YC Spring 2025 (P25) companies

Acolite builds AI teammates for insurance agencies, handling the repetitive back-office work of processing submissions, issuing certificates of insurance, and mapping statements of values across any agency management system. Agents forward an email and Acolite completes the task, replacing manual workflows that would otherwise consume staff time. The company also operates as a licensed construction insurance broker, writing coverage for contractors across general liability, workers' comp, and related lines.

Adentris builds AI compliance and revenue integrity software for hospitals and clinics. It checks clinical documentation against regulatory requirements in real time, inside the EHR, catching coding errors, compliance gaps, and denial risks before a claim is submitted. The product replaces manual chart review and billing audits that today require large teams of coders, compliance officers, and revenue cycle staff.

Aegis builds an AI platform that helps hospitals and health system billing teams recover revenue from denied insurance claims. It automates the full appeals cycle, from identifying denials as they arrive to drafting responses, submitting appeals, and tracking them through to payment. The platform also flags rising denial patterns to prevent repeat losses and checks remittances against payer contracts to catch underpayments, replacing manual billing team work across each of those steps.

Airweave builds a context retrieval layer for AI agents, letting them query data across productivity tools, databases, and document stores in a single request. Developers connect their apps and data sources to Airweave, which indexes the contents into searchable knowledge bases that agents pull from at runtime. This replaces custom retrieval pipelines that teams would otherwise build and maintain for each data source separately.

Alkali builds AI takeoff software for structural steel fabricators. It reads construction bid sets, automatically detects and counts beams, columns, joists, braces, and connections, and outputs a bill of materials with tonnage calculations, 3D model, and nesting plans. It replaces manual quantity takeoff and scope review that estimators previously did by hand across large drawing packages.

Almond builds industrial-grade dual-arm robots for teams developing physical AI applications. Its flagship robot, Axol, is a 7-DOF-per-arm platform with 860mm reach, a 500Hz control rate, and fully internal cable routing, targeting robotics builders who need a capable, real-world-ready manipulation platform. Axol ships with an open-source Python SDK, VR teleoperation, and bimanual inverse kinematics, replacing the need to design and integrate custom hardware from scratch.

Aravolta builds monitoring, control, and optimization software for data centers, serving colocation operators and enterprise infrastructure teams. A single utility node onboards every building, electrical, and compute asset across any vendor or protocol, replacing manual tracking and fragmented tools like SCADA, EPMS, CMMS, and BMS systems. Operators get unified visibility into hardware health, power usage, energy metering, and resource utilization through one platform.

Atlog builds voice AI agents for B2B companies that run outbound and inbound call operations, handling collections, customer service, and lead reception. The agents are designed with TCPA, FDCPA, and state-level compliance rules built in, replacing human-staffed call centers and manual outreach workflows. Businesses use Atlog to run automated call and text campaigns without exposure to regulatory violations.

Attune builds drop-in acceleration for software build systems, targeting developers and engineering teams bottlenecked by slow compile and deploy cycles. Instead of requiring a migration to a new build tool, Attune wraps existing toolchains: a single command prefix like "hurry cargo build" replaces the standard invocation and delivers dramatically faster builds. The product eliminates the waiting time that compounds across every code change a developer makes each day.

Atum Works builds scalable 3D lithography equipment for semiconductor manufacturing, targeting the fabrication of arbitrary 3D metal structures at submicron resolution. The technology extends chip scaling into the vertical dimension, replacing the planar constraints of conventional lithography with a foundry-grade 3D printing process. It addresses the work currently done by 2D lithography systems that cannot produce complex three-dimensional metal geometries at wafer scale.

Auctor builds an agentic operating system for software implementation teams and system integrators. It centralizes decisions, artifacts, and project context across the full implementation lifecycle, from pre-sales through go-live, replacing the scattered documents, siloed memory, and manual coordination that cause delays and rework. The platform uses AI to generate deliverables and carry forward institutional knowledge so each project runs more consistently than the last.

Avallon AI builds AI agents that automate claims operations for insurance carriers, third-party administrators, and their partners. Its agents handle the work that claims teams do manually: taking first notice of loss by phone, email, or file upload; answering inbound status calls with live claim data; and auto-dialing employers, providers, and injured workers to collect responses. The platform covers workers' comp, life and health, and property and casualty lines.

Aviro builds post-training datasets and reinforcement learning environments for frontier AI labs and large enterprises. It designs complex, long-horizon task environments across coding, computer use, knowledge work, and research workflows, then converts them into the training signal that teaches AI models to explore and solve multi-step problems. The work it replaces is the manual construction of challenging, realistic environments that AI developers need to push models beyond current capability ceilings.

Beluga runs influencer marketing campaigns for B2B companies, connecting brands with creators and managing the sponsorship process end to end. It replaces the manual back-and-forth of outreach, negotiation, and coordination that in-house teams or agencies otherwise handle. The result is attributable performance from creator partnerships at a scale most marketing teams cannot reach on their own.

Besimple AI builds the audio data layer for AI voice and speech model development. The company collects, annotates, and licenses conversational audio datasets spanning 15-plus languages, dialects, and accents, with human expert annotators handling transcription and speaker diarization. It replaces the work of sourcing and processing raw audio in-house, giving AI teams ready-to-use training and evaluation data for automatic speech recognition and voice agent models.

Better Auth builds an open-source authentication framework for TypeScript developers, covering everything from basic credential login to enterprise SSO, SAML 2.0, SCIM, and AI agent auth, all running inside the developer's own backend against their own database. It replaces hand-rolled auth code and third-party auth services by shipping sessions, social sign-on, multi-tenancy, passkeys, and fraud detection as composable, plugin-based modules. An infrastructure layer on top adds user management, transactional messaging, and global session storage for teams that need to scale.

BitBoard builds an AI-native spreadsheet and analytics tool for business teams who need to model, analyze, and visualize data through natural language. Instead of manually writing formulas or queries, users describe what they want in chat and get traceable answers backed by connected data sources. It replaces the back-and-forth between analysts and stakeholders by generating live, repeatable dashboards that carry their sources and logic forward.

