YC Summer 2025 (S25) companies

Acrely builds enterprise voice agents that handle customer conversations across customer service, sales, and operations. The platform, sold under the name Lupitor, lets large enterprises build and deploy AI agents that take over inbound and outbound phone interactions previously handled by human agents or basic IVR systems. It runs in both cloud and on-premises environments, targeting industries such as banking, telecoms, and insurance.

Adaptional builds an AI platform for insurance carriers that automates underwriting and claims quality review. Its Underwriter AI pulls data from submission files, cross-checks it against public sources, and summarizes risk against a carrier's own guidelines, replacing manual pre-processing by underwriters. On the claims side, it continuously audits every file against configurable scorecards, surfacing compliance gaps, reserve issues, and leakage in real time rather than through the monthly spot-checks that traditional QA teams run on a fraction of open claims.

AgentHub builds a simulation and evaluation engine for AI agents, letting engineering teams catch and fix agent errors before they reach production. Teams run agents through curated reinforcement learning environments, trace behavior, generate annotated datasets, and compare prompts, models, and configurations in a single interface. The product replaces ad-hoc manual testing and scattered evaluation tooling with a structured feedback loop for iterative agent improvement.

AgentMail builds an email inbox API that gives AI agents their own dedicated email addresses and inboxes. It targets developers building agentic systems that need to send, receive, and manage email as part of automated workflows. The product replaces custom email infrastructure that developers would otherwise have to wire together themselves.

Alara builds a procurement platform for dental offices that consolidates supply purchasing across multiple vendors into a single workflow. It pulls live pricing from a range of suppliers so dentists can compare costs and place a combined order in one cart. This replaces the manual process of checking each supplier separately and managing multiple purchase orders.

Albacore Inc. builds mass-producible, long-range drone submarines for military operators conducting undersea deterrence and sensing missions. Its first vehicle, Ghostfin, is a 6-foot, 425-pound autonomous submarine that can be launched from land within minutes, reach targets up to 1,000 nautical miles away, and carry precision strike or sensor payloads across configurable mission modules. The system replaces crewed or shorter-range undersea platforms in littoral and open-ocean conflict scenarios.

Alter builds a privileged access management platform for AI agents, serving security and engineering teams that run agentic workflows against production systems. Every tool call an agent makes is authenticated, checked against fine-grained policies, and executed with a short-lived, scoped credential that expires in seconds, replacing the standing secrets and manual access reviews that teams currently rely on. Unsafe actions are blocked before they reach production, and a tamper-evident audit log covers every request for SOC 2, HIPAA, and GDPR reporting.

Altur builds voice AI agents that run outbound debt collection and sales calls for B2B clients in Latin America, replacing human call center operators. The agents handle the full collection workflow: segmenting accounts, placing calls through a proprietary dialer, negotiating payment plans in real time, and following up via WhatsApp and SMS. Companies that previously staffed call centers to chase overdue balances hand that operation to Altur's agents instead.

Ambral builds AI account managers for B2B companies, aggregating signals across customer interactions into a live model of each account. It identifies which accounts need attention and acts on that, drafting outreach, flagging upsell opportunities, and stepping in before customers churn. The product replaces the manual monitoring and follow-up work that human account managers can no longer sustain as a customer base grows.

April builds a voice AI executive assistant that handles email and calendar management for busy professionals who lack dedicated human support staff. Users interact hands-free, letting April take on the scheduling and inbox work that a human executive assistant would otherwise own.

Astor gives retail investors an SEC-registered AI advisor that connects to their brokerage accounts, analyzes their holdings, and answers investment questions around the clock. It targets busy professionals who trade actively but lack confidence in their decisions, replacing the ad-hoc research, news-scanning, and guesswork that currently fills that gap. The AI surfaces tailored opportunities and filters market news to only what is relevant to a user's actual portfolio.

Autosana builds an AI-powered end-to-end testing layer for mobile and web apps, serving engineering teams that ship iOS, Android, and web products. Engineers describe test coverage in natural language, and Autosana's agents execute and maintain the tests across CI pipelines, replacing the manual work of writing, running, and updating UX test scripts. It catches bugs before they reach production without requiring engineers to maintain brittle automation code.

