YC Summer 2026 (S26) companies

6thSense builds tactile sensing hardware for dexterous robots. Its two products, Skin and Eye2 Pro, capture force and first-person video in sync, giving robot hands the pressure feedback that vision alone cannot provide. The hardware replaces the guesswork that causes robots to crush or drop objects during precision manipulation tasks.

83 Sciences turns unpublished experimental data, the majority of which labs currently discard, into new materials, papers, and patents. Its platform ingests raw lab outputs, from voice notes to instrument readings, structures them into a queryable record, and uses AI to identify discoveries and propose optimized process conditions. The company serves both university labs seeking publications and commercialization, and industrial partners looking to bring new materials to market faster and at lower cost.

Aerogen Systems builds point-of-use contamination control hardware for semiconductor fabs. Its Talos system mounts directly inside process equipment and chemically destroys airborne molecular contaminants before they reach the wafer, replacing the ultra-high-purity nitrogen purge that fabs use today. The longer-term product, Cleanpipe, extends that controlled environment across the wafer transport path between tools, with the goal of making billion-dollar cleanroom buildings optional.

Agency Tool Company builds deployment and build infrastructure for robotics teams shipping software to physical robots in the field. Its first product, ATC Deploy, sends only changed bytes over the air and resumes on network drops, updating robot software 20x faster than a standard Docker pull across fleets in agriculture, construction, and logistics. Its second product, ATC Build, runs tests and software validation on actual target embedded hardware, including Nvidia Jetson and Raspberry Pi, replacing ad-hoc CI workarounds that were never designed for constrained, intermittently connected machines.

Agentcard builds payment infrastructure for AI agents, letting them spend money online without human intervention. Companies integrate Agentcard's API to issue single-use or multi-use debit cards, store customer or corporate cards in a vault, and run a purchase agent that buys directly from retailers like Amazon and DoorDash. It replaces the manual step of a human authorizing and executing each transaction on an agent's behalf.

Agnost AI builds monitoring and analytics software for companies running AI agents. It ingests every agent-user conversation to surface silent failures, the points where an agent breaks down without any visible error. It then uses that conversation data to train custom models that replace frontier models, cutting both cost and latency for the agent's tasks.

Akon Labs builds GitNexus, a self-hosted knowledge graph that indexes enterprise codebases into a structured map of every dependency, call chain, cluster, and execution flow. It gives AI coding agents a deterministic, compiler-grade view of code structure in place of file-by-file grepping, so agents resolve cross-repo relationships accurately before making edits. The product replaces brittle text search with static analysis built on Tree-sitter parsing and graph-based clustering.

Aktoria Robotics builds teleoperation software and runs a staffed remote operations center so robotics teams can deploy and drive any robot from anywhere. Its platform delivers sub-8 ms local latency and around 30 ms cross-country, replacing the need for on-site human pilots. It also collects research-grade training data, covering real robot episodes, UMI handheld gripper episodes, and stereo-plus-IMU egocentric footage, so AI teams can skip the expensive, time-consuming work of sourcing and capturing that data themselves.

Alkera AI builds an agentic platform for data engineering and data science that runs inside an IDE or CLI. It serves data teams who need to build, analyze, and maintain pipelines across tools like Snowflake, dbt, Airflow, and BigQuery, replacing manual pipeline work and ad-hoc analysis with agents that carry column-level lineage, data history, and team knowledge. VPC and on-premises deployment options cover teams whose data cannot leave their environment.

Allia Health builds an AI-native electronic health record and practice operating system for behavioral health clinicians. The platform unifies intake, scheduling, documentation, prescribing, and billing into a single workflow, replacing the disconnected point tools that fragment operations across independent therapy, psychiatry, and multi-site group practices. Clinicians using Allia run on one shared system of record that handles AI-assisted documentation and claims submission automatically.

Alloovium builds a voice-driven site operations platform for construction teams. Workers phone an AI to report site updates, ask project questions, and trigger actions; the system drafts forms and follow-ups against the project's contracts, specs, and procedures, flagging missing inspections or approvals before work proceeds. It replaces manual paperwork, document hunting, and compliance checking that site teams currently do by hand.

Almanac builds an AI assistant that operates across a company's tools, conversations, and accounts to complete tasks and maintain a living knowledge base. It connects to Gmail, Calendar, GitHub, Attio, meeting notes, and AI conversations, then compiles everything into a self-updating wiki that tracks projects, decisions, customers, and next steps. Workers message it through Slack or iMessage instead of manually chasing context across apps or keeping documentation up to date themselves.

Amorphic Labs builds AgentMuxer, a capability marketplace where AI agents discover, pay for, and call external APIs, data sources, and software tools at runtime through a single MCP connection and unified balance. It targets developers and teams building AI agents who today must manually provision accounts, manage API keys, and select providers for every third-party service their agents need. AgentMuxer replaces that human-in-the-loop setup by routing each agent request to the best available provider, scored on real-task benchmarks and outcome feedback from the agents themselves.

Amulet builds a high-performance filesystem for parallel AI agent workloads, giving training runs, evals, and autonomous agents fast read access to petabyte-scale cloud data. Workspaces mount as standard Linux filesystems in under 100 ms, fork without copying data, and persist across disposable compute. It replaces ad-hoc data staging and manual checkpoint management for data-intensive AI teams running agentic pipelines.

Applied Electrodynamics builds WaveSight, a through-wall imaging system that maps structural elements, mechanical, electrical, and plumbing systems inside walls and ceilings with millimeter accuracy. Construction teams use it to capture as-built conditions and identify hidden issues that thermal cameras and LiDAR cannot reach. The same hardware platform extends to manufacturing inspection and security screening of bags and packages.

Arbital builds a unified trading terminal for retail traders who currently juggle separate platforms to trade meme coins, perpetual futures, and tokenized equities. A single app covers all three asset classes alongside market-making tools, replacing the need to switch between crypto exchanges, perp venues, and stock brokers. The roadmap points toward a social, mobile-first experience that brings these markets together for a broader audience.

Archal builds stateful API sandboxes for developers testing AI agents against third-party integrations. Each sandbox spins up isolated, production-like environments for services such as Stripe, GitHub, HubSpot, and Slack, replacing shared test accounts and the dirty state they accumulate. Agent developers, connector teams, and integration engineers use it to run tests, CI jobs, and reinforcement-learning rollouts against a known baseline that resets in milliseconds.

Assemble builds AI agents that perform development and configuration work inside enterprise systems such as ERP, CRM, and HRIS platforms. IT teams use it to scope dependencies, implement logic changes, and validate those changes across systems like Salesforce, Oracle EBS, and Oracle Integration Cloud. It replaces the manual, cross-system coordination work that IT developers and administrators handle when delivering changes to core business software.

Async builds AI agents that run day-to-day operations for small businesses in law, healthcare, and real estate. The agents handle lead intake, document processing, customer follow-up, and back-office tasks by operating directly inside the tools and websites a business already uses. This replaces work that was previously too fragmented or low-volume to justify a human hire.

Atlas Discovery builds autonomous AI agents that identify new therapeutic uses for existing drugs and optimize the execution of clinical trials. The company targets the drug repurposing pipeline, replacing manual hypothesis generation and trial design work done by pharma researchers. Its models predict how humans respond to drugs, reducing the guesswork and cost that drives clinical-stage attrition.

Atlia manages short-term rental properties end to end for Airbnb hosts who want professional operations without running them personally. AI agents handle guest communication, pricing, turnover coordination, maintenance, and reporting, with local operators supervising on the ground. The model replaces traditional property management companies, which typically charge fees that Atlia cuts roughly in half.

Atomarine builds offshore floating data centers powered by shipborne generation units, starting with gas turbines and transitioning to marine nuclear reactors. The platforms are manufactured in shipyards, towed to site, and moored at sea, replacing the land-based construction process that requires grid interconnection, freshwater cooling, and local permitting. AI operators get scalable compute capacity, from a single 75-100 MW barge up to multi-hundred-megawatt campuses, without competing for land, grid access, or community approval.