BitPatrol scans code repositories in real time to catch exposed credentials, API keys, and tokens before they cause a breach. It integrates directly into GitHub workflows and CI/CD pipelines, analyzing every commit automatically and triggering remediation through PagerDuty, Slack, or custom webhooks. The product replaces manual secret audits and fills gaps left by conventional scanning tools.

Blaxel builds cloud infrastructure for AI agents that execute code inside isolated sandboxes. It replaces traditional sandbox providers that impose cold-start delays and expiration limits by keeping microVM sandboxes suspended indefinitely, resuming them in under 25ms with memory state intact. Agent logic runs co-located with those sandboxes, cutting the network round-trips that add latency between an agent and the code it runs.

Bloom lets anyone build and deploy cross-platform native mobile apps through AI, without writing code. It targets founders and makers who want functional MVPs fast, replacing the traditional development cycle of writing frontend and backend code by hand. Unlike similar tools, Bloom automatically provisions a backend, so the output runs as a real app rather than a static prototype.

Bluejay tests, monitors, and improves voice and chat AI agents for B2B companies deploying conversational AI at scale. It runs lifelike simulated interactions across voice, chat, and text to catch defects before and after production, replacing manual QA review of agent calls. Businesses use Bluejay to validate agent workflows, replay real production calls, and get continuous feedback on agent performance.

Blueshoe builds an AI platform that runs legal matters end to end for attorneys and their clients. On the attorney side, it handles intake, records gathering, drafting, deadline tracking, and client updates, with the attorney reviewing each output before it goes out. On the consumer side, it lets individuals describe a legal problem in plain language and moves the matter forward without billable hours, replacing the manual administrative work that consumes law firm capacity.

Bond builds an AI chief of staff for CEOs and executives, pulling tasks, follow-ups, and commitments from email, calendar, Slack, and project tools into a single prioritized list. It monitors activity across those sources in real time, surfaces what needs attention, and handles or delegates routine items. The product replaces the manual work of triaging inboxes, tracking open threads, and deciding what to act on next.

Boost Robotics builds autonomous mobile manipulation robots that inspect and maintain data centers. The robots handle physical tasks, such as triaging and addressing hardware failures, that today require on-site technicians. Data center operators use Boost Robotics to run inspections more frequently, cut response times to failures, and reduce exposure to SLA violations and insurance costs driven by staffing constraints and accelerating infrastructure growth.

Bramante Biologics sells cell culture media and buffers, such as DMEM, RPMI, and DPBS, to research labs at uniform list pricing with no negotiation required. It also produces custom media and buffers to a lab's own formulation and fill size, quoted within 24 hours. The company replaces the standard supplier model, where pricing is opaque and turnaround is slow, with same-day shipping on stocked items and a public catalog accessible via an AI agent interface.

Cactus builds an AI-powered call center for home service businesses, covering trades such as HVAC, plumbing, and roofing. The system answers inbound calls around the clock, qualifies callers, schedules jobs, and runs follow-up outreach, replacing the work of a human dispatch and customer-service team. Business owners capture leads and re-engage past customers without adding headcount.

Capacitive builds a secure data gateway that gives AI coding agents and agent frameworks structured access to a company's internal data. Engineering teams connect it once to their existing data sources, then define fine-grained permissions, monitor usage, and filter sensitive content across platforms. It replaces ad-hoc, manual context-passing that developers otherwise handle themselves each time an agent needs internal knowledge.

Cascade Space builds ground station infrastructure and software for lunar and deep space missions, serving commercial spacecraft operators who are locked out of the oversubscribed Deep Space Network. Its array-based architecture uses clusters of smaller dishes to deliver scalable, redundant communications capacity that a single large dish cannot match. The integrated platform cuts spacecraft iteration cycles from weeks to hours, replacing fragmented and manual ground operations.

Casco builds an AI penetration testing platform for software teams shipping web apps, APIs, cloud infrastructure, and AI agents. It continuously runs simulated attacks against a company's full attack surface, proves exploitable vulnerabilities, and retests every fix, replacing the slow, point-in-time engagements of traditional human pentesters.

Caucus builds a constituent services platform for government offices, replacing fragmented inboxes, manual casework, and slow response cycles with a unified system that triages, routes, and responds to constituent inquiries automatically. Government staff connect email, web forms, mail, and phone into a single queue, then let AI agents handle routine requests and batch responses. The work displaced is the manual triage and drafting that can take public servants weeks per case.

Chestnut originates mortgages using AI agents, handling purchases, refinances, and cash-out loans for homeowners in Texas and Colorado. The AI handles the underwriting and processing work that loan officers and processors do at traditional lenders, cutting the interest rate customers pay. The founding team has prior operating experience scaling a mortgage business that handles a significant share of U.S. loan volume.

Chiron builds an iPad app that reads handwritten math in real time and gives students just enough guidance to move forward on their own, replacing the moment-to-moment intervention a human tutor would provide. The app targets students working through algebra and similar subjects, stepping in only when they are stuck rather than solving problems for them.

Chonkie builds an open-source data ingestion library for developers building AI applications. It splits large documents into structured, context-appropriate chunks so language models receive relevant input within their token limits, replacing hand-rolled preprocessing pipelines. Developers use it to improve the accuracy and speed of retrieval-augmented generation and similar AI workflows.

chrt builds a logistics platform for freight forwarders and shippers that need to move time-critical cargo, including pharmaceuticals, medical devices, and specialized manufacturing parts. The platform coordinates every leg of a shipment, from ground pickup through airline handoffs to final delivery, inside a single live order. It replaces manual tracking across carrier portals and fragmented courier coordination with end-to-end visibility and early exception management.

Clado builds an agentic people-search platform that gives sales, recruiting, and operations teams instant access to a database of over 800 million professional profiles. Users query the database in plain language to surface candidates, prospects, or contacts, replacing manual research across LinkedIn, spreadsheets, and disparate data sources. The product is aimed at any team that needs fast, structured intelligence on people at scale.

Claim Health builds a revenue operations platform for post-acute and home-based care agencies. It connects documentation to billing outcomes, catching coverage gaps, authorization issues, and claim errors before they trigger denials or payment delays. The platform replaces manual portal logins, fragmented referral intake, and reactive claims workflows with automated digital workers that move a patient from referral to discharge without billing leakage.

