workspace apps, letting enterprise teams automate workflows, from templated presentations to financial models, without leaving the tools they already use. It replaces manual, repetitive document and spreadsheet work that teams currently handle by hand or with disconnected add-ons.

OctaPulse builds an AI vision platform that automates quality inspection in fish hatcheries, replacing manual phenotyping and deformity screening done by trained technicians and geneticists. The system uses off-the-shelf cameras to assess broodstock and juvenile fish, cutting per-fish inspection time from roughly five minutes to under thirty seconds. Target customers are vertically integrated finfish farms, with the platform designed to expand beyond hatchery QA into feeding, health monitoring, and processing across the full production lifecycle.

One Robot builds world model-based simulation environments for robotics teams training and evaluating manipulation policies. The environments replicate both the visual appearance and physical behavior of real scenes, covering contact-rich tasks, deformable objects like textiles, and precise assembly. This lets teams run experiments, surface edge cases, and generate targeted synthetic training data without consuming physical robot time or setting up real-world trials.

Opalite Health builds an AI medical interpreter that gives healthcare providers real-time, speech-to-speech interpretation in over 150 languages and dialects, available around the clock with no wait time. It fits inside existing clinical workflows and handles interpretation, patient-safety checks, and clinical documentation at the point of care. The product replaces on-demand human interpreter services, which carry per-minute fees and introduce delays that can defer treatment for patients with limited English proficiency.

OpenSpec builds an open-source, spec-driven framework that helps developers plan complex software features before and during implementation. Developers write lightweight spec documents that keep both teammates and coding agents aligned as requirements evolve and code changes. The framework replaces ad-hoc, scattered planning by acting as durable, structured context that coding agents read throughout the build.

Origami Robotics builds a general-purpose dexterous robotic hand and a paired data-collection glove designed to match each other exactly, eliminating the gap between data capture and robot deployment. The system targets manual tasks in manufacturing and logistics, replacing human operators on manipulation work that conventional industrial robots cannot handle. Real-world data collected across deployed units trains a general manipulation model, which feeds automation solutions back to those same industries.

Origin builds patient-derived cancer single-cell datasets that map how individual cancer cells respond to drugs and genetic medicines under defined DNA switches. Pharmaceutical and AI companies license these datasets to predict which patient populations will respond to a given therapy before a trial is designed, replacing the costly process of learning that only after a trial reads out. The goal is to raise oncology trial success rates by identifying responder profiles at population scale ahead of enrollment.

Orthogonal builds a unified payments and access layer for AI agents that need to call external APIs. Agents connect through a single key, discover the right endpoint from a catalog of 60-plus providers, and pay per call, replacing the manual work of managing provider accounts, authentication, and billing across dozens of services. API providers list once and collect revenue each time an agent hits their endpoint.

Overdrive Health automates end-to-end medical billing for healthcare providers, covering insurance discovery, eligibility verification, clinical documentation QA, coding, patient collections, and denials management. The platform replaces manual revenue cycle work that billing teams and outsourced billing companies traditionally handle by hand, letting providers collect more while reducing administrative overhead.

Overshoot builds low-latency AI infrastructure for developers creating real-time vision applications. It connects live video feeds to a library of Vision Language Models through a minimal API, returning responses in under 200ms and replacing the need to stitch together inference platforms, streaming pipelines, and scaling infrastructure. Target use cases span physical security, robotics, gaming, and consumer products.

Oximy builds device-level AI governance software for enterprises, giving security, finance, and IT transformation teams visibility and control over every AI request, file transfer, and agent action employees make across their fleet. Its proprietary on-device model, Oxi-1, reads prompts, files, images, and spreadsheets in real time, then applies company policy to allow, warn, redact, or block the action before it leaves the machine. The product replaces manual audits and reactive policy enforcement with automated, pre-transmission controls across all AI tools in use.

Oxus automates the execution work inside internal audits for pre-IPO and public companies, covering industries from fintech and biotech to enterprise software. Its platform ingests walkthrough recordings, transcripts, policies, and supporting documents to produce standardized flowcharts, narratives, control inventories, and risk mappings in minutes, replacing weeks of manual work by audit teams. The result is faster audit delivery without the outsourcing costs that understaffed teams typically absorb.