Audun automates debt collection for businesses, replacing manual outreach and one-size-fits-all recovery workflows. It builds a profile of each debtor from invoice history, demographics, credit data, and prior conversations, then runs a tailored AI-driven contact sequence across channels to assess willingness and ability to pay. When a dispute arises or a debtor requests human help, a case specialist steps in.

Avoca Systems builds an AI platform that handles the full patient access workflow for radiology networks, from inbound calls and GP referrals to scheduling, reminders, results delivery, and follow-up recall. It connects to a clinic's existing systems and encodes each site's clinical and operational rules, so complex radiology bookings that generic tools escalate to staff are resolved automatically. The platform replaces front-desk call handling, manual scheduling, and fax-based referral processing across every channel, around the clock.

Axelrod builds an AI operating layer for boutique, independent, and luxury hotels, coordinating front-of-house and back-of-house work across arrivals, rooms, rates, maintenance, supplies, and owner reporting. It sits above a property's existing systems, replacing the manual coordination that hotel staff perform across guests, rooms, teams, and channels. Customers include Michelin Key awarded and Marriott Luxury Collection properties.

Baud designs custom AI chips and a companion developer platform that train and serve neural networks far more efficiently than current hardware. The core innovation is an arithmetic representation that replaces multiplications with additions, packing an order of magnitude more compute and memory into the same die area and cutting power consumption. A PyTorch-compatible compiler lets AI teams run existing code on Baud hardware without rewriting kernels or training recipes, removing the main adoption barrier that has historically stalled custom silicon.

Belvedir builds a custom model factory for teams running AI agents. Its SDK captures agent traces, tool calls, and replies, then automatically trains task-specific models, evaluation harnesses, and routing logic, deploys them privately, and retrains them on live usage. It replaces the manual work of dataset curation, fine-tuning pipelines, and inference optimization that today keeps custom models out of reach for most builders.

Bernard builds AI employees for home and commercial appliance repair companies, replacing the human labor that handles customer calls, scheduling, dispatch, and technician coordination. Its Pre-ID system cross-references the specific appliance, reported symptoms, repair history, similar past jobs, manufacturer documentation, and parts availability to produce a diagnosis and a ranked parts list before a technician leaves the shop. Servicers arrive on-site prepared, cutting the repeat visits that result from missing parts or incomplete diagnostics.

Billow AI Labs builds an AI-native accounting firm that executes the day-to-day work of accounting and FP&A staff: month-end close, reconciliations, accruals, consolidations, audit workpapers, and variance analysis. It operates directly inside tools companies already run, including NetSuite, QuickBooks, Coupa, and complex Excel workbooks, replacing the manual coordination and data entry that finance teams and their outside accountants currently do by hand. Enterprises and accounting firms are the target customers.

Bizmark builds an AI assistant, Otto, for salespeople at manufacturers of physical goods, covering sectors such as injection molding, packaging, and extrusion. Salespeople describe a customer interaction and Otto handles the downstream admin: researching the account, building a quote in the existing CRM or ERP, checking ETAs with operations, and routing approvals, work that currently consumes the majority of a rep's day. The product connects to existing systems, including Salesforce, Dynamics 365, Oracle, HubSpot, Excel, and Outlook, so manufacturers get a working workflow without replacing their stack.

Blaise builds enterprise software end to end, handling scoping, development, and deployment through an AI system. It targets businesses that need custom software built, replacing the work traditionally done by development teams or outsourced engineering shops. Companies bring an idea and Blaise takes it to a production-ready product.

Bloomy builds an AI-powered mastery learning platform for K-12 students, covering English Language Arts, Math, and Writing. It ingests existing benchmark data from tools like i-Ready and NWEA MAP to pinpoint each student's skill gaps, then delivers adaptive lessons, mastery-gated practice, and a Socratic AI tutor, replacing one-size-fits-all curriculum pacing and generic test-prep drills.

Bylaw builds evidence enforcement for AI agents, ensuring agents act only on verified, correct sources before taking action. It targets teams deploying autonomous agents where acting on wrong or missing evidence carries real cost. The product replaces manual review and ad-hoc prompt constraints with a structured guarantee layer that sits between an agent and its decisions.

Callbook AI runs an AI-powered collections agency for lenders managing late-stage delinquent portfolios. Its AI agents locate borrowers, then contact them across calls, SMS, and email in a single continuous conversation, negotiating repayment plans adapted to each borrower's ability and willingness to pay. The product replaces human collections agents and the manual triage work of segmenting accounts, deciding outreach intensity, and routing edge cases.

Care GP builds AI agents that run day-to-day operations for Australian general practice clinics. Its current agents handle medical document allocation, filing inbound emails, faxes, and scans, and manage inbound and outbound phone calls and appointment bookings around the clock. The product replaces the manual administrative work that clinic staff otherwise perform across these workflows.

CarSignal builds an AI-native operating system for independent auto repair shops, replacing the patchwork of management software, scan tools, parts catalogs, and spreadsheets that shops currently juggle. A diagnostic copilot weighs fault codes against vehicle history and customer complaints to recommend repairs with step-by-step confirmation tests. A single connected record carries each job from intake through booking, inspections, estimates, parts ordering, customer approvals, payment, and follow-up, cutting the admin that pulls technicians away from the bay.

Caution gives developers a hosting platform for sensitive workloads that demand cryptographic proof of what is actually running in production. It wraps confidential computing hardware in a git-based workflow that builds reproducible enclave images and lets customers, auditors, and counterparties verify the exact source code, build, configuration, and runtime, down to the compiler layer. It replaces ad hoc attestation setups, bespoke security engineering, and single-cloud lock-in for teams in finance, healthcare, AI, and regulated data collaboration.

Cerenovus builds an AI operating partner that ingests every record a company produces, including contracts, invoices, system exports, and email, to surface and rank operational inefficiencies and value creation opportunities by financial impact. It serves private equity deal teams, operating partners, and advisory and consulting firms doing diligence, portfolio reviews, and post-merger integration. The work it replaces is weeks of manual investigation spread across finance, sales, and operations teams.

Chromie builds an operating system for government contractors that centralizes opportunity discovery, qualification, and pursuit across federal and SLED markets. It replaces manual searches and siloed document workflows by continuously monitoring SAM.gov and state portals, scoring solicitations against a firm's specific profile, and surfacing recompete signals before awards hit the market. Contractors use it to run a higher volume of bids without adding headcount.

Click builds an MCP (Model Context Protocol) layer that gives AI assistants like ChatGPT and Claude access to data sources beyond standard web search, including LinkedIn profiles and posts, YouTube transcripts, social reactions, live flight fares, financial filings, and more. It targets professionals using AI chat for tasks like sales meeting prep, lead generation, company research, and travel planning. The product replaces manual cross-platform research by pulling structured, source-backed context directly into the chat interface.

COACH builds an AI coaching platform for in-person, field sales reps. Reps record their meetings and the system analyzes what top closers do differently, then delivers personalized feedback to every rep after each meeting. Sales managers get visibility into 100% of field conversations, with the AI flagging where reps are struggling so managers can intervene selectively and cover more of their team.

CoArena builds a crowdsourced benchmarking arena for computer-use AI agents, serving AI labs that need honest, real-world evaluations of their models. Users pit two agents head-to-head on live tasks and judge which performs better, generating labeled comparison data that labs cannot produce through internal testing alone. The platform replaces hand-crafted, lab-controlled evals with blind human judgment at scale, publishing every result alongside the methodology that produced it.

Codag compresses the tool output that coding agents consume, stripping test passes, redundant file listings, and verbose logs down to only the signal an agent needs to act on. It sits between a developer's toolchain and the agent, replacing raw, bloated command output with a filtered version that cuts token volume. The target customer is any team running coding agents at scale whose costs climb with every noisy tool call.