Clarm builds AI coworker agents for non-technical operations teams in industries such as FMCG, banking, and aviation. Its Atlas platform lets those teams assemble agents from building blocks pre-approved by IT and compliance, then deploy them to handle repetitive back-office work: reading order emails and entering data into ERPs, assembling client review and fraud-case packs, and building employee training from approved manuals. The work it replaces is manual rekeying, spreadsheet-chasing, and document assembly done by hand.

Cleon builds AI voice agents that handle inbound and outbound phone calls for businesses around the clock. The agents answer questions, book appointments, and take orders, replacing the need for human staff to manage routine call volume. Complex calls are escalated to a human team member.

Clidey builds WhoDB, an open-source database interface that gives engineering and data teams direct visibility into and control over their data without routing through a traditional application layer. It replaces purpose-built internal tooling and heavyweight database clients that teams otherwise use to inspect, query, and manage production data.

Code Four builds AI software for law enforcement that converts body-camera footage, transcripts, and case context into ready-to-file police reports, replacing the manual transcription and writing that consumes a significant share of every officer's shift. Its platform also covers investigative search across video and digital evidence, automated redaction for public release, and mobile field capture. The work it eliminates is largely clerical: documentation tasks that pull officers away from frontline duties and introduce errors when done by hand.

Cohesive builds an AI-driven sales and marketing platform for blue-collar service businesses, including janitorial, HVAC, landscaping, and pest control companies. It identifies local businesses and property owners in a service area, then runs outreach across email, SMS, and phone, with AI agents handling replies, missed calls, and follow-ups. It replaces the manual prospecting and outreach work that field-based operators have no time to do themselves.

Combinely builds an AI coworker, Coco, for US accounting firms. Coco handles the preparation side of accounting work, including document collection, client communication, bookkeeping, payroll, sales tax, and individual and business tax return prep, across the tools firms already use. Accountants review the output rather than doing the groundwork themselves.

Cotool builds an agentic platform for corporate security operations teams, replacing manual investigation, alert triage, and documentation work. It assembles context across a team's existing security stack, lets practitioners define detection logic in natural language, and generates response automations and reports without coding. The work it displaces includes manual alert review, detection rule maintenance, and threat intelligence filtering that keeps teams reactive instead of proactive.

Crimson builds a case intelligence platform for litigation lawyers handling complex disputes. It ingests the full case file to answer factual questions, surface key evidence, and draft legal documents, replacing hours of manual document review and drafting that lawyers previously did by hand. Every output is cited back to source documents, keeping the work auditable.

Cua builds open-source infrastructure for teams developing computer-use AI agents. It provides sandboxed desktop environments, fleet orchestration, SDKs, and benchmarks across Linux, Windows, macOS, and Android, replacing the need to provision and manage real machines for agent training, evaluation, and synthetic data generation. AI teams use it to run parallel rollouts, reproduce failures from snapshots, and measure agent performance against expert-level desktop tasks.

cubic builds an AI code review platform for software engineering teams whose pull request volume, accelerated by AI-generated code, outpaces their reviewers. It automatically scans every PR against the full codebase, catches bugs that cross repository boundaries, and converts senior engineers' past review comments into rules that run on every future PR. The platform also restructures how humans review remaining changes, replacing the default alphabetical file order with a logical sequence and generating architecture diagrams.

Delty builds an AI workforce platform for value-based care organizations, serving health plans, IPAs, ACOs, and medical groups. It deploys role-specific AI copilots across risk adjustment, quality and Stars, pharmacy, provider network, and care management, each drawing from a single shared record ingested from claims, EHR, pharmacy, lab, and ADT feeds. The platform replaces manual gap-closure workflows, BI ticket queues, and meeting-prep work that currently consumes care operations teams.

Den builds a no-code platform that lets business teams create and run AI agents to automate knowledge work across tools like Notion, Slack, Gmail, and Linear. Teams describe tasks in plain language, and agents execute them autonomously, handling recurring workflows or responding to external triggers. Den replaces manual coordination and repetitive white-collar work that would otherwise require custom engineering or human effort.

Docket builds an AI-driven QA testing platform for engineering and QA teams across web, iOS, Android, and desktop. Instead of writing and maintaining brittle script-based tests, teams describe test flows in plain English or record interactions by screen coordinates, and Docket executes and self-heals those tests as the application changes. It replaces manual scripting, selector maintenance, and the QA engineering hours spent keeping test suites current.

Dolphin builds full-stack web applications for startup founders who need a production-ready SaaS product fast. It generates a complete app, including database schemas, user authentication, billing, file storage, and an embedded AI assistant, from a single prompt. Founders get what would otherwise take an engineering team weeks: a correct, secure, and scalable codebase delivered in hours.

Eloquent AI builds AI agents that automate complex, regulated back-office and customer operations for financial services firms, including core banking and insurance workflows. The agents learn by observing existing standard operating procedures and team actions, then take over tasks such as resolving customer inquiries, updating CRMs, and processing transactions across browser and desktop applications without any API integration or engineering work. This replaces manual operations staff handling high-volume, compliance-sensitive processes.

Flott HQ builds an AI-powered fleet management platform for transport and logistics companies, covering route planning, live GPS tracking, fuel monitoring, invoicing, and payments in a single workspace. AI agents monitor connected vehicles continuously, detect anomalies like delays, excess idling, or unusual fuel consumption, and prepare the next action for the operator to approve. This replaces the manual work fleet managers do to spot problems, chase context across tools, and coordinate responses across drivers, dispatchers, and customers.

Foundation Industries machines precision components for frontier hardware companies, delivering parts in as little as two days via 5-axis CNC milling. It targets the prototype and low-volume production stage, where lead times from domestic suppliers typically stretch far beyond what hardware development cycles allow. The service replaces slow domestic machining and the logistical complexity of sourcing parts overseas.

Frekil builds an AI-powered real-world evidence platform for clinical researchers, biostatisticians, and epidemiologists. It connects to EHR systems, claims data, and registries, then automates the full epidemiological workflow, from cohort building and study design to statistical analysis and publication-ready outputs, replacing manual data querying, statistical coding, and report assembly. Users interact via natural language, review causal diagrams, and get auditable traces from raw data to final results.