Autumn builds open-source billing infrastructure for AI startups, sitting on top of Stripe to handle pricing models like usage-based billing, credits, subscriptions, and rollovers. It replaces custom billing code by moving pricing logic into configuration, letting teams change plans, run migrations, and manage entitlements without engineering work. The target customer is a high-growth AI or usage-based SaaS company that would otherwise wire this together manually.

Avelis Health applies machine learning and clinician review to scrutinize medical claims on behalf of self-insured employers and health plans, catching billing, coding, and contract errors that legacy rule-based audit tools and manual reviewers miss. The platform screens every claim, not just high-dollar outliers, before and after payment. It replaces slow, partial manual audits and static rules engines that leave most erroneous payments undetected.

Avent builds AI agents that handle quoting and order entry for manufacturers and industrial distributors. The agents process inbound RFQs, match products, apply pricing rules drawn from past deals and institutional knowledge, and update records directly in the company's existing ERP, replacing the manual sales inbox and data-entry work that sales teams currently do by hand.

b12 Labs builds an AI copilot that automates synthesis planning and condition optimization for medicinal chemists at pharma and biotech companies. Its tools, Selenium and Palladium, take a target molecule and generate ranked retrosynthesis routes and optimized reaction plates, replacing the manual iteration cycles that typically consume years of early-stage drug discovery. The platform compresses what had required eight to twelve experimental rounds into a single, grounded experimental plan.

bitrig lets anyone build, test, and deploy native Swift iOS apps directly from an iPhone using natural-language prompts, replacing the need for a Mac, Xcode, or coding experience. It targets creators and entrepreneurs who want to ship mobile apps without a traditional development setup. Where other AI coding tools generate web-based or cross-platform output, bitrig produces the same native Swift code used by top App Store apps.

Bizzy AI builds an outbound calling agent, Otto, for banks and fintechs that negotiates recurring bill reductions directly with service providers on behalf of customers. Otto handles provider phone trees, account verification, and retention conversations, then returns a structured result with transcript evidence to the financial platform. This replaces manual call center operations and removes the phone call barrier that prevents financial platforms from acting on identified savings opportunities.

RNA-based AI for better drugs and smarter clinical trials.

Blue builds a voice assistant for iOS and Android that operates directly inside any app on the phone: tapping, typing, and swiping on the user's behalf without requiring app-specific integrations. Consumers use it to clear inboxes, reply to messages, pay bills, attach files, and manage calendars entirely hands-free. On iPhone, a small USB-C hardware accessory gives Blue precise input control, replacing the manual tap-through work users would otherwise do themselves.

BootLoop builds an AI agent for firmware development and testing on embedded hardware. The agent ingests datasheets, schematics, and design files to write firmware, then compiles, flashes, and validates it on real hardware using hardware-in-the-loop test benches, replacing the manual work that typically takes firmware engineers days or weeks. It targets hardware engineering teams shipping complex embedded products.

burnt builds an agentic operating system for food supply chain companies, targeting distributors that run on legacy ERPs they cannot easily replace. Its AI agent, Ozai, takes over order management end-to-end: reading incoming orders across email, phone, WhatsApp, fax, and other channels, learning each customer's preferences and shorthand, and entering orders directly into the existing ERP, handling clarifications with customers autonomously. Additional agents cover procurement, credit control, and sales workflows, replacing manual back-office work across departments without requiring a full system migration.

Cactus builds an open-source inference engine and foundation models for running AI directly on consumer devices, from phones and wearables to microcontrollers and AR glasses. Developers embed it in products that need on-device AI without a cloud dependency. It replaces cloud-routed inference for applications where connectivity, latency, or footprint make a remote model impractical.

Candytrail builds AI agents for in-person sales teams in pharma, real estate, and retail. Its platform records customer interactions, scores calls, and surfaces the talk tracks, objections, and revenue drivers that separate winning reps from losing ones. It replaces ad-hoc sales manager observation and quarterly review cycles with automated, timestamped coaching delivered daily.

CareSwift builds an AI writing assistant for EMS crews that drafts compliant patient care narratives during or after each call. Medics enter a call description, answer a short set of clarifying questions surfaced by CareSwift, then review and copy the finished narrative into their existing ePCR system. The tool replaces manual free-text documentation and the back-and-forth required to fix errors that trigger insurance claim denials.