Computable builds an open-source price index for GPU compute, aggregating published on-demand rental rates from a fixed panel of providers into a single, transparent reference rate updated every 15 minutes. It targets the gap every mature commodity market fills with a benchmark: GPU hours are rented, resold, and financed at scale but have had no shared reference price. The index replaces closed, unverifiable figures by publishing the data, methodology, and code so any party can reproduce every value.

Conifer builds an AI inference router that selects the model, provider, and cache strategy for each query automatically. It sits inside existing development harnesses, so tools like Claude Code and Codex require no reconfiguration. The product replaces manual model selection and provider management, cutting the inference costs companies pay across a fragmented landscape of hundreds of models and providers.

Context.dev builds a web data API that gives AI agents structured, real-time content from any website. Developers use it to scrape pages, crawl entire domains, monitor for changes, and get sourced answers from the web, replacing ad-hoc scraping pipelines and stale training data. It serves teams building RAG systems, agentic products, and data-enrichment workflows that require fresh, accurate web content.

Control Seat builds FactoryOS, an AI operating system for factories that combines predictive maintenance, SCADA and HMI, virtual PLC, and operations intelligence in one modular platform. It serves manufacturers, utilities, and energy companies, connecting existing plant equipment and sensors into a single source of truth to replace the patchwork of disconnected tools operators use to monitor equipment health, catch failures, and understand plant performance.

Cosmic Robotics builds autonomous heavy-lift robots for large-scale construction on critical infrastructure. Its first machine, Cosmic-1, handles solar panel installation on utility-scale farms and is expanding into data center construction. The company also develops lunar construction robotics under a NASA contract, with a long-term goal of building on Mars.

Cova runs an AI-native home care agency that enrolls Medicaid recipients into Michigan's Home Help Program and arranges for a family member or trusted person, rather than a stranger, to be their paid caregiver. The company handles eligibility checks, Medicaid billing, timesheets, and compliance on behalf of families who would otherwise navigate that paperwork alone. AI agents operate the back office, cutting overhead and letting Cova pass the top caregiver pay rate in the state directly to the people doing the work.

Daqstra builds an AI-native API and orchestration layer for physical R&D test infrastructure, serving engineering teams in aerospace, energy, and robotics. It connects fragmented hardware, such as NI DAQ systems, PLCs, and custom devices, into a single platform that unifies live and historical test data across every run. The product replaces the manual work of rebuilding configurations, calibrations, and data pipelines between tests.

Datoric builds training datasets for robotics and physical AI, serving frontier model teams that need high-quality labeled data for voice, robotics, and world models. It runs its own collection experiments across video, motion, and tactile modalities to identify the data causes of model failures, then converts validated methods into repeatable, rights-cleared data products. This replaces ad hoc data procurement and manual debugging cycles that teams at the research-to-production boundary would otherwise run themselves.

Dawn Industries builds VIM, an automated fault diagnosis and recovery system for industrial automation cells. VIM connects to PLCs, robots, CNC machines, and sensors, identifies the root cause of a fault, and presents the operator with a staged fix requiring a single approval click. When a machined part drifts out of tolerance, VIM detects it immediately after the cut and recommends the corrective adjustment, replacing the manual troubleshooting and downtime that can stretch a recovery to half an hour or more.

Decawork gives IT teams a control plane for every AI agent running across a company. It routes agent traffic through a company-trained model that evaluates each action against plain-English policies, the current task, and prior actions before allowing or blocking it in real time. This replaces ad-hoc agent deployments where IT has no visibility, no consistent credentialing, and no way to audit what agents do or touch.

DeepReach Inc. builds a distributed data network that captures real-world human labor on video for robotics and world-model training. The company designs and manufactures wearable stereo cameras, then places them with local operators, called data partners, who run collection as a business inside warehouses, farms, kitchens, repair shops, and workshops. The result is egocentric physical-world footage spread across thousands of distinct environments, replacing the narrow, single-site datasets that robot policy developers have had to work with.

Denta builds and operates a full-stack dental insurance carrier that sells plans directly to employers, replacing legacy carriers whose claims processing runs on paper checks and takes weeks. Employees receive a Denta debit card with a selectable annual coverage cap, and claims settle electronically in seconds. The product targets the administrative overhead that consumes roughly half of every premium dollar collected by incumbent dental insurers.

Derya runs supply chain operations for robotics, commodities, and food and beverage companies, handling the work that would otherwise require dedicated procurement, sales, and logistics staff. Its AI agents source leads, manage industrial sales conversations, handle procurement, and coordinate shipments from end to end. The product replaces in-house supply chain and sales operations teams across the full cycle, from finding buyers to moving goods.

Dialogus builds production-grade voice agent infrastructure for enterprises, covering the full stack from phone system connectivity and internal tool integrations to human escalation, monitoring, and compliance. Enterprises use it to run inbound and outbound calls for orders, customer support, and collections, replacing manual call-handling and fragmented IVR workflows. Every call completes real actions in connected systems and feeds outcome data back to improve the next interaction.

Dipole Labs builds photonic hardware for AI clusters and quantum infrastructure, replacing conventional silicon interconnects with optical switching to remove the speed, density, and efficiency limits that silicon imposes. The hardware targets the data-movement bottleneck inside large-scale compute environments, delivering the optical layer that next-generation AI and quantum systems run on.

Dock builds a multiplayer workspace where teams of specialized AI agents handle department-level work: coordinating tasks, unblocking one another, and managing execution across workstreams so operators are not manually routing every step. It targets business operators and founders who currently juggle multiple point tools or chat interfaces to direct AI. The agents communicate with each other and with humans in a shared environment, replacing the manual oversight of running each task one prompt at a time.

Donkey buys goods directly from Chinese factories and sells to US importers at a single delivered, duty-paid price to their dock. It replaces traditional trading companies, which mark up containers and conceal factory sources, by using AI to identify the actual manufacturer from a product photo and return a landed cost quote within 72 hours. Donkey takes title, places its own inspectors on factory floors before payment, and handles freight, duties, and logistics under one contract.

Dream builds pocket-sized, battery-powered AI cameras that document the condition of high-value physical assets every time they move. Car dealerships, equipment rental companies, and fleet operators point the cameras at vehicles and machinery; each pass is photographed from multiple angles, the asset is identified, and a 3D condition record is built automatically. On return, the system flags new dents and scrapes that a manual walk-around inspection would miss, replacing ad-hoc visual checks with a continuous, timestamped record of every asset's condition.

Earendil Robotics builds counter-UAS systems for small military units, covering detection, interception, and destruction of hostile drones. Its three integrated products, the Aeglos interceptor, the AURION networked situational-awareness platform, and the Holo thermal sight, replace ad-hoc, fragmented air-defence solutions at the squad level. A companion Drone Warfare Academy trains soldiers to operate the systems in contested environments, drawing on frontline combat experience from active operations including deployments in Ukraine.

Edgerun builds hip-mounted exoskeletons for soldiers operating under combat load. The system weighs under 10 pounds, reads a soldier's gait via onboard sensors, and powers each step to cut physical effort by 30 to 45 percent, extending range and reducing fatigue on missions like assault, patrol, resupply, and casualty evacuation. It replaces raw physical endurance as the limiting factor in how far and fast a unit can move.

EdotEnv builds reinforcement learning environments from financial market data to train agents toward recursive self-improvement. The company uses markets as a self-scaling research ground: as agents discover useful strategies, competition raises the difficulty, forcing continuous improvement. Its environments support hypothesis formation, experiment design, and iterative verification, replacing hand-crafted AI evaluation setups with a dynamic, self-escalating benchmark.