Galen AI builds a personal health companion app for patients managing chronic or complex conditions. It pulls together medical records from hundreds of healthcare institutions and data from wearable devices into a single view, then uses that context to answer health questions, surface proactive recommendations, and provide on-demand support. This replaces the experience of waiting days for a care team response or settling for generic search results that know nothing about the patient's actual history.

Godela builds an AI physics engine for engineers working on complex physical systems. It trains on a team's existing simulation data to learn the governing behavior of their system, then answers what-if queries and finds optimal configurations in seconds, replacing both computational fluid dynamics solvers and physical prototypes. The target user is an engineer who currently waits days for simulation results or builds hardware to test a design change.

Golf builds a security gateway for enterprise MCP servers, giving security, IT, and platform teams a single control point to discover, monitor, and enforce policy over AI agent connections to production systems. It targets the blind spot created when engineers wire tools like Cursor, Claude Code, and Copilot directly to codebases and databases via MCP, bypassing existing identity, DLP, and SIEM controls. Golf replaces manual, ad-hoc oversight of those connections with real-time visibility, policy enforcement, threat detection, and audit-ready evidence export.

GroundControl builds quality documentation software for manufacturers in regulated industries such as aerospace, defense, and medical. Its platform automates the generation, review, and analysis of inspection reports, including AS9102C first article inspections and PPAP packages, replacing manual data entry from PDF drawings into spreadsheets. Suppliers complete required documentation in minutes rather than days, reducing source rejections and accelerating approvals from OEM customers.

HABIT builds conversational robots for physical retail and consumer-facing businesses. The robots greet customers, answer questions, and upsell products, replacing the front-of-house staff who would otherwise handle those interactions. Operators train the robots the same way they would onboard a human employee.

HelixDB builds a combined graph and vector database that gives AI agents persistent memory, connected knowledge, and grounded retrieval in a single system. It replaces the patchwork of separate vector stores, graph databases, and search indexes that developers otherwise stitch together to support agent workflows. Customers range from dev-tool teams linking code, docs, and incidents to enterprises grounding AI answers in permissioned company records.

Hemut builds back-office software for trucking companies, replacing manual dispatch coordination, paperwork, and accounting work with AI-driven automation. The platform handles inbound and outbound calls via AI agents, extracts data from documents to cut manual entry, sources loads, and automates accounting. It targets trucking operators who currently manage these workflows through a mix of phone calls, spreadsheets, and disconnected tools.

Human Behavior builds an AI-powered session replay and analytics platform for product and engineering teams. It records every user web session and uses vision models to surface issues, behavioral patterns, and funnel breakdowns that standard dashboards miss. Beyond surfacing problems, it acts on them: opening pull requests and contacting affected users directly.

Janus builds simulation-based evaluation infrastructure for AI agents. It runs agents through high-fidelity simulated environments to surface failures in reasoning, compliance, tool use, and performance before deployment. The resulting datasets serve as benchmarks and feed post-training pipelines, replacing manual or ad-hoc evaluation workflows.

Jazzberry builds reinforcement learning environments used to train frontier AI models. The product gives model developers ready-made simulation environments that would otherwise require custom engineering to construct from scratch. It targets AI labs and research teams that need structured training grounds to develop and refine large-scale models.

Jeevy Fabrication acts as a weld fabrication contractor for industrial clients, matching complex weld jobs to vetted shops and managing the full project lifecycle. Its software converts manufacturing drawings into bills of materials and resource-loaded schedules, cutting bid and scheduling time for its shop network and eliminating the errors that drive rework. Clients get a single portal to submit purchase requests, track schedules, pull key documents, and receive real-time photo updates, replacing the fragmented coordination typical of traditional outsourced fabrication.

Kaelio builds an AI-native business intelligence layer that connects directly to a company's data warehouse, letting business teams explore data and build dashboards through conversation. It replaces rigid tools like Metabase and Looker, which require tickets to data or engineering teams for every new question or dashboard change. The core product, ktx, auto-generates a governed context layer of business definitions and metric logic that data agents and analysts query against, producing consistent, accurate SQL without manual configuration.

Kaizen builds browser-based automation for healthcare clinics, handling the administrative work between patient referral and payment. It logs into payer portals directly, without APIs, to verify insurance eligibility, submit prior authorizations, process referrals, and enroll providers. This replaces the manual back-office labor that revenue cycle and operations staff currently perform across fragmented web portals.

Kanava AI builds voice AI agents for wholesale distributors, handling the inbound sales work that human reps currently do manually: stock and price checks, order calls, field note logging, and CRM updates pulled from live ERP data. As the system trains on real rep workflows, it takes on fuller sales calls, surfaces upsell opportunities, and reactivates dormant accounts. Distributors use it to grow sales volume without adding headcount.

Kashikoi builds a simulation engine that benchmarks AI agents for B2B software teams. It generates world models that autonomously conduct interviews with agents and produce detailed behavioral assessments across accuracy, context understanding, and edge-case handling. This replaces manual evaluation workflows, letting teams test and ship agents without writing evaluation code or monitoring runs by hand.

Kestral builds an agentic product development platform for B2B software teams. It ingests customer signals, such as sales call transcripts and feature requests, then surfaces prioritized opportunities, links them to an existing roadmap, and automatically follows up with customers when their requests ship. This replaces the manual work of synthesizing customer feedback, grooming backlogs, and coordinating task updates across product, sales, and engineering.

Kirana AI builds an AI store manager for physical grocery retailers, starting with an on-premise GPU that monitors stores around the clock to detect theft, flag workplace safety issues, and improve customer service. Planned integrations with POS and ordering data will extend the system to automate inventory ordering, pricing, assortment decisions, and task delegation, replacing the day-to-day operational work currently handled by human store managers.

Klavis AI builds an open-source MCP integration platform that lets AI agents call external tools reliably at scale. It provides hosted, secure MCP servers via API, removing the need for teams to manage authentication or write client-side integration code. The platform also produces high-quality coding and agentic tool-use training data, including long-horizon workflows across live SaaS apps, programmatic verification, and Dockerized environments, replacing brittle local setups and manual data curation for AI model post-training.