Certus AI builds an AI voice agent that handles restaurant phone lines end to end, around the clock. It takes orders, books reservations, processes delivery requests, upsells, and manages complaints in multiple languages, integrating directly with POS and delivery platforms including Toast, Square, Clover, UberEats, and DoorDash. It replaces the combination of front-of-house staff answering phones, IVR systems, and third-party call centers.

Channel3 builds a universal product graph that lets AI agents search, recommend, and buy products across the web. It serves developers, brands, and enterprises who need shopping infrastructure without scraping or licensing data from closed sources like Google or OpenAI. The product replaces custom data pipelines and fragmented affiliate integrations by providing a single API covering a broad product catalog with cross-merchant matching and variant recognition.

Clodo builds a prompt-driven people-sourcing platform for go-to-market teams, recruiters, and expert-sourcing operations. Users describe a target persona in plain language, and Clodo finds, enriches, and sequences outreach to matching contacts, replacing manual list-building and multi-tool prospecting workflows.

Closera builds AI employees for commercial real estate, replacing the manual work that leasing, brokerage, and property teams handle today. The program places Closera in B2B commercial real estate, while the company's website shows a separate AI personal shopping agent product for e-commerce. The profile is based on the program description: Closera deploys AI agents that carry out commercial real estate work otherwise done by human staff.

cocreate builds automated ingest and organization software for professional video editorial teams. Before an editor touches a timeline, assistant editors spend days offloading cards, syncing audio, grouping camera setups, and structuring projects by hand. cocreate runs that entire process automatically at the point of media backup, delivering a fully transcribed, searchable, NLE-ready project in Premiere, Avid, Final Cut, or DaVinci Resolve the moment footage arrives.

Comena builds AI agents that handle order and quote entry for industrial distributors and manufacturers. The agents read incoming purchase orders, match line items to the correct catalog articles, and push the completed entries directly into the ERP, replacing the manual inbox-to-system work that inside sales and customer service teams currently do by hand.

ContextFort gives security teams OS-level visibility and controls over AI coding agents, such as Cursor and Claude Code, running on engineers' machines. It monitors at the kernel level using eBPF, macOS Endpoint Security Framework, and Windows ETW, independently of the agent itself, to log every file touched, network connection made, and process spawned. It replaces reliance on agent-reported logs and traditional EDR tools, which have no concept of AI-generated actions and produce no audit trail for what an agent accessed.

Convexia builds an AI-driven drug sourcing and diligence platform for pharma and biotech buyers evaluating preclinical and abandoned drug assets. A pipeline of specialized agents handles discovery, in silico scientific evaluation, commercial analysis, and clinical risk scoring, replacing months of manual work by internal teams or consultants. Human experts, including PhDs and key opinion leaders, validate each stage before a go/no-go decision is made.

Cyberdesk builds a computer-use agent for Windows that lets developers automate workflows inside EHRs, ERPs, and other legacy desktop software. Developers describe a task in plain English, trigger it via API, and the agent executes it, replacing teams of workers who would otherwise click and type through desktop apps manually. The agent caches repeated tasks to run them faster and deterministically, while still handling unexpected screen changes mid-run.

datafruit builds an AI workspace for software implementation teams, specifically system integrators and services firms handling enterprise deployments. It captures discovery calls, scoping decisions, and handoff notes into a structured, traceable record, replacing the scattered docs, emails, and tribal knowledge that cause estimation errors and project misalignment.

Dedalus Labs builds persistent virtual machines for AI agents, giving developers full Linux environments that start in under 50ms with filesystem and memory preserved across sessions. It targets developers who need to run agents in production without managing infrastructure, replacing manual container setup, cold-start penalties, and custom autoscaling work. The platform provides VM-level isolation stronger than Docker or V8, billed only on active compute time.

DeepAware AI builds automation software and physical robotics services for GPU-intensive AI data centers. Its reinforcement-learning scheduler connects to real-time energy markets and cuts power waste, while a unified dashboard gives operators visibility across their infrastructure. Planned autonomous robot inspections and maintenance aim to replace manual staffing rounds and enable continuous, around-the-clock data center operations.

DeepGrove brings frontier AI model capabilities to any device, targeting businesses that need advanced intelligence without being constrained to specific hardware. The product lets enterprise teams run powerful inference at the edge or on existing equipment, replacing the requirement for dedicated cloud infrastructure or specialized machines.