Palus Finance builds a treasury management platform that moves company cash from money market funds into short-duration bond portfolios, delivering higher yields with sub-3-day liquidity and setup through existing bank accounts. The platform also provides fractional CFO services, pairing AI-powered financial workflows with human finance professionals to handle cash flow planning, margin analysis, and day-to-day finance decisions. It targets startups and SMBs that lack dedicated finance staff, replacing the work of both a standing treasury function and an in-house CFO team.

Panta brokers commercial insurance for contractors, trucking fleets, logistics operators, and manufacturers, combining licensed brokers with AI-powered internal workflows. It targets heavy-industry businesses that standard markets routinely decline or underserve, handling placement across general liability, workers comp, commercial auto, cargo, and related lines. The product replaces the slow, generalist brokerage process with a focused, class-specific placement workflow built around each industry program.

Patientdesk.ai builds an AI-powered operating platform for dental practices, covering the front desk, back office, and marketing in a single system. It handles inbound patient calls, appointment scheduling, online booking, recall follow-ups, insurance, payments, Google review requests, paid ad campaigns, and SEO content, replacing the combination of a human receptionist, a traditional practice management system, and a separate marketing agency. Dental practices use it to keep their appointment calendars full and reduce administrative overhead.

Pax Historia builds an AI-powered alternate-history sandbox game where players pick a country and a moment in time, then rewrite how events unfold. Scenarios range from classical antiquity to speculative fiction, with tools to create and edit custom maps, actors, and historical context. The game replaces static, scripted strategy titles with open-ended, player-authored historical simulations.

Payna builds an AI licensing agent for regulated financial services companies, covering mortgage, lending, debt collection, and money transmission. It handles state license applications, renewals, amendments, individual seat licensing, and regulator follow-ups across all 50 US states. The product replaces the spreadsheets, email chains, and outside counsel that compliance teams currently use to track and file licensing obligations.

Perfectly automates end-to-end recruiting for high-growth startups, replacing the work of a full recruiting team across sourcing, outreach, screening, and candidate coordination. Its AI agent, Paul, runs personalized outreach on LinkedIn, email, and text, then delivers interview-ready candidates directly to the hiring team. The product targets understaffed startups that need to fill roles fast without building an internal recruiting function.

Piris Labs builds a full-stack AI inference service that pairs proprietary photonic hardware with a vertically optimized software stack. The system targets the memory bandwidth bottleneck that makes large-model inference expensive on GPU clusters, delivering comparable throughput at lower cost. B2B customers running trillion-parameter models get the same performance without the unit economics that make GPU-based inference unsustainable at scale.

Pocket builds a pocket-sized AI hardware device that captures spoken conversations in real time and converts them into structured notes, action items, and searchable records. It targets professionals, founders, doctors, and engineers who work on the move and currently rely on manual note-taking or separate recording apps. The device combines custom hardware with on-device AI to eliminate the gap between live conversation and organized, retrievable information.

Pollen builds AI employees for financial services firms, trained on each firm's specific workflows to execute document-heavy tasks, from portfolio reviews to tax provisions, in minutes. The agent, called a Pollenator, learns a firm's processes once and then runs them autonomously, replacing the manual analyst work behind recurring deliverables like LP reports, credit packages, and underwriting memos.

Pollinate builds procurement automation for supply chain teams, starting with three-way match: pulling supplier invoices from email and reconciling them against purchase orders and receipts across ERPs. The product replaces the manual spreadsheet and inbox work that finance and procurement teams do today because ERP systems are too rigid to handle messy, inconsistent data. The company has rebranded to Khotan, which delivers bespoke procurement software built around each team's specific approval rules and exception handling.

Polymath builds simulation environments used to train and evaluate autonomous AI agents on long-horizon tasks. The simulated worlds let agents practice real-world applications and services workflows repeatedly before deployment, replacing ad-hoc, live-environment testing. Its customers are frontier model labs seeking to improve agent reliability and reduce the need for human supervision.

Polymorph builds personalization infrastructure for consumer and self-serve B2B apps, letting developers integrate AI-driven personalization into user experiences and notifications without building it from scratch. The work it replaces, tailoring content and outreach to individual users at scale, has historically required years of effort and dedicated teams at companies like Facebook and Google. Polymorph packages that capability as developer-ready infrastructure.