Labric builds data infrastructure for research scientists, automatically ingesting raw instrument outputs, structuring them with full experimental context, and connecting them to AI workflows. It replaces manual data exports, scattered file transfers, and the institutional knowledge lost when team members leave. The result is a single, queryable source of truth that lets researchers run AI-powered analysis on their own lab data without first cleaning it by hand.

Lapel builds a customer operations platform for B2B teams, unifying account data from across systems into a real-time view and coordinating action across sales, success, and support. It replaces the patchwork of dashboards, funnels, and templated replies that teams currently use to manage customer relationships. The goal is to let online businesses deliver the kind of attentive, context-aware service that was previously only practical face-to-face.

Leeroo builds an AI platform that ingests every system an enterprise operates, constructs a continuously updated model of the business, and deploys agents that autonomously handle data engineering, opportunity discovery, and ML engineering tasks. It targets enterprises that today rely on teams of data engineers, data scientists, and ML engineers to move from raw data to production use cases one at a time. Leeroo's platform, Uniforge, collapses that multi-team chain into a single system that ships agents and their operational interfaces on the customer's existing stack.

LineWise captures how physical products are designed, machined, and assembled on real factory floors, then converts that video and process data into structured training data for AI and robotics models. It records design decisions, CNC machining, and hands-on assembly across live production environments, replacing the manual and largely absent documentation of manufacturing know-how. The target customer is the physical AI and robotics industry, which lacks the real-world, in-the-wild data that language models drew from the web.

Lucis builds a preventive health membership for European consumers who want to go beyond standard blood panels. Members get twice-yearly lab tests covering 110+ biomarkers across 11 body systems, combined with wearable data and a health questionnaire, then receive personalised recommendations on what to change. The product replaces the fragmented loop of generic GP blood tests and self-interpreted results with a single, ongoing health companion.

Lumari builds AI agents for direct procurement teams in manufacturing and related industries, covering supplier collaboration, RFQ and purchase order management, and order monitoring. The agents handle the day-to-day operational work, including supplier emails, quote analysis, ERP updates, and shipment booking, that procurement staff currently do manually across the sourcing-to-delivery cycle.

Lyra builds an AI-native meeting platform for go-to-market teams, replacing the combination of separate video tools, notetakers, and CRM manual entry. Meetings run natively in the browser with per-participant transcription, and Lyra handles prep research, real-time document collaboration, and post-call actions like CRM updates and follow-up emails from a single workspace.

Mbodi AI builds software that lets industrial operators teach robots new tasks through plain-language instructions and brief physical demonstrations, replacing the traditional cycle of specialist reprogramming and production downtime. The platform uses generative AI and agent orchestration to translate those instructions into reliable robot actions that run consistently on the factory floor. When one robot in a fleet learns a skill, that knowledge propagates to all others, so manufacturers can adapt to new product lines or process changes in minutes rather than months.

Mesmer gives founders and CTOs a live view of their engineering organization's health by reading code, tickets, and workflows to pinpoint where development slows down. Each month it identifies the top bottleneck, implements fixes directly in the codebase, and then remeasures, replacing the manual work of status meetings, update-chasing, and ad hoc process debugging. The product targets technical leaders who need engineering visibility and continuous improvement without dedicating headcount to it.

MindFort builds autonomous AI agents that conduct web application penetration testing, replacing human pen testers and legacy scanning tools. The agents attack live applications through their real interfaces, confirm exploitability with reproduction evidence, and generate patches, working at the speed and scale that AI-generated code demands. Customers include enterprise security teams, startups facing security reviews, and organizations maintaining aging applications across regulated industries such as FinTech, healthcare, and manufacturing.

Minerva runs an AI-native accounting firm where human accountants work alongside AI agents to deliver financial advisory and bookkeeping services to businesses. The model replaces the slow, manual workflows of traditional accounting firms, closing books significantly faster. Minerva's pitch is that it makes professional financial services faster and more affordable without sacrificing the human judgment clients expect.

mlop builds an open-source MLOps platform that tracks and optimizes machine learning experiments, targeting ML engineers and data science teams. It covers experiment logging, parameter and gradient tracking, model versioning, git status, and real-time performance alerts, replacing proprietary tools like Weights & Biases. The API is fully compatible with Weights & Biases, so teams can migrate without rewriting instrumentation code.

Moby Analytics builds an AI platform for financial auditors, operating as an Excel add-in. Auditors use it to automate audit testing, general ledger analysis, and statement tie-outs: the product fills workpapers, links every figure to its source document, and flags variances, replacing manual data gathering and cross-referencing done by audit teams.

Morphik builds an open-source multimodal search platform for AI applications and a separate AI worker product for skilled nursing and senior living operators. The AI worker product handles accounts payable, billing, collections, and care management workflows, replacing manual back-office work done by facility staff. Customers range from space-tech research teams and developer-built brokerage agents on the search platform to multi-facility senior living operators automating their procure-to-pay cycles.

MorphoAI builds an AI-powered design platform for mechanical engineers creating robots and machines. Engineers input parts, constraints, and target tasks, and the platform evaluates hundreds of design options to produce and fine-tune a machine design in minutes rather than years. It replaces the manual, iterative design work that slows hardware R&D teams at OEMs and system integrators from bringing products to market.

nao Labs builds an open-source analytics agent builder for data teams. It lets data engineers define a structured context layer, connecting a warehouse, dbt repos, docs, and business definitions, so an AI agent can answer SQL questions reliably against real production data. It replaces ad-hoc analyst queries and the manual work of wiring LLMs to data sources.

neoncoral builds an AI-native consumer social product. The company operates in stealth, so the specific feature set, target user, and workflow it replaces are not yet public.

Nimbic AI builds an AI agent that generates and maintains internal code documentation for software development teams. It connects to a GitHub repository, scans the codebase, and produces structured docs in a separate repository, updating them automatically on every commit. It replaces the manual work of writing, organizing, and keeping technical documentation current as codebases change.

Nomi builds a real-time AI copilot for live sales calls, serving sales teams from early-stage founders to mature revenue operations. During a call on Zoom, Meet, or Teams, Nomi listens, diagnoses prospect signals, and surfaces specific responses, competitive angles, and next-best moves on screen, replacing the mental load of recalling playbooks, battle cards, and objection-handling scripts mid-conversation. It also analyzes top-performer patterns across recorded calls and automatically updates the team playbook to spread winning behaviors across every rep.