Design Arena builds a platform where users describe what they want to create, from websites and games to images and presentations, and immediately see outputs generated by the leading AI models side by side. Designers, developers, and general creators use it to compare model performance across dozens of creative categories and identify which model best fits a given task. The platform also powers a crowdsourced benchmark that ranks AI models on design quality, replacing manual model-hunting and scattered one-off tests.

Doe builds an AI-powered command center that unifies a team's tools and lets AI agents handle routine operational work across sales, finance, legal, and marketing. It serves startups, enterprises, and government teams, replacing the manual overhead of searching across disconnected software and executing repetitive tasks. When users act directly, built-in features for spreadsheets, analytics, deep research, and document editing make that work faster.

dScribe AI builds an autonomous drone and 3D computer vision platform that measures bulk stockpiles of grain, gravel, and similar materials for agricultural and mining companies. It replaces manual surveys and rough estimates that routinely misstate inventory volumes by wide margins, giving operations teams accurate figures for logistics planning and financial reporting.

Duranium produces titanium, magnesium, aluminum, zirconium, and hafnium in the United States using a novel carbochlorination reactor that recycles CO₂ emissions back into process feedstock and generates bleaching agents as a co-product. The closed-loop design cuts carbon output and production costs, targeting price parity with Chinese producers without subsidies. It replaces dependence on foreign primary metal production at a time when the U.S. has no domestic output of magnesium or titanium sponge.

Eden builds an autonomous marketing platform for ecommerce brands. Its AI agents handle campaign scheduling, A/B testing, and audience personalization across email, SMS, and on-site channels, replacing the work a marketing team or agency would otherwise own. Brand founders get full-stack marketing execution without hiring specialists.

EffiGov builds voice AI agents for local governments, handling inbound constituent calls across every department around the clock. Residents dial the same phone numbers they already know, and the AI answers permit inquiries, animal control requests, and other 311-style calls without staff involvement. The product replaces the manual triage and routing work that previously kept government call center teams occupied with routine, repeatable requests.

Embedder builds an AI coding agent for firmware engineers at hardware companies, handling board bring-up, hardware debugging, HIL testing, platform migrations, and performance optimization. It reads datasheets and schematics, writes part-specific drivers, and runs tests against live hardware, replacing weeks of manual bring-up and root-cause analysis work that embedded engineers would otherwise do by hand.

Epicenter builds an ecosystem of open-source, local-first productivity tools where every app reads from and writes to a single folder of plain text files and SQLite on the user's machine. Current tools include Whispering, a meeting transcription app, and epicenter.sh, which lets users query a codebase through their own hardware. The suite targets individuals who want to replace siloed, subscription-based apps with interoperable tools they fully control, including their choice of AI model.

F2 builds an AI platform for private credit, private equity, and commercial banking teams. It ingests unstructured deal data, such as management decks, credit agreements, and financial models, and produces investment-grade analysis and IC memos, replacing manual diligence and underwriting work. Past deal context is retained across the platform so analysts can benchmark new opportunities against historical precedents.

F4 builds a drawing analysis platform that checks engineering drawings for GD&T compliance, interprets symbols and datums against ASME Y14.5, and generates inspection plans and tolerance analysis inputs. Design engineers and quality teams use it to catch errors before parts are fabricated, replacing manual drawing reviews, redlining sessions, and ad-hoc standard lookups. It processes PDF and STEP files locally without uploading them, targeting hardware teams across defense and industrial manufacturing.

Finto builds AI accounting software for enterprise finance teams, automating the full invoice-to-pay workflow from email inbox through to ERP posting. Its AI agents handle invoice capture, account coding, validation, and approval, replacing the manual, click-through work accountants currently do step by step in traditional invoice platforms. The result is a finished journal entry delivered without human intervention.

Flai builds AI voice agents for automotive dealerships that answer inbound calls and run outbound campaigns around the clock. The agents book service appointments, schedule test drives, and execute recall outreach, replacing the work a business development center rep would handle during and after business hours. Dealerships use Flai to capture leads that would otherwise go unanswered and convert to a competitor.