Prana delivers AI-powered primary care, pairing a continuously available AI physician with licensed human doctors. The AI monitors connected medical records and wearable data to catch gradual health changes between visits, handling intake, triage, and lab explanations. Human doctors step in for high-stakes decisions, prescription management, and telehealth visits, replacing the traditional episodic care model.

Protent builds real-time video intelligence software for law enforcement and security teams. It analyzes live footage to detect early escalation patterns and threat signals, giving operators the context to prioritize alerts and act before incidents worsen. The product replaces manual video monitoring and reactive alert triage.

Proximitty builds an AI operating system for commercial lending that handles the full loan lifecycle for FDIC-insured banks and fintechs. Its regulator-grade AI agents ingest and reconcile borrower documents, spread financials, generate credit memos, monitor covenants, and service borrower requests across C&I, CRE, and SBA loan portfolios. The product replaces manual analyst work across underwriting, portfolio monitoring, and loan operations.

Q2Q builds an AI-powered deal sourcing engine for private equity teams. Buyers provide their acquisition criteria, and Q2Q identifies target companies, writes personalized outreach emails, and books first calls with business owners. It replaces the manual sourcing and outbound work typically done by junior deal team members or third-party researchers.

RamAIn builds AI agents that automate repetitive UI-level tasks across web portals, legacy systems, and desktop applications, without requiring API access. Operations teams in mortgage, finance, healthcare, and insurance use it to replace manual work like copying data between systems, checking vendor portals, and chasing handoffs. Users describe a task in plain language, and RamAIn compiles it into a reusable, self-healing workflow exposed as an API.

Reframe builds an AI-native marketplace for hardware teams to procure both off-the-shelf and custom-manufactured components. Procurement agents handle the end-to-end sourcing process, replacing the manual work that typically requires dedicated procurement staff. The result lets hardware teams source parts at scale without growing their operations headcount.

Remix builds an AI content engine for creators and brands who need to produce social media posts, short-form videos, carousels, and articles at scale. Users supply existing assets, such as photos, voice notes, or account data, and a fleet of parallel AI agents handles photo selection, copywriting, and editing simultaneously. The product replaces the manual work of adapting trends and formatting content across platforms.

Remy AI builds a computer vision intelligence layer for warehouse operators, connecting to their existing security camera networks to surface operational metrics that managers currently have no systematic way to track. The product monitors trailer fill rates and dwell times, workforce productivity by zone, and security events such as unauthorized access and segregation breaches. It replaces manual floor walks, ad-hoc reporting, and passive CCTV footage with automated dashboards that let off-floor decision-makers benchmark sites and hold ground-level teams accountable.

Ressl AI builds autonomous back-office agents for trades businesses, including HVAC, plumbing, electrical, roofing, and collision repair shops. The agents handle quoting, estimating, parts procurement, insurance coordination, scheduling, and customer follow-ups by working across the software these businesses already run, replacing the "glue" employees whose jobs consist of copy-pasting between tools and routing work between systems. Each agent takes ownership of a task end-to-end, from reading the incoming request to completing the update and passing context forward, so field crews and shop operators do not have to.

REV1 automates the conversion of 3D CAD models into fully annotated, production-ready 2D manufacturing drawings for mechanical engineers. Its software applies GD&T controls and tolerance values according to ASME Y14.5-2018 standards, capturing functional intent automatically. The product replaces a manual, hours-long drafting process that consumes over a quarter of a mechanical engineer's working time.

Revion builds software for automotive retail service departments. Its first product automates the administrative tasks that consume a large share of automotive technicians' working hours, letting dealerships process more vehicle repairs without adding headcount. The company targets the operational layer of auto retail that has traditionally run on manual paperwork and disconnected tools.

Rhizome AI builds a regulatory research agent for life sciences teams that need to know what health authorities require. Users ask a regulatory question and receive a cited answer drawn from up to 1,000 primary source documents across 84.6 million indexed pages from 17 countries, with every fact tied to its source. The product replaces manual searching of FDA, EMA, and other health authority databases by regulatory affairs professionals.

Ritivel builds AI agents that draft regulatory submission documents for life sciences companies, producing first drafts of filings such as CTDs, CSRs, INDs, and BLAs directly in Microsoft Word. The agents pull clinical data and prior submissions from connected systems like SharePoint and Veeva, embed citations traceable to source documents, and handle follow-up communications through Outlook. The product replaces a manual, months-long process in which regulatory and medical writing teams must gather scattered documents, consolidate clinical data, and compose dense submissions by hand.