Edviro builds an AI platform that operates energy infrastructure in commercial buildings, replacing the manual work facilities managers do to keep sites efficient. It connects to meters, HVAC systems, batteries, solar, and other equipment, then handles the full loop from detecting and diagnosing problems to generating work orders, dispatching technicians, controlling equipment, and verifying savings against actual utility bills. Facilities teams at data centers, K-12 districts, hospitals, and commercial real estate portfolios are the primary customers.

Egoist Machines builds AI Passport, a portable personal context layer that lets users carry their preferences, history, and identity across AI apps instead of rebuilding that profile from scratch inside each product. Consumer AI apps integrate it via a "Sign in with AI Passport" button and request specific context for a stated purpose and duration, rather than holding broad permanent access. The memory backend is a proprietary vector database with cryptographically enforced access controls, designed for fast, compact per-user storage.

Ekho Labs builds a freight intelligence layer that monitors geopolitical events, strikes, and tariffs to identify which shipments are exposed to disruption before it hits. For freight teams managing high-risk trade corridors, it delivers purchase-order-level rerouting options, carrier alternatives, and booking windows directly into their TMS, replacing manual news monitoring and reactive decision-making with advance, actionable guidance.

Ekpa connects retail investors' existing brokerage accounts into a single consolidated view and delivers AI-generated stock analysis and personalized trade recommendations. It replaces the manual work of tracking positions across multiple brokers and interpreting market data. The product is read-only: it pulls portfolio data without ever touching or moving funds.

Enact builds the post-training layer for physical AI, generating targeted data that makes robotics models reliable in real-world deployment. The company runs a closed improvement loop: deploying policies on tasks, identifying failure states, and producing demonstration and recovery datasets to fix those failures. It replaces the manual, ad hoc process robotics teams use to find model gaps and collect the data needed to close them.

Erinys launches and operates AI-native law firms for plaintiff-side litigators who want to go independent. It provides the full operating stack, including software, back-office functions, staff, working capital, and client acquisition infrastructure, so a lawyer can open a practice without first building an organization. The work it replaces is the years of firm-building overhead that has historically stood between an attorney and equity partnership.

Ethos Space Resources develops lunar mining and construction infrastructure to produce solar-grade silicon, rocket propellant, and construction materials directly from Moon regolith. It targets the bottleneck that will follow Earth's grid constraints: once data centers move to orbit, rocket payload capacity becomes the ceiling, and Ethos removes that ceiling by supplying mass already outside Earth's gravity well. The process melts and electrolyzes regolith in hard vacuum, replacing the need to launch raw materials from Earth.

Evergrove builds AI voice agents purpose-built for workers' compensation care coordination, serving managed care organizations, TPAs, specialist PT networks, and ancillary providers. The agents run end-to-end phone calls across intake, clinical, status, billing, and investigative workflows, replacing the contact center staff who currently rely on phone, fax, and email to move cases forward. Each delay in coordinating an injured worker's care slows rehabilitation, reduces provider throughput, and raises costs for insurers and employers.

Executor builds an MCP gateway that connects AI agents to any integration, whether OpenAPI, GraphQL, or MCP, through a single endpoint. Developers configure an integration once and every agent client gets access to it, replacing the repetitive per-client OAuth setup and API key management that multiplies across tools like Claude Code, Cursor, and Codex. At runtime, Executor surfaces only one tool to the model and fetches what it needs on demand, keeping context windows small regardless of how many services are connected.

Exosat builds low-Earth-orbit communication satellites designed with no ITAR/EAR-controlled components, allowing manufacture outside the United States and launch from non-U.S. sites. The constellation targets network operators and device OEMs in Central Asia, Southeast Asia, and the Middle East that want satellite connectivity without dependence on U.S.-controlled infrastructure. Satellites support direct-to-cell, IoT, and fixed broadband links using existing ground hardware and standard 3GPP modulations, replacing the need for custom gateways or new spectrum filings.

Experiential Labs builds an open-source AI gateway that routes traffic across hosted providers, customer-supplied API keys, and private GPUs through a single endpoint at provider cost. On top of routing, an intelligence layer watches live traffic to recommend cheaper or faster models, improve cache hit rates, and fine-tune a customer-owned model on their own workflows, replacing frontier models at a fraction of the cost and latency. The target customer is any team spending on AI inference that wants cost control, model flexibility, and continuous optimization without rebuilding their stack.

Fabraix builds automated red-teaming agents that continuously find security vulnerabilities in customer-facing AI agents. Its product, Nyx, replaces periodic, manual red-team engagements by running multimodal adversarial attacks through controlled replicas of SaaS products and websites, inserting malicious payloads into pages, documents, files, and tool outputs to probe how a target agent behaves. Enterprise security and AI teams use it to test every model, prompt, or permission change as it happens, not just at a point in time.

Familiar builds an AI dubbing system that translates video content across 99 source languages into 30 output languages while preserving each speaker's voice, face, lips, and delivery. A single proprietary diffusion omni-model processes voice, person, and scene together, replacing the patchwork of separate models and manual pre-processing that rivals and human dubbing pipelines depend on. The target customers are movies and TV studios, broadcasters, education platforms, and ad teams that need localized content at scale without losing the original performance.

Financial Datasets builds a financial data API that connects AI agents to real-time global stock market data, sourced directly from primary filings and reports. It serves investors and AI teams that need earnings figures, fundamentals, KPIs, and news indexed within seconds of publication. The product replaces manual data aggregation and slower, less auditable data feeds that agents would otherwise depend on.

Floracene lets healthcare operators build and deploy HIPAA-compliant internal tools without an engineering team. Staff describe the tool they need, and Floracene deploys it connected to live production data, with authentication and permissions scoped by user role. It replaces the manual workaround where employees build useful dashboards locally but cannot share them or connect them to patient data because IT blocks any deployment that touches protected health information.

Florin operates as a full-stack insurance carrier that replaces human underwriters with an AI model. When a submission arrives, the model reads ACORD forms, pulls satellite imagery, permits, public records, and social media to surface exposures the application omits, then runs tens of thousands of simulations to price the risk and return a bindable quote in under a minute. It targets liability, property, and products coverage across technical sectors including robotics, space, advanced energy, and defense.

FlowManual builds takeoff and estimating software for general and MEP specialty contractors. It replaces disconnected spreadsheets and manual quantity counting by letting contractors measure drawings, count items, and price quantities in a single workspace, from the first sheet to the final bid proposal. The platform targets the pre-construction buying process, covering mechanical, ductwork, concrete, and general scopes.

Gamgee builds a personalized mRNA vaccine platform for dogs diagnosed with cancer. Starting from a dog's tumor biopsy, Gamgee sequences both cancerous and healthy DNA, pinpoints the mutations specific to that cancer, and manufactures a vaccine designed to direct the immune system against those targets. The platform handles each step from initial vet coordination and tissue collection through genomic analysis, vaccine production, administration by a veterinary oncologist, and response monitoring, replacing the ad hoc, trial-and-error approach that defines current veterinary oncology.

GitCafe builds a Git forge designed for AI coding agents and the developers who work alongside them. It provides repository hosting, pull request workflows, and CI infrastructure, replacing the bottlenecks that traditional forges create when hundreds of agents are committing and running jobs concurrently. Teams import existing repositories from GitHub, with optional two-way sync, and agents get the same CLI-driven workflow as human engineers.

Glen builds a shared memory and context layer for companies running AI coding agents. It continuously captures agent sessions from tools like Claude Code, Codex, and Cursor alongside Slack messages, GitHub pull requests, tickets, calls, and docs, then automatically injects relevant prior context into each new agent prompt. This replaces the manual work of re-briefing agents and eliminates the repeated token cost and quality loss that comes from agents starting every task cold.

Graphify Labs builds an on-device knowledge graph engine that indexes codebases, documentation, and project management data into a single, continuously updated graph. Engineering teams use it to give AI coding assistants a shared, queryable map of every repository, replacing ad-hoc codebase re-learning and context fragmentation across teams. An enterprise layer sits on top to formally verify pull requests, proving that a code change preserves prior behavior or returning the exact input that breaks it, which targets high-risk work like refactors, framework upgrades, and large-scale migrations.