Fleetline builds an algorithmic load planner for mid-to-large trucking fleets, replacing the manual coordination dispatchers do across siloed tools to balance regulations, customer demands, live fleet data, and driver schedules. Its planning engine combines combinatorial optimization with large language models to adapt to each fleet's rules and capture driver-specific details such as preferences and schedule exceptions. Errors in load planning cost fleets thousands of dollars per mistake, and Fleetline targets that loss directly.

Floot builds a platform that lets non-technical founders and entrepreneurs create full-stack web apps by describing what they want inside AI tools like Claude and ChatGPT. Instead of wrapping existing developer frameworks in AI, Floot created its own coding framework and controls every layer of the stack, including backend, database, and hosting, so the AI can write, debug, and deploy without the user ever touching a terminal. It replaces the patchwork of developer tools that previously stopped non-coders before they could ship anything real.

Fluidize builds an AI platform for scientific computing that automates the setup, execution, validation, and scaling of simulations and experiments. Scientists and engineers use it to accelerate R&D pipelines, replacing manual orchestration of simulation software, cloud compute, dependencies, and versioning. The platform wraps existing simulation stacks or runs workflows end-to-end.

Flywheel AI retrofits existing excavators with remote-operation systems for construction contractors, replacing the need for on-site machine operators. Contractors can control multiple machines or sites from a distance, improving worksite safety and cost predictability. The robotics data generated through operations feeds a dataset used to train autonomous excavator policies.

Freya builds a text-to-speech voice AI platform for enterprise call centers in fintech and retail. Its Adam V1 English voice model is designed to be indistinguishable from a human speaker, replacing live agents on inbound calls for tasks like payment support, transfer inquiries, and fraud checks. Enterprises use Freya to handle customer phone interactions without staffing human agents.

Frizzle grades handwritten student assignments across math, ELA, science, and social studies using AI, then turns each paper into per-student misconception data and live class dashboards. Teachers photograph a stack of papers and get results in roughly eight minutes, replacing manual grading that consumes a quarter of their working week. The product targets K-12 teachers and districts who want continuous insight into student understanding without changing how students work.

Fulcrum builds red-teaming agents for developers working with reinforcement learning environments and AI systems. Its tools probe agent behavior to surface failures and dishonest outputs, replacing ad-hoc manual debugging. The target customer is a developer whose production agents return unreliable or fabricated results.

GETASAP delivers wholesale fresh produce to retailers and restaurants on an 8-hour same-day cycle, using micro-fulfillment centers positioned near demand. It monitors live sales, purchasing, and waste data to drive sourcing and replenishment decisions, replacing fragmented supplier management and in-house procurement teams. The model cuts spoilage and safety stock by tying every buy to actual, near-real-time demand rather than forecast guesswork.

GhostEye runs simulated social engineering attacks, across voice, email, SMS, and chat, to expose where a company's employees, workflows, and trust assumptions break down. Security and engineering teams use it to find the exact gaps real attackers would exploit, replacing generic phishing templates and annual compliance training with live-tradecraft simulations tied to current threat techniques.

gitdot builds a code hosting and collaboration platform for developers, positioning itself as a direct alternative to GitHub. Developers host repositories, discover trending projects, and collaborate on code through gitdot's own infrastructure. The product targets the full workflow that GitHub currently owns, from repository management to community browsing.

Golpo converts documents and text prompts into whiteboard-style explainer videos and podcast-style audio explainers, replacing manual animation and video production work. Marketers, educators, and developers use it to turn written content into narrated, hand-drawn-style videos up to ten minutes long, with support for custom scripts, real images, and 45-plus languages. Both the video and audio products are available via API for programmatic use.

GoRiff builds an AI-powered music editor aimed at both working artists and hobbyists new to music production. It handles the technical and time-consuming parts of arranging and editing audio, replacing the steep learning curve and manual workflow of traditional digital audio workstations. The goal is to let anyone produce a finished track without the overhead those tools normally demand.

Halluminate builds datasets and sandbox environments that train and evaluate computer-use and browser-use AI agents. Foundation model labs and enterprises use these tools to benchmark agent failures and run realistic, scaled training against simulated versions of software like Salesforce. The product replaces ad hoc, manual evaluation cycles that bottleneck teams trying to improve agent accuracy and speed in knowledge-work tasks.

Hera builds an AI motion design tool that turns text prompts into polished, editable animations. It targets creators, such as YouTubers producing infographic videos, who would otherwise spend hours manually keyframing and clicking through Adobe After Effects. Hera automates the repetitive, technical steps in motion graphics production while keeping full creative control with the user.