Robby builds a sales and service platform for home services businesses, including HVAC, plumbing, and electrical contractors. It equips technicians with structured talking points and personalized homeowner presentations during every visit, replacing ad-hoc upselling with a consistent, data-informed pitch. General managers get a single command center to track and act on revenue opportunities surfaced from each customer interaction.

RoboDock builds robotic systems that automate depot operations, specifically charging and vehicle inspections, for electric and autonomous vehicle fleets. Its robots handle plug insertion, charge verification, and pre-redeployment inspections that depot workers currently perform manually, multiple times a day, across mixed fleets. The system replaces manual labor at the depot level, reducing vehicle downtime and cutting energy and operational costs at scale.

Rubric AI builds reasoning and verification infrastructure for AI developers, translating expert human judgment into the training signals that shape high-stakes AI systems. The work it replaces is the ad hoc, unstructured process of turning domain expertise into reliable feedback for model training.

Ruma Care builds the administrative operations platform for biologic infusion clinics, handling prior authorizations and copay assistance enrollment. Infusion clinics buy specialty biologics costing $80,000 to $150,000 or more per patient per year upfront, then seek reimbursement through insurers, but a 37% annual denial rate and paper-heavy workflows leave clinics absorbing drug costs when authorizations fail. Ruma pulls patient, diagnosis, medication, and insurance data from a clinic's EHR, matches each case to payer-specific submission criteria, auto-fills and submits prior auth forms, and follows up with insurers until approval, replacing a process that previously stretched across 70-plus portals, paper forms, and manual phone calls.

RunAnywhere builds a unified inference stack for deploying AI models on-device across enterprise hardware fleets. One SDK, built on a shared C++ core with six language bindings, runs models locally using the NPU silicon already in the device, while a hosted API option keeps cloud execution opt-in. A control plane lets enterprises manage model distribution, enforce policies, and monitor performance across thousands of devices, replacing the per-device porting and runtime fragmentation that breaks on-device AI at scale.

Salus builds a runtime authorization layer for AI agents, sitting between an agent and the systems it acts on. Before any action executes, Salus checks it against policy, live evidence, and intent, blocks ones that fail, and returns the exact correction the agent needs to retry. It replaces the gap left by evals and observability tools, which can flag mistakes only after they have already happened.

Samora AI builds multilingual voice agents for B2B organizations that run high volumes of inbound and outbound calls: customer support, outreach, follow-ups, screening, and intake. The agents handle real-world speech including dialects and mid-sentence language switching, enforce strict policy controls to prevent unauthorized commitments, and write outcomes directly to CRMs, ticketing tools, and ATS platforms. When a call reaches a sensitive or uncertain point, the agent escalates to a human, replacing manual call center operations without requiring engineering resources.

Scheduling Wizard builds and delivers physician and residency schedules for hospital departments, GME programs, provider groups, and private practices. It replaces the manual, time-intensive work done by chiefs, administrators, and program coordinators by handling complex rotation rules, call blocks, vacation requests, and coverage requirements end-to-end. The company is expanding from schedule creation into patient triaging and provider management.

Seeing Systems builds autonomous strike and training drones for military operators in contested combat environments. Its two hardware platforms, Bandit and Banshee, pair a low-cost attritable FPV drone with a modular, fibre-optic-guided autonomous platform that resists jamming. The Aerie AI core manages multi-platform swarm operations, reducing the cognitive demands on operators that current drone systems place on them.

Sentrial builds a monitoring and debugging platform for AI agents and products running in production. It detects failures such as infinite loops, hallucinations, and poor user experiences in real time, then diagnoses root causes and surfaces specific fixes. Engineering teams use it to replace guesswork from offline evaluations with direct visibility into how their AI behaves with live users.

Sequence Markets builds low-latency execution infrastructure for traders operating across crypto exchanges, prediction markets, and tokenized-asset venues. Its routing software consolidates a single market view into a single trade, replacing the manual work of managing positions across fragmented platforms. The system targets programmatic traders who need unified, fast access to perpetuals and prediction markets through one interface.