Notus Autonomous Systems builds autonomous drone and ground robot systems for military use, designed to operate in coordinated swarms under AI-driven command. Commanders issue high-level mission objectives rather than piloting individual units, replacing skilled human operators for each vehicle. The system compresses tactical response times by shifting the command model from manual control to fleet-level, software-directed autonomy.

Novoflow builds computer-using AI agents that operate electronic health record systems on behalf of medical clinics and hospitals. The agents read screens, click, and type inside existing EHR software to handle scheduling, billing, and claims work that staff would otherwise do manually. Deployment requires only a single staff login and no changes to how the clinic's team works.

Opine builds AI chatbots trained on a specific person's content so that anyone can have a conversation with that person's expressed views and opinions. It replaces the need to read through a creator's or expert's back catalogue to understand their stance on a topic. The product targets consumers who want quick, direct access to someone's perspective without the person being present.

Opusense AI builds an AI-native field operations platform for construction inspectors, engineers, architects, and facility managers. Inspectors capture voice notes and photos on a job walk, and the platform converts that raw input into structured, branded reports in real time, replacing manual write-ups that previously consumed a significant share of the work week. The product serves three distinct teams: field review firms, general contractors, and owners or property managers, each with a tailored workspace for managing documentation and sharing project information across stakeholders.

ParaQuery runs large-scale Spark and SQL workloads on GPU-accelerated, fully managed infrastructure, delivering twice the throughput at roughly half the compute cost of platforms like Databricks and BigQuery. It targets data engineering teams spending too much time managing pipelines and too much money on cloud compute. Teams point ParaQuery at their existing storage and run ETL or analytics jobs without migrating data, rewriting logic, or reconfiguring tools.

Parsewise builds an AI platform for insurance and reinsurance teams that extracts, resolves, and validates data from document packages, emails, spreadsheets, and other sources into a shared data layer covering party, policy, and claim records. Its AI assistant, Navi, prepares submissions, compares quotes, checks policies, and analyses portfolios, tracing each finding back to its source document. The product replaces manual document review and data gathering that underwriting and investment teams would otherwise do by hand.

Partcl builds GPU-accelerated electronic design automation tools for chip designers. Its physics-informed models replace legacy EDA solvers, compressing simulation and optimization work that previously took weeks into minutes. The tools target hardware engineers who need faster design iteration without sacrificing physical accuracy.

Percival builds an AI copilot for researchers who need to analyze and transform real-world data. It targets research and data science workflows, replacing manual data wrangling and analysis work that researchers currently do by hand or with ad-hoc scripts. The product sits at the intersection of developer tooling and applied AI, accelerating the path from raw data to insight.

PgDog builds a horizontal scaling layer for PostgreSQL that handles sharding, connection pooling, and load balancing in a single executable. It sits between an application and managed Postgres databases like AWS RDS, routing and distributing queries automatically across shards without requiring changes to application code or schema. It replaces tools like PgBouncer and RDS Proxy while adding distributed query execution that those tools do not provide.

Photonium builds software and provides engineering consulting that automates optical system design for hardware teams in quantum computing, automotive, aerospace, life sciences, AR/VR, and semiconductor manufacturing. Its Insight product takes a written spec or existing design file and runs optimization, analysis, and tolerancing inside Zemax, CODE V, or its own ray tracer, replacing the manual iteration that occupies optical engineers. A consulting arm pairs clients with senior optical engineers supported by the same AI tooling.

Plexe builds predictive machine learning models for enterprise teams, taking a plain-language problem description as input. Its agents connect to a customer's data sources, run experiments, evaluate candidates, and deploy a finished model to an API endpoint, replacing the work a data science team would otherwise spend quarters doing by hand.

PowerMatrix builds high-efficiency, compact power supply systems for AI hardware, including GPUs, xPUs, AI servers, and data centers, as well as drones, electric vehicles, and aerospace applications. Its designs use a patented circuit topology, embedded packaging, and novel magnetics to cut energy loss in half compared to conventional power supplies. The product targets the growing demand for dense, power-hungry compute infrastructure where thermal and efficiency constraints make standard power delivery inadequate.

Prism builds an AI media studio that generates short-form video ads and branded image content for marketing teams. It consolidates dozens of AI video and image models into a single platform, replacing traditional production workflows that required studios, shoots, and manual editing. Marketing, social media, and real estate teams use it to go from a creative brief to launch-ready ads, product visuals, and UGC-style clips.

Probo handles end-to-end compliance for B2B software teams, replacing the need for a dedicated compliance hire. It collects requirements, manages documentation, risks, controls, and evidence, runs audits, and keeps everything current afterward. The work it replaces spans weeks of founder and engineering time spent on frameworks like SOC 2 and ISO 27001.

Propolis builds autonomous browser agents that run quality assurance testing for software teams. Swarms of agents simulate real user behavior across web applications, replacing manual QA work and human-in-the-loop test execution. The product targets B2B software teams that need continuous, hands-off coverage without dedicated QA staff running tests themselves.

QFEX operates a 24/7 exchange for US equities, commodities, and FX pairs, targeting retail and professional traders who are priced out or locked out by traditional brokers and exchange hours. The platform offers direct market access with leverage up to 100x through perpetual futures across all listed asset classes, replacing the broker layer and the constraint of market-hours-only trading that defines incumbent exchanges.

QualGent builds an AI-powered quality assurance tester for mobile apps, running automated tests that simulate real human behavior to catch bugs before release. Mobile development teams use it to scale QA coverage without adding headcount, replacing manual test cycles that cannot keep pace with AI-assisted development. The product closes a feedback loop between bug discovery, fixes, and test coverage so each release improves on the last.

Ralo builds an automated mortgage platform for homebuyers and refinancers, replacing the traditional broker with software that shops multiple lenders, reads every quote for hidden fees, and ranks loans by true cost. It handles the paperwork, pricing, and negotiation that brokers typically charge for, passing the savings to the borrower.