Greypoint Industries builds LEGION, an autonomous drone swarm for military use that detects, geolocates, and tracks battlefield emitters by exploiting the enemy's own radio signals. The system targets previously hard-to-find threats, including jammers, command posts, and drone operators, giving warfighters precise, real-time location data on those targets. It shifts forces from reacting to incoming threats to proactively engaging the source.

Grip builds robotic systems that sort and pick objects from live waste and recycling streams. Its hardware and learning-based manipulation software train on every successful and failed grasp, growing more reliable across different facility types and material mixes over time. The systems replace manual sorting, reducing contamination and recovering more recyclable material from streams that change constantly in composition.

Grocalo runs content operations for social media creators, replacing the manual work of organizing libraries, analyzing performance, and distributing posts across platforms. It ingests every video, caption, and post a creator has published, applies video understanding and platform analytics to model what resonates with their specific audience, and handles distribution and engagement across Instagram, TikTok, YouTube, Facebook, Threads, LinkedIn, and Snapchat. The product targets professional creators who publish at scale and want to reclaim time spent on operational tasks.

GUILD manufactures aerospace parts for the U.S. defense market, competing directly for government contracts rather than serving as a sub-tier vendor. Its AI-native operating model connects requirements, pricing, supplier sourcing, compliance, and delivery inside a single system, replacing the fragmented manual coordination that slows traditional defense supply chains. Each contract feeds the system so that cost and lead time drop with scale.

Gutgutgoose formulates personalized probiotics for consumers who want strains that will actually establish in their gut. Customers mail in a stool sample, which Gutgutgoose sequences and uses to build a metabolic model that identifies open colonization niches before selecting and blending the right strains. A follow-up sequencing at 90 days confirms what colonized and what shifted, replacing the guesswork of off-the-shelf probiotic supplements.

Hardware Intelligence builds AI-powered developer tools for semiconductor and chip engineering teams. Its products target the hardware development workflow that underlies AI compute, replacing manual, high-overhead processes used during chip design and verification. Its second product, WaveZip, captures lossless waveforms at a fraction of the size and instrumentation cost of conventional tooling.

Hebbian Robotics builds an open-source SDK and hosted model API for quality control of robotics training data. Data teams use it to run automated checks across video episodes collected from teleoperation, egocentric rigs, and autonomous rollouts, catching problems such as missing hand visibility or inactive manipulation. It replaces ad-hoc scripts and general-purpose vision models with purpose-built checks that scale from thousands to millions of hours of footage without requiring teams to manage their own infrastructure.

HERA automates engineering back-office work for heavy industry teams, handling drawing checks, assembly and BOM reviews, revision comparisons, document drafting, and P&ID review. It serves manufacturers in sectors such as data centers, renewable energy, robotics, heat exchangers, and pressure vessels. The product replaces hours of manual drawing reading and document preparation, returning cited findings traced to governing requirements.

hiloop builds an autoresearch platform that automatically improves AI models and agents against a metric the customer owns. Engineering teams bring a task, an existing model or agent, and an evaluation function; hiloop runs thousands of directed experiments across training data, post-training methods, prompts, tools, and system configuration, then returns verified improvements. It replaces the manual, iterative ML research loop that teams otherwise run themselves.

Hilstart builds an autonomous hardware testing platform for electronics engineers. Its physical agent harness, Tracer, connects to a board and runs bring-up, validation, and firmware tests on real hardware by reading design files such as schematics, netlists, and BoMs into a single approved specification. The product replaces the manual work of writing test scripts, operating bench instruments, and debugging firmware iterations by hand.

Hop Aero builds rocket cargo vehicles that deliver supplies point-to-point into contested and austere locations hundreds of miles away in minutes at hypersonic speeds. The vehicles serve U.S. and allied military customers who need to move critical cargo into environments where conventional air or ground logistics are denied or degraded. The product replaces slower, higher-risk delivery options in those scenarios.

Hoplite builds a cloud platform that runs swarms of AI coding agents for software engineering teams. Developers connect a GitHub repo, and Hoplite mirrors their local environment in the cloud, running agents in isolated sandboxes to handle tasks from issue triage and bug fixes to complex migrations. It replaces the manual work of spinning up environments, triaging Sentry alerts, and shepherding code changes through review by producing built, tested, and verified pull requests automatically.

Hubble retrieves patient medical records for developers building in digital health, legal, insurance underwriting, and clinical trials. Its AI agents pull records from EHRs, payers, and HIE networks using direct API connections, and fall back to voice and browser agents for portals those connections miss. It normalizes everything into a single structured API response, replacing the slow, manual work of chasing records across thousands of disparate systems.

HyperProbe builds a runtime observability layer that lets AI coding agents diagnose and resolve production failures without redeploying code. Its SDK injects read-only probes into live services to capture exact variable states at the point of failure, giving agents precise evidence that logs alone never record. The product targets engineering teams whose on-call cycles and production war rooms are driven by missing runtime data rather than missing fixes.

Illume Labs builds a 24/7 AI health companion that unifies data from wearables, blood panels, and genomic tests into a single model, then delivers personalized, actionable insights via text. It replaces the manual work of cross-referencing fragmented health platforms by detecting trends across data sources those platforms analyze in isolation. Customers can ask conversational health questions and dig deeper into their data through a full companion platform.

IMPACT Drones builds autonomous interceptor drones that protect fixed civilian infrastructure, including data centers, power grids, refineries, and airports, from hostile drone threats. The system ships as a sealed container pre-loaded with interceptors: when sensors detect a threat, an interceptor launches and neutralizes it without a crew on site. It replaces legacy kinetic effectors that cost over a million dollars per shot with a lower-cost interceptor sized to match the threat.

Induction Labs builds foundation models that are intrinsically motivated to learn, updating continuously from their own observation rather than relying solely on static training data. The research targets a core limitation of current AI: models that cannot discover knowledge beyond what they were originally trained on. The goal is superintelligence that learns the way curious humans do, compounding knowledge across experience.

Inkbox gives AI agents a persistent identity, complete with a dedicated email address, phone number, iMessage, and an internet address, so people and services can reach them the way they reach human teammates. Agents can send, receive, and initiate conversations across those channels from a single identity, and coordinate with other agents through a built-in agent-to-agent communication layer. The product replaces ad-hoc workarounds developers use to wire AI agents into real-world communication flows.

Instance builds automated verification for robot training pipelines. Given a task description and camera footage, it returns a success verdict with evidence for each rollout episode, replacing manual human judgment of whether a robot completed its task. Robotics teams use it to filter and trust training data faster, accelerating policy improvement.

Instaplay builds a multiplayer AI game maker that turns text prompts into playable browser games, no coding required. Hobbyists and game creators use it to generate party, co-op, competitive, and solo games directly in the browser. It replaces the manual work of writing game code, setting up multiplayer infrastructure, and iterating on prototypes.

Insurf builds AI infrastructure for health insurance decisions, serving companies, families, and medical clinics. Its model Surely prices every filed plan in a given county to identify the lowest true cost for an employer or individual, replacing manual plan comparison. Its product Inveto generates source-cited prior authorization requests and denial appeals for clinics, replacing the administrative work that delays or loses reimbursed care.

Isengard Industries Inc builds AI-guided precision munitions and combat UAVs for allied militaries, supplying the designs, autonomy stack, tooling, and supply chain so that partner nations can manufacture strike and counter-drone systems at scale on their own soil. The product replaces dependence on large defense primes and foreign imports for front-line strike capability. Three systems, including the AQV 114, AQ 101, and XQ 500, are fielded and in active combat use in Ukraine.