Servo7 builds autonomous robots that unload loose-loaded shipping containers end to end, replacing manual dock crews in warehouses and logistics facilities. The robots deploy at existing dock doors without facility redesign, learn new tasks from short demonstrations, and improve performance on the job. They handle boxes of mixed sizes and weights, removing the need for pre-mapped configurations or dedicated integration engineers.

Shofo builds a large-scale video and multimodal data platform for AI labs that need custom training datasets. It has aggregated billions of videos into a searchable index spanning computer use, talking-head footage, speech, gameplay, and expert physical labor, each paired with structured metadata and annotations. When a lab specifies an exact data requirement, Shofo's agents search the index, extract the matching clips, run them through a labeling pipeline, and deliver a finished dataset in days rather than months.

ShortKit builds a full-stack video infrastructure layer for consumer app teams that want short-form video feeds without building the underlying systems themselves. It provides client SDKs, video processing, recommendation engines, and monetization tooling, replacing months of in-house engineering work. Marketplaces, community apps, and publishers use it to run owned video surfaces rather than ceding that audience to third-party platforms.

Signals builds an AI customer engagement agent for direct-to-consumer brands, communicating with shoppers over iMessage and RCS across the full purchase lifecycle. It answers pre-purchase product questions, handles order and delivery issues, and prompts repeat purchases, replacing the work a skilled in-store associate would otherwise do. The product is designed to reduce returns, deflect support tickets, and drive incremental repeat revenue without requiring discounts.

Sila builds a messaging and collaboration platform that puts AI agents and human teammates in the same conversations. Enterprise teams use it to coordinate work without switching between a chat tool and a separate AI interface. It replaces the pattern of toggling between tools like Slack or WhatsApp and bolt-on AI assistants by making agent participation native to every thread.

sitefire builds a marketing suite that helps B2B brands get discovered and used by AI agents such as ChatGPT, Perplexity, and Gemini. It monitors how AI models mention a brand, identifies which content drives AI visibility, and then executes both on-page and off-page actions, including researching, writing, and publishing content directly to a brand's CMS. The product replaces the manual work of tracking AI search presence and producing the content needed to influence it.

Skillsync builds a tool that makes AI coding session context portable across agents and teammates. Engineers working with Claude Code, Cursor, Codex, and other local coding agents can resume any session on a different agent mid-stream without re-explaining prior work. It replaces the manual effort of reconstructing context each time an engineer switches tools or collaborators.

Sonarly builds an autonomous incident-resolution layer for engineering teams. It connects to monitoring tools such as Sentry, Datadog, and Grafana, triages incoming alerts to cut noise and duplicates, then traces root cause across logs, metrics, traces, and code. The output is either a ready-to-merge fix pull request or a refined alerting rule, replacing the manual on-call investigation cycle.

Sparkles gives every employee at a company the ability to make code changes without knowing Git or engineering workflows. A technical admin connects a GitHub repository once, and from then on non-engineers, such as marketers or support staff, can open pull requests through a web app or Slack by describing what they want in plain language. Coding agents execute the work inside isolated sandboxes, replacing the back-and-forth between non-technical teammates and the dev team for small but recurring code edits.

Sponge builds payment and financial infrastructure for AI agents, letting them hold and spend money autonomously via bank accounts, cards, and crypto. On the agent side, it provides wallets and spending controls so agents can purchase services, like browser tools or premium data, without human involvement. On the business side, its Gateway lets companies accept payments from agents using protocols like x402 and MPP, with no code changes required.

SpotPay builds a stablecoin-based neobank for users who need to hold, send, and spend money across borders from a single account. The app handles international transfers, local payments, and everyday card spending through the SpotPay Calypso Card, replacing the need to juggle multiple bank accounts or remittance services. It targets anyone moving money globally who wants one account for receiving funds from abroad, paying locally, and spending anywhere cards are accepted.

Squid builds a model repository for power-grid planning, serving the engineers and teams who plan and operate electricity networks. It ingests models from tools like PowerFactory, PSS/E, IPSA, CYME, and ETAP alongside GIS layers, spreadsheets, and studies, resolving them into a single versioned, governed network model. It replaces the fragmented workflow of disconnected specialist tools, queues, and static spreadsheets that cause plans to drift from the underlying network model.