Refresh builds simulation environments and evaluation benchmarks for AI agents doing real computer work, including navigating enterprise software, writing code, and operating 3D animation tools like Blender. Its customers are AI labs and enterprises that need to benchmark and post-train frontier models on economically valuable tasks. The environments replace hand-crafted, manually scored test suites with high-fidelity software worlds that generate verifiable, automatically scored rewards for reinforcement learning and supervised fine-tuning.

Relixir builds an autonomous generative engine optimization (GEO) platform for B2B companies that want their brand cited when buyers search on ChatGPT, Perplexity, Claude, and Google AI Overviews. Its AI agent, Rex, produces and refreshes structured content, builds domain authority through automated backlink outreach, and tracks which prospects convert after encountering a brand mention in an AI response. It replaces the manual work of SEO content teams and traditional search optimization workflows.

Rimba builds environmental compliance software for industrial and energy companies, starting with low carbon fuels. The platform pulls data from ERPs, SCADA systems, and document stores to calculate carbon intensity, track fuel chain of custody, and generate regulatory reports across programs like LCFS, RFS, 45Z, and CORSIA. It replaces manual data gathering and spreadsheet-based emissions accounting that legacy tools cannot handle at scale.

RunRL delivers reinforcement learning as a service, helping AI teams improve the performance of their LLMs and agents on specific tasks. Customers supply a model, a prompt, and a reward function; RunRL handles the training to push that reward higher. The service replaces manual prompt engineering and the guesswork of repurposing observability data for model improvement, and has been applied to domains including interpretability, formal verification, drug design, and automated red-teaming.

Saffron Health builds AI agents that handle specialist referrals for primary care practices. The agents identify available in-network providers, complete prior authorization forms, retrieve medical records, and help patients book appointments. This replaces hours of daily manual coordination work performed by primary care staff.

Scalar Field builds an AI-powered trading terminal for market participants who need to test and act on market hypotheses. Instead of static dashboards, it gives users deployable agents that monitor market events, run backtests across equities, options, prediction markets, and pre-IPO names, and execute trades when set conditions are met. It replaces the manual, fragmented workflow of hypothesis validation that conventional terminals leave to the trader.

sieve cleans and structures financial data for hedge funds and investment firms, replacing manual engineer review loops with an API that plugs directly into existing data pipelines. It combines AI extraction with human expert review to produce validated, structured data from unstructured sources such as websites, PDFs, and filings. The product is also accessible via Excel for teams that work outside code.

Sim builds an open-source AI workspace that lets development teams design, deploy, and govern AI agent workflows. Teams connect to over a thousand integrations and any major LLM, then build agents visually, through chat, or in code. Sim replaces ad-hoc agent tooling and scattered oversight by centralizing control over access, spend, data, and deployment in one platform.

Sixtyfour builds an identity intelligence platform for trust, risk, and investigation teams. Starting from a single identifier, such as an email, phone number, or company name, its AI agents browse public records, official filings, and the open and dark web to produce a fully cited intelligence profile. It replaces manual research workflows used in contractor vetting, fraud investigations, financial compliance, and child safety operations.

StarSling builds an agent-native CI platform that runs and continuously optimizes GitHub Actions workflows for AI-first development teams. Its agents analyze pipelines, cut queue times and test durations, and open pull requests with concrete workflow improvements, replacing the manual tuning and infrastructure management developers would otherwise do themselves. The product also serves as a developer homepage where agents surface and auto-fix issues from GitHub, Linear, and Sentry without requiring the developer to leave the interface.

stratify builds an AI-powered user research platform for product, UX, and marketing teams. It automates the full research workflow: recruiting participants, running video interviews, analyzing responses, and delivering insights. The platform replaces manual research operations that traditionally take weeks, returning findings in hours.

Sureform builds post-training datasets and reinforcement learning environments for frontier AI developers, drawing on real enterprise workflows to train multimodal and embodied AI systems. The data captures diverse human interactions across varied settings, targeting the specific gap between generic training data and the grounded, real-world inputs that world models and physical AI require.

Sygaldry Technologies builds quantum-accelerated AI servers designed to speed up model training and inference for organizations developing AI. The hardware combines multiple qubit modalities in a fault-tolerant architecture, replacing conventional compute infrastructure that makes large model development and deployment prohibitively expensive. Sygaldry also produces tooling that lets AI researchers incorporate quantum capabilities directly into their models and workflows.

SynthioLabs builds a voice AI platform for pharmaceutical and life sciences commercial teams, automating engagement with healthcare professionals and patients around the clock. Its suite of specialized agents handles HCP outreach, patient support, field team assistance, and on-demand market research simulations. The platform replaces manual field rep calls, live call center interactions, and ad hoc research workflows that pharma commercial operations currently run through human labor.

Tegore builds an AI math tutor, Noor, for K-12 students covering grade 6-12 curriculum. Each lesson targets a specific topic, such as polynomial long division, and delivers content through voice and interactive visuals while using dialectic teaching to check comprehension. The product replaces one-on-one human tutoring and supplements classroom instruction, giving students a structured, curriculum-aligned practice session on demand.

Text.ai builds Alfi, an AI agent that lives inside group chats on SMS, WhatsApp, and iMessage. Alfi learns each member's preferences, humor, and history to help the group plan events, find places to eat, set reminders, and create images, replacing the back-and-forth coordination that normally falls on individual people. It requires no download or setup, operating as a persistent presence in existing conversations.

The LLM Data Company researches post-training data for frontier AI models, producing bespoke tasks, reward signals, and training environments at scale. It works directly with frontier AI teams, replacing ad-hoc internal data curation with purpose-built pipelines tailored to specific model behaviors. Its work feeds the reinforcement and fine-tuning stages that shape how large models reason and act.

The Robot Learning Company builds a general-purpose robot platform that automates repetitive, stationary physical tasks for businesses that cannot justify the cost of conventional industrial automation. The platform targets work that today relies on human operators performing the same manual motions at a fixed station. By keeping the hardware affordable, the company opens robotic automation to smaller or cost-sensitive operations that have historically been priced out of the market.

Theorem builds AI-accelerated program verification tools for engineers working on high-stakes code. Its models cut the cost of producing formal proofs, letting developers verify correctness and security across GPU-accelerated code, cryptography implementations, and legacy systems without requiring specialized verification expertise. The work it replaces is manual auditing and ad-hoc testing, which leave entire classes of bugs and exploits undetected.