Jcode builds an open-source, terminal-based AI coding agent written in Rust, aimed at software developers who want to run many coding tasks simultaneously. It focuses on the "harness" surrounding the language model: the tools, context, memory, and feedback loops that determine what the model can actually accomplish. Instead of watching a single agent finish a task, developers spin up dozens of parallel sessions, replacing the serial, one-task-at-a-time workflow that characterizes most AI coding tools today.

Kandor pays private medical practices immediately after a procedure is performed, replacing the 30-to-90-day wait for reimbursement from insurance payers. The product sits between practices and insurers, taking on the billing and collections cycle so clinicians collect on day one instead of chasing receivables for months.

Kara grows single-crystal diamond wafers in its own reactors and handles all growth, processing, and characterization in-house. Its engineered diamond targets thermal management, power electronics, RF systems, quantum computing, and photonics, replacing conventional semiconductor materials like silicon, GaN, and SiC in applications where heat dissipation, breakdown tolerance, and carrier mobility are limiting factors. Customers include hardware builders in data centers, aerospace, and quantum hardware who are constrained by the physical ceilings of existing semiconductor substrates.

Kebra captures data from field jobs, sites, and assets and builds it into a continuously updated operating memory that AI can query and act on. AI agents use that context to automate the follow-on work across the systems field service companies already run. Each completed job deepens the company-specific context, replacing the tribal knowledge and manual coordination that currently limits what software can do in field operations.

Lamb Labs builds MPUs (Model Processing Units): custom chips that hardcode an entire LLM's weights directly into silicon, eliminating the memory-bandwidth bottleneck that forces GPUs to shuttle weights between memory and compute. The first MPU carries Larry, a one-bit quantized 27B-parameter model compressed to 3.8 GB, targeting throughput above 20,000 tokens per second. The product replaces GPU-based inference stacks for workloads where speed and energy efficiency are the primary constraints.

Last Accounting Company delivers a full-stack accounting service, covering bookkeeping, VAT filings, payroll, and financial statements, to startups and SMBs. Proprietary AI agents handle the day-to-day work, from reconciling transactions to closing the books, while a certified accountant reviews and signs off each output. This replaces both the traditional outsourced accountant and standalone accounting software in a single service.

LATO builds an agent-native research platform for investors, replacing the weeks-long, six-figure process of commercial due diligence. Its agents source and conduct voice interviews with hundreds of domain experts, customers, and operators, then combine those interviews with thousands of public and proprietary documents into a single study. That study feeds a live market simulation investors can query to run scenarios and forecast how a market will move.

LemonLime automates sales and marketing for small businesses, replacing the daily work of prospecting, writing outreach, managing social media, and running promotions. Each day the platform surfaces new leads, ranks the best fits, and drafts personalized messages across email, LinkedIn, and social channels for owner review. Small business owners get a full go-to-market operation without hiring a sales or marketing team.

Levocred AI builds AI agents for credit funds, fintechs, and banks that automate the analytical and operational work lenders assign to back-office staff and analysts. The agents handle underwriting memo drafting, portfolio monitoring, compliance certificate generation, and borrowing base calculations, taking a loan tape and credit agreement as inputs and producing cited, ready-to-send outputs. The product replaces the manual analyst hours that those workflows currently require.

Libra Robotics builds collaborative robotic crews for utility-scale solar farm and data center construction. Its AI control system coordinates specialized robots to handle panel transportation, alignment, placement, and fastening, end to end, across daytime and nighttime shifts. The system replaces manual installation crews on large-scale infrastructure projects, delivering consistent cycle times and throughput under supervised field operation.

Litmus builds a technical hiring assessment platform that replaces manual screening, take-home problems, and early-round interviews. It ingests a company's own codebase, tickets, and job descriptions to generate async coding tasks that reflect real work, then evaluates candidates on how they build, how they use AI, and how they reason through decisions. Engineering teams see a candidate's full working trajectory before any engineer spends time on a call.

Locke builds an AI-native government affairs firm that helps companies lobby policymakers, win public contracts, and monitor legislation and regulations. Its LockeOS platform layers AI agents trained on expert policy data over the largest assembled U.S. government dataset, with human policy staffers and lobbyists reviewing outputs and managing government relationships. The product replaces traditional lobbying firms and manual policy-tracking work for organizations seeking to operate in the public sector.

Luca IQ builds tax software and an API-first tax engine for accounting firms and fintechs. It covers the full return workflow, from client document intake and automated workpaper preparation through CPA review, e-signature collection, and IRS e-filing, replacing the disconnected desktop tools and manual data re-entry that currently consume firm staff time. The underlying tax calculation and e-filing engine is also available as an API for fintechs building tax products.

Lumeria builds a multispectral phone-clip camera, the Lumoscope, that gives consumers clinical-grade skin imaging at home. The device reads redness, pigmentation, and sebum in a single scan, then tracks changes over time to surface what is driving breakouts, wrinkles, and inflammation before they are visible to the naked eye. The companion software cross-references that scan history against an ingredient database of 18,000 products to tell users what is actually working in their routine, replacing guesswork and periodic dermatology visits.

machine0 builds persistent cloud VMs purpose-built for AI agents running long workloads, from multi-hour coding tasks to multi-day research jobs. Agents provision and drive the machines directly through a CLI or a remote MCP server, replacing ad-hoc infrastructure cobbled together from general-purpose cloud offerings. Configurations scale from a single vCPU to 60 vCPUs with 240 GB RAM and up to eight H200 GPUs.

Magma turns the interaction logs that AI agent builders already produce into a licensed data asset sold to AI labs and researchers. Agent developers connect their production data, approve the scope, and Magma packages traces, corrections, and completed workflows that labs buy to understand how real work happens inside real products. The work Magma replaces is discarding that data entirely.

Maingen builds simulation environments for industrial operations so that AI frontier labs can train and benchmark models on running factories and other industrial systems. The product replaces the need for real-world trial-and-error in high-stakes physical environments by providing a safe, structured setting to develop models that can operate industrial infrastructure. Customers are AI labs working to deploy models in industrial contexts.

Manifold deploys autonomous robots into warehouses to handle case picking, trailer loading and unloading, and palletizing, replacing manual labor for physically demanding case-handling tasks. The robots integrate with existing ERP, WMS, and WES systems, map operating zones on arrival, and use remote human operators to resolve edge cases. The service targets warehouse operators who cannot or do not want to undertake a large retrofit project to adopt automation.

Marble builds an AI-native operating system for restaurant back-of-house teams, replacing the sprawl of spreadsheets, clipboards, and disconnected tools that most restaurant groups rely on today. Computer vision and voice AI handle inventory counts, machine-learning models forecast demand at the item level, and AI agents run end-to-end workflows across procurement, scheduling, prep, menu engineering, and invoice reconciliation. The net result is a single system that replaces the manual labor and guesswork behind food cost control, waste reduction, and labor optimization.

Marengo designs data centers for large-scale builders, cutting the time and cost of engineering work that traditional firms handle manually. Its AI-native approach compresses the design process, replacing the slow, labor-intensive workflows that conventional architecture and engineering teams use to produce data center plans.

Marker embeds with B2B companies, maps how work actually happens inside their teams, and ships custom AI agents into production in weeks through its Marker Command Center platform. It replaces the slow, incomplete work of traditional consultancies and internal AI hiring by keeping engineers on-site until the client runs as a fully agent-first organization. Each deployed agent accelerates the next, compressing rollout timelines from weeks to days over successive cycles.

Markov captures how people use computers and converts those recordings into structured training data for AI labs building computer-use models. Each dataset pairs screen recordings with synchronized mouse and keyboard events, narrations, problem statements, and verifiable gold outputs across professional tools such as AutoCAD, Salesforce, Blender, and Photoshop. The product replaces the slow, ad-hoc process frontier labs would otherwise use to source and label high-quality interaction data at scale.