Stilta builds agentic AI software for patent litigation, serving IP law firms and in-house IP teams. It runs invalidity, infringement, and freedom-to-operate analyses across patents, scientific literature, and web archives, replacing manual document review with fully source-cited, auditable outputs. Every finding links back to the original passage, giving attorneys a defensible work product without the search and retrieval labor.

Strand AI builds multimodal foundation models that predict biological data missing from patient tissue samples. Its first product, Lattice, generates spatially resolved protein expression maps from standard H&E slides, replacing the need to run multiplex immunofluorescence staining on archival tissue. Pharmaceutical companies use the output to identify better-matched patients for clinical trials and to run retrospective studies across entire sample archives.

Supafax builds an AI agent that works inside Gmail and Outlook to handle email and calendar on behalf of professionals. It drafts replies in the user's own voice, schedules meetings, and automatically labels and prioritizes incoming mail. The product replaces the manual work of sorting, responding to, and organizing a busy inbox.

Synthetic Sciences builds an AI platform for computational scientists, letting them delegate end-to-end research workflows, from literature review and hypothesis generation through experimental design and results drafting, to coordinated AI agents. It replaces the manual, sequential work a researcher would otherwise do alone, running tasks continuously in the background while the scientist is away.

Syntropy builds an autonomous coding agent for software teams tackling complex, long-horizon engineering tasks. Developers describe a feature in plain language and the agent produces a fully tested, production-ready pull request, replacing hours of prompt management and context-switching at the terminal. It integrates with Slack for progress updates and supports custom MCP integrations.

Talking Computers builds multimodal interaction models for voice agents, giving AI systems the ability to converse, listen, and respond with human-like fluidity. Its models target the latency bottleneck between foundation models and real-time voice interaction, optimizing for time to first audio token. The work it replaces is the manual engineering required to make capable AI models genuinely interactive across high-stakes fields such as healthcare, education, law, and public safety.

Tensol builds requirements and traceability software for hardware test and systems engineering teams. It links each requirement to its test cases, runs, and verification evidence, then automatically flags stale coverage, recomputes gaps, and queues re-runs when anything changes. It replaces manual traceability work and the spreadsheet or document workflows teams use to demonstrate compliance with standards such as ASPICE, ISO 26262, and DO-178C.

Tepali builds an all-in-one operating system for medspas and wellness clinics, replacing a patchwork of scheduling software, paper charts, and manual front-desk work. Its AI handles inbound calls, texts, and social DMs to book leads within seconds, generates compliant visit charts with injection maps and signed consents, and automates post-visit outreach to bring clients back. The target customer is the medspa or wellness clinic operator losing revenue to slow lead response, charting gaps, and checkout errors.

Terminal Use rebuilds entire operational departments for construction, logistics, manufacturing, and retail companies using AI agents, replacing the manual workflows those departments run today. Starting with a process audit, the team re-engineers a targeted process end to end so that agents handle the work previously done by human operators. Past deployments include large engineering, procurement, and construction firms as well as national-scale medical billing operations.

Terranox AI uses machine learning trained on over 70 years of exploration data to identify high-grade uranium deposits in North America. The company handles the full exploration stack in-house, replacing the manual geological analysis and guesswork that traditional uranium prospectors rely on to select drill targets. Its customers are nuclear fuel supply chains that need domestic, high-grade uranium sources.

The Token Company builds compression middleware that sits in front of large language models and strips filler tokens from context windows before requests are sent. B2B software teams drop it into their LLM pipelines to cut token spend and reduce end-to-end latency without changing model outputs. It replaces manual prompt engineering and bloated context handling with a sub-50ms, deterministic compression layer.

Traverse builds reinforcement learning training environments for frontier AI labs, supplying the data needed to teach models judgment and taste in ambiguous, non-deterministic domains such as law, healthcare, sales, and strategic decision-making. The work it replaces is synthetic prompt-response data generation: instead of contrived expert prompts, Traverse captures real experts operating in real environments, preserving the full reasoning process behind their decisions to create training signals that reflect how judgment actually works.

Travo builds a unified real estate data platform that gives private equity firms, developers, and brokers instant access to rental comps, ownership records, zoning details, and financials for any property or parcel. It replaces the fragmented, manual process of pulling data from multiple sources by consolidating it into a single searchable layer. On top of that data, Travo develops AI applications purpose-built for real estate decision-making.