Theta builds reinforcement learning environments that train AI agents to perform domain-specific knowledge work, replacing human labor on tasks like financial modeling, contract redlining, patient documentation, and IT resolution. The platform converts a business's first-party data and expert workflows into simulation environments where agents learn the tools, constraints, and multi-step processes of that specific operation. Frontier AI labs and enterprises use Theta to deploy agent workforces that can execute long-horizon tasks with the depth of trained domain specialists.

Third Chair automates music IP enforcement and licensing workflows for media and entertainment companies, combining software with an on-demand network of lawyers. It monitors content use across platforms, converts unauthorized ads into licensing revenue, and accelerates sync deal flow. The product replaces manual processes and outsourced data labelers that deliver raw CSV files with no actionable enforcement layer.

throxy runs fully managed outbound sales for B2B companies selling into traditional industries such as manufacturing, logistics, construction, and finance. It combines AI agents with human operators to find sector-specific buyers, deliver personalized outreach at scale, and book qualified meetings directly onto client calendars. The product replaces in-house SDR teams and generic outbound tooling that was built for tech-native buyers and fails to reach offline-heavy industries.

Ticket Wallet automates ticket resale for season ticket holders and brokers, replacing manual workflows built around spreadsheets, single-site listings, and outsourced pricing teams. Sellers upload inventory once; the platform sets prices using live demand data, pushes listings to eight or more marketplaces simultaneously, and reprices continuously until each ticket sells.

Tinfoil builds a confidential computing platform that lets companies run AI workloads on cloud infrastructure with cryptographically verifiable data privacy. It targets businesses handling sensitive data that need guarantees beyond contractual promises, replacing on-premises deployments, PII redaction hacks, and unenforceable "trust us" policies from cloud AI providers. The platform runs on NVIDIA Hopper and Blackwell GPUs in confidential compute mode, so neither Tinfoil nor the underlying cloud provider can access user data.

Trapeze builds AI voice agents for medical practices that handle inbound patient calls around the clock, replacing human receptionists and call center staff. The agents are modeled on a practice's existing front-desk workflows, so doctors can book appointments and answer patient questions outside of office hours without adding headcount. The product targets independent physicians and small practices looking to grow their patient volume without scaling their administrative staff.

TypeOS builds an AI writing platform that generates text in the user's personal voice, for individuals and enterprise teams who want AI output that reads as authentically their own. On the integrity side, it also gives teachers, recruiters, and security teams tools to detect AI-generated content, flag hallucinated citations, check plagiarism, and verify authorship. The product replaces manual style editing of AI drafts and the patchwork of separate detection tools.

ValueMate builds AI-powered appraisal software for valuation firms, banks, appraisal management companies, and independent appraisers. Appraisers use a LiDAR-enabled mobile app to capture a property in a short walkthrough, and the platform automatically assembles the inspection data, MLS records, comparable sales, and sketch into a regulator-ready UAD 3.6 report, replacing manual measurement, data entry, and form-filling.

Vassar Robotics builds gearbox-free robotic arms for robot learning, replacing conventional gear-driven joints with transverse-flux direct-drive actuators. The design eliminates backlash, stiction, speed caps, and wear drift from every joint, giving robotics developers cleaner demonstration data and actuators whose behavior a simulator can accurately model. The arms are aimed at researchers and teams training home robot policies, removing the need to work around gearbox artifacts in collected data and sim-to-real transfer.

Vesence builds an agent-native workspace for law firms, banks, investment firms, and consulting teams, embedding AI agents directly into Word, Excel, PowerPoint, and Outlook. Lawyers use it to draft and review contracts, run due diligence, and prepare transaction documents, work that today is done manually inside plain Microsoft Office with no intelligent assistance. The product replaces that unassisted drafting and review process by letting agents do the execution while the professional reviews and decides.

VoiceOS builds a voice operating system for Mac and Windows that lets users control apps, dictate text, and execute multi-step tasks by speaking. It replaces manual typing and clicking across every app, handling actions like sending emails, scheduling meetings, finding files, and setting reminders through a single voice command. The cursor acts as context, so users point at anything on screen and speak rather than switching apps or explaining the situation.

Vybe builds AI agents that live inside Slack, giving every team in a company a shared, collaborative AI coworker rather than a private chatbot. Each agent carries its own memory, permissions, and integrations, replacing the scattered mix of manual workflows, rigid SaaS tools, and one-off AI chat sessions that currently fragment company operations. The platform is SOC 2 Type II certified and supports SAML SSO, targeting B2B companies that need AI deployed securely across departments.

Wavedash delivers browser-based gaming that lets players launch titles instantly, with no download or installation required. It targets mainstream gamers who want to jump into games, including multiplayer sessions, without the friction of local setup. The platform replaces the traditional install-and-patch workflow with cloud-streamed play accessible from any browser.

Willow builds an AI voice dictation app for professionals who write heavily across email, documents, and AI prompting tools. Users speak naturally into any application and Willow converts speech to polished, edit-ready text, replacing keyboard typing as the primary input method. It targets knowledge workers, lawyers, healthcare professionals, developers, and sales teams who spend significant time writing.

YouLearn builds an AI tutoring platform for college students that ingests their own course materials, including PDFs, YouTube videos, lecture slides, and recorded lectures, and converts them into summaries, quizzes, and a conversational tutor. Students use it in place of manual note-taking, self-study review sessions, and generic tutoring services. The product adapts directly to each student's uploaded content rather than relying on generic curricula.

Zeon Systems builds AI-powered robotic systems that run complete experiments in scientific labs. Scientists describe a protocol in plain English, and Zeon translates it into a workflow, detects the lab's existing equipment, and executes the full procedure autonomously using robotic arms, including steps like liquid handling, incubation, and sample disposal. It replaces the manual bench work scientists currently perform themselves, freeing them to focus on analysis and design.

Zero builds an open-source AI email client for knowledge workers who spend significant time triaging and acting on email. It automatically summarizes incoming messages, surfaces tasks, and lets users interact with their inbox through a chat interface, replacing manual sorting, reading, and follow-up triage.