Mass Magnetics manufactures high-performance permanent magnets in the United States from recycled rare-earth scrap, targeting aerospace, robotics, defense, and industrial manufacturers. Its NdFeB magnets, graded up to N52, are fully milled and sintered domestically, making them compliant with DFARS 252.225-7052 rules that restrict covered-country materials. The product replaces foreign-sourced magnets for customers who must meet domestic supply requirements without accepting lower magnetic performance.

Mentlio builds an engineering intelligence platform that gives enterprises visibility into AI coding tool spend and its effect on developer output. It measures productivity signals locally on-device, so prompts and source code never leave the machine, then routes requests, compresses context, and trims prompts to cut token waste across every engineer's seat. The platform replaces ad-hoc AI bill reviews and gut-feel ROI assessments with per-engineer metrics tying AI spend directly to delivery.

Meteoric builds drones that disperse low and mid-altitude clouds over solar farms, replacing lost generation caused by cloud cover. The drones fly into clouds and alter droplet properties to reduce reflectivity, letting more sunlight reach panels below, without chemicals or new ground infrastructure. Modelled output uplift ranges from 10% to 30% depending on grid region and cloud type.

Mireye builds a unified API and MCP layer that gives AI agents accurate, cited data about the physical world. Spatial facts today are scattered across thousands of government portals, GIS maps, parcel records, and regulatory PDFs, each built for human navigation. Mireye indexes those sources and exposes them through a single endpoint so agents can query elevation, land use, environmental signals, and more without hitting each silo.

MOCHI.TV produces and distributes vertical anime shorts, each under one minute, for mobile audiences who scroll through episodic content the way they scroll social feeds. The platform uses AI to ship new shows daily, replacing the months-long production cycles of traditional animation studios. Viewers browse a catalog of original action, romance, and fantasy series and watch episodes in bite-sized runs.

Molagri develops biopesticides engineered to hit a molecular target found only in pest insects, killing crop pests while leaving pollinators and beneficial insects unharmed. The products are designed with predictive resistance modeling so they remain effective as pest populations evolve, replacing conventional broad-spectrum pesticides that degrade ecosystem health and accumulate resistance over time. Production is biobased, cutting emissions at both the manufacturing and field-application stages.

Mosaic builds shared memory infrastructure for teams that work with AI coding agents. It captures every session's context, decisions, and history in one place, so any team member or agent can pick up an ongoing project without losing thread. The product replaces ad-hoc context-passing between developers and disconnected agent sessions.

Moving Atoms builds simulation and data infrastructure for physical AI, giving robotics teams the assets, synthetic training data, and evaluation platforms they need to train and deploy robot policies at scale. Its RobotGym platform covers the full stack: robot ingestion, world modeling, skill creation via teleoperation or video-to-sim transfer, and deployment evaluation with regression testing. The company's world model, Atom 1, is trained on internet-scale video to predict physical interactions, replacing the need for costly real-world data collection at each stage of robot development.

Nebula Security builds an AI-powered code and cloud security platform backed by a team of competitive hackers with confirmed zero-days in Chrome and the Linux kernel. Nebu continuously monitors code and cloud infrastructure, surfaces verified vulnerabilities, and delivers merge-ready fixes, replacing periodic manual audits with always-on automated coverage. When findings are critical, the same researchers step in directly.

Neuromorphic builds autonomous robots that operate inside existing wet labs, handling the physical work that falls between automated instruments: moving samples between workstations, pipetting, loading equipment, and running end-to-end experiments. The robots work with a lab's current infrastructure and instruments rather than requiring dedicated hardware or reconfigured spaces. Biotech and life-science labs use them to replace the manual technician hours spent connecting discrete automated steps into complete workflows.

Neuron Industries builds the Cortex AIC, a purpose-built industrial controller that replaces the PLC, HMI, historian, and vendor IDE with a single device. Controls engineers program machine behavior in plain English through Synapse, Neuron's on-device IDE, instead of writing ladder logic across multiple legacy tools. The hardware targets any machine-level controls project, from engineered wood lines to chemical processing, and compresses commissioning from weeks to hours.

Nex builds an AI RevOps engineer for B2B go-to-market teams, running the high-volume, multi-step workflows that skilled GTM engineers typically handle manually across sales, revenue operations, and customer retention. Its agent roster covers outbound prospecting, pipeline management, churn prevention, win/loss analysis, and more, connecting across a broad set of CRM and sales tools. The product replaces the operational labor of configuring, maintaining, and executing those workflows by hand.

Nori builds a bimanual mobile robot, the A3, that handles household and light commercial tasks such as tidying, folding clothes, loading dishes, and restocking shelves. At $1,688, it targets homes and small businesses priced out of existing robotics hardware. Users train the robot on custom workflows through the Nori Lab app, and skills trained by one user feed back into a shared pool that makes the broader fleet more capable over time.

Olam Labs builds multi-agent simulation environments for evaluating and training AI models on character and agentic behavior, work that current benchmark datasets handle poorly. The company works with frontier AI labs and researchers, running models through complex simulated worlds where social strategy and real-world reasoning can be measured and improved. Its public platform, Multi-Agent Arena, pits humans against multiple AI agents in social strategy games and generates the behavioral datasets that power those evaluations.

Omanta runs a personalized research program for patients with cancer, rare diseases, or unresolved diagnoses who have run out of standard options. For each case, a dedicated team of PhD scientists collects clinical records, performs multiomic analysis, and builds a therapeutic roadmap, replacing the patchwork of one-off consultations and generic diagnostic workups those patients would otherwise navigate. Clinicians receive a continuously updated evidence base to guide treatment decisions for their most complex cases.

OneCLI gives every employee a personal AI agent connected to their existing tools, including Salesforce, GitHub, Gmail, Notion, and Snowflake, operated from chat or Slack. Each agent runs in its own sandbox under admin-defined policies, with credentials injected at the network layer per request rather than stored in the agent itself. It replaces manual, tool-by-tool work on CRM updates, BI queries, and incident response with audited, approval-gated automation.

OneTriangle builds an inference engine that cuts the cost and latency of running large language models by splitting the prefill and decode steps across model sizes. During prefill, a small model processes the input and its KV cache tensors are reprojected into the large model's attention space, so the large model begins decoding against a populated cache it never had to compute. This replaces the standard approach where the full large model handles every token of a long prompt at large-model rates.

Ooak Data builds reinforcement learning environments for AI labs and model developers who need to evaluate agents on realistic enterprise work. The company sources multimodal data directly from real companies, including documents, communications, and project management tools, then anonymizes it into digital twins that preserve organizational structure and complexity. The resulting RL environments replace synthetic benchmarks with multi-step, multi-tool tasks calibrated to expose gaps in frontier models rather than confirm what they already do well.

OpenRelay runs an inference delivery network that routes AI model requests across a distributed mesh of GPUs from multiple hardware providers, automatically selecting the best node by latency, cost, availability, and model performance. It targets engineering teams running high-volume inference workloads who need resilience and cost control beyond what a single cloud provider offers. The product replaces manual provider management, custom failover logic, and the operational overhead of stitching together disparate GPU sources.

OpenTag builds an AI coworker that operates inside Slack and Microsoft Teams, giving enterprise teams a single assistant with full context across their tools, knowledge base, and communications. It answers questions, executes delegated tasks, joins calls, and runs automated workflows, replacing the manual work of routing requests, searching documentation, and keeping company wikis current.

OpenTrade turns stock market investing into a game-native experience, letting users practice trading through daily games and social competition rather than traditional brokerage interfaces. It targets a generation accustomed to mobile games and social apps, replacing the dry, form-heavy onboarding of conventional trading platforms with tap-to-play formats like Tickerdle and Higher/Lower. The product sits at the intersection of financial literacy and entertainment.

OpenVector connects existing cameras to frontier vision-language models, enabling businesses to automate monitoring and follow-up actions without new hardware. Operators describe a task in plain English, and OpenVector builds a workflow that watches the camera feed, detects physical events such as an empty staging bay or a safety violation, and acts on business software in real time. It replaces manual camera monitoring and the human coordination work that follows each observation.

Operon builds a structured data layer for heavy industry plants, ingesting the documents a facility runs on, such as P&IDs, isometrics, schematics, and compliance records, and converting them into a single queryable graph of equipment, lines, loops, and instruments. Process and manufacturing engineers use it to search plant documentation in natural language, with every answer traced back to a specific tag and drawing region. It replaces manual document hunting and disconnected drawing archives across safety reviews, compliance workflows, and engineering change management.

Orvixo builds an AI planning and optimisation system for manufacturers, covering automotive, food and beverage, defence, electronics, and other industrial sectors. It models a plant's machines, tooling, people, shifts, materials, and customer orders into a live solver that re-plans automatically when conditions on the floor change. The product replaces static spreadsheets and manual scheduling work that cannot react in real time to order amendments, supply constraints, or production disruptions.

OS3 builds US-manufactured bimanual mobile semi-humanoid robots, training them on 150,000 hours of video data to handle physical labor tasks. The first target is hospitality, where OS3 operates deployments directly, with biotech as a second vertical. The robot replaces human labor on repetitive physical tasks such as towel folding, learning new tasks from as few as two to four hours of demonstrations.

Osmaura builds an AI-driven business development platform for law firm partners. Its agents continuously scan the web to surface new client leads and cross-selling opportunities, delivering daily contact lists with context on who to reach, why now, and what to say. It replaces the manual prospecting and pipeline management that partners are neither trained nor incentivized to do themselves.

Osseus builds a development platform for robotics teams that connects their existing tools, CAD, ECAD, GitHub, Jira, docs, and test runs, into a live dependency graph spanning every part, board, requirement, test, and decision. When engineers make a hardware change, the platform traces the impact across mechanical, electrical, and software layers, runs affected tests and simulations, and generates fixes on branches. It also hosts a component marketplace where teams source or substitute parts during prototyping, and it centralizes design review by surfacing supplier feedback, past decisions, and prior engineering context inline.

Pacific mass-produces modular, factory-built data center pods and deploys them on stalled construction sites that already have grid connections but no active project. Those sites bleed capital on idle debt; Pacific's pods land within days and convert the unused power into compute revenue. Diamond-wafer cooling lets the pods operate in extreme climates, removing the geographic constraints that ground conventional data center builds.

Palette builds a multimodal AI studio that consolidates video, image, and structured content generation into a single workflow. Creative and marketing teams at enterprises use it to produce audience-ready assets from one source, replacing the patchwork of separate generation and editing tools. The platform tracks the logic behind each iteration so outputs improve automatically over successive passes.

Palisade builds an AI security platform that detects and remediates OS-level vulnerabilities across enterprise device fleets. It monitors at the kernel layer, using syscalls, eBPF traces, process and memory events, and live network packets to catch threats, including privilege escalation, remote code execution, and memory corruption, that network, cloud, and application-layer tools cannot see. Once a vulnerability is found, Palisade pushes the fix fleet-wide in minutes, replacing a remediation process that otherwise takes weeks across hundreds or thousands of machines.

Palisade builds AI sales agents for online marketplaces, deploying a dedicated agent for every buyer, seller, and staff member on a platform. Each buyer agent learns individual preferences, tracks live inventory and pricing, and follows a shopper from discovery through checkout across web, SMS, and email. The product targets trading card platforms, vintage retailers, and short-term rental operators, replacing the manual work of personal sales outreach and passive browsing that ends without a conversion.

Pango builds an agentic operating system for e-commerce back-office logistics, covering multi-carrier delivery management, order tracking, and returns and exchanges. It targets e-commerce brands that currently manage these workflows manually or across fragmented tools. Pango replaces the coordination work across shipping carriers, customer-facing tracking, and returns processing with autonomous agents that handle each step end to end.

Parasma trains living human brain cells to perform AI compute tasks, replacing silicon GPUs with biological neurons. Text is converted into electrical stimulation, the neurons process it, and their responses are decoded directly into tokens. The work targets the core inference loop that GPU clusters currently handle.

Peer operates a truckload freight brokerage for shippers moving dry van, reefer, flatbed, hazmat, and port drayage loads. It quotes rates within the hour, vets and books carriers, tracks shipments live, and closes each load with matched proof of delivery and invoice the same day. The product replaces the manual coordination, carrier vetting, and exception-handling that shippers otherwise do themselves or delegate to traditional brokers.

Pennant builds a corporate governance operating system for institutional investors, public companies, and corporate advisors. It centralizes proxy meeting analysis, policy-aligned vote recommendations, and audit trails in one platform, replacing manual research and fragmented workflows that governance, stewardship, compliance, and portfolio management teams use today.

We make law firms AI-native.

Petrarch sources and prepares proprietary internal data from manufacturing and industrial companies, including codebases, project files, and financial records from bankruptcy courts, for use in AI model training by frontier labs. It handles de-identification and data preparation, replacing the manual, fragmented work that labs would otherwise do to acquire and clean this kind of enterprise data.

Pluto builds an AI voice agent that converts a 10-minute career conversation into a structured professional profile, capturing competencies and job intent that resumes and LinkedIn profiles leave implicit. Hiring companies and AI recruiting agents query that profile to find and reach candidates who match what they are actually looking for. The product replaces manual sourcing and the guesswork of interpreting static credentials.

PokerClubHub builds a licensed platform for hosting private online poker clubs, serving both club owners and players who have moved away from regulated sites overrun by professionals and bots. Club owners on competing platforms like ClubGG handle deposits manually in crypto, exposing players to scams and slow withdrawals. PokerClubHub takes custody of player deposits directly, automates all payments, and pays rake back to club owners as affiliate fees, replacing a fragmented, trust-dependent cash flow with a managed, licensed system.

Poth Labs builds an agentic customer intelligence platform for B2B product and research teams. It connects to existing customer data sources, such as support tickets, calls, and usage events, identifies the questions that data cannot answer on its own, and then runs adaptive surveys to collect the missing evidence directly from customers. The platform replaces manual research workflows by automatically forming hypotheses, targeting the right users, and folding each response back into the evidence base.

Praxis Robotics collects the real-world training data that physical AI developers cannot source on their own: egocentric video, 3D scans, and multimodal capture from inside active industrial and residential environments. It embeds within large conglomerates to reach workers across a wide range of environment types, then supplies that data to frontier labs and humanoid robotics companies. The work it replaces is the costly, fragmented effort those developers would otherwise spend building proprietary data-collection pipelines from scratch.

Prescience, Inc. builds an AI-driven employer health benefits platform that creates, operates, and delivers health plans for startups, SMBs, and growing companies. Its core system steers employees toward care pathways that reduce total medical spend, replacing the role that traditional insurers and benefits brokers play in designing and administering employer coverage. The result is a plan Prescience positions as higher-quality than Fortune 500 offerings at materially lower premiums.

PRINCEPS builds insurance products for data centers and neoclouds, covering GPU residual value, SLA breaches, and outage risk. It operates as a managing general agent, underwriting the compute infrastructure risks that traditional insurers do not price. The coverage is structured to reduce financing costs for operators building out large-scale compute capacity.

Prized builds a cloud development environment that mirrors a local Mac setup. Developers get a persistent remote machine with repos, dependencies, and tools pre-configured: files sync bidirectionally in roughly 100 ms, remote ports appear on localhost, and terminal sessions survive after the laptop closes. It replaces the need to maintain a local dev machine, letting both developers and coding agents work continuously without being tied to the device in front of them.

Prodigy Research trains a foundation model purpose-built for quantitative trading and deploys it as a proprietary autonomous trading system in live financial markets. The lab targets alpha generation across global equities, replacing the manual research and signal development that human quant teams perform at traditional trading firms. Its co-founders bring hands-on experience training state-of-the-art models at Google DeepMind and Jane Street.