Humoniq builds AI-native customer support for airlines, handling passenger requests across email, chat, and phone from first contact through to resolution, including rebookings and system changes. It replaces traditional outsourced contact center operations by combining AI automation with travel specialists to execute end-to-end workflows inside an airline's existing systems. The target customer is airlines that currently pay third-party BPOs to manage high-volume, repetitive passenger support.

Hyprnote builds an open-source AI notetaker for enterprise teams in sectors where data privacy is legally or operationally mandatory, including finance, healthcare, and government. It runs fully on-device or on-premises, transcribing and summarizing meetings automatically without sending data to external servers. It replaces manual note-taking and cloud-based transcription tools that cannot meet strict data-sovereignty requirements.

Idler builds reinforcement learning environments that train AI models to code at expert human levels. The environments are constructed from real production coding workflows, giving models hands-on exposure to the kinds of complex, economically meaningful tasks they will encounter in deployment. This replaces the need for AI labs to design and maintain their own coding training infrastructure from scratch.

Imprezia builds an ad network for AI-native apps, connecting advertisers to users inside chatbots, agents, and copilots. Publishers drop in an SDK and Imprezia handles real-time context matching, billing, and attribution, placing brand ads that fit the conversation rather than interrupting it. The product replaces manual monetization deals and generic display ad integrations for AI app developers.

Interface builds OneStop, a city-based social app that lets people post plans, find nearby strangers who want to do the same thing, and coordinate in real time. It targets urban residents looking for spontaneous or scheduled activities, from tennis games to karaoke nights, replacing the friction of group chats, event apps, and social media posts to fill a calendar slot.

Interfere gives engineering, product, and design teams real-time detection and diagnosis of production bugs, surfacing the exact code change responsible and the scope of user impact as issues emerge. It replaces the manual process of correlating error logs, user reports, and deployment history to find what broke and why. Teams use it to catch regressions immediately after a ship rather than hours or days later through support tickets.

Iron Grid builds AI-native insurance products for hardware, starting with energy storage. The founders combine deep battery degradation modeling and financial engineering to price and underwrite risk that traditional insurers cannot accurately assess. The product replaces manual, assumption-heavy underwriting with physics-based and data-driven models of how batteries and related hardware actually fail.

IronLedger.ai builds AI agents that handle accounts payable and card expense management for multifamily and commercial real estate operators, including property managers, developers, and contractors. The platform automates invoice processing, receipt collection, expense coding, and approval routing, replacing manual accounting work done by property accounting teams. It connects to existing property management systems such as Rent Manager to push expenses directly into the books.

Janet AI builds an AI-native ticket management system for engineering teams, replacing manual work in tools like Jira and Linear. It captures ideas, bug reports, and feature requests automatically from Slack, Discord, meetings, and SMS, and advances ticket status as GitHub pull requests open and close. Coding agents in Cursor and Claude Code get full ticket context so engineers spend less time on project administration and more time shipping.

Juxta builds a software-based universal positioning system that tracks people, assets, and robots indoors, outdoors, and underground without GPS, beacons, cameras, or new hardware. It runs on existing devices using custom machine learning models and synthetic fingerprinting to make any space location-aware in minutes. Target customers include defense operators tracking personnel in GPS-denied environments, warehouse and port operators managing assets, and logistics or healthcare teams that rely on manual or infrastructure-heavy tracking today.

Kalinda builds AI-powered research and intake software for class action and mass tort plaintiff law firms. It extracts, summarizes, and aggregates data buried across thousands of pages of medical, financial, and employment records, replacing the manual work done by paralegals and legal nurses. What once took months to compile across an entire litigation, Kalinda compresses into hours, letting attorneys quickly determine claimant eligibility and file cases faster and at greater volume.

Kernel builds open-source browser infrastructure that gives AI agents reliable, high-speed access to the web. It spins up sandboxed cloud browsers in under 30ms, handles CAPTCHA solving, proxy management, authentication, and session persistence so that agent developers do not have to build or maintain any of that themselves. Teams building web automation in QA, healthcare, e-commerce, and GTM use Kernel in place of managing their own browser fleets.

Kestroll builds AI agents that automate video editing and content production for creators and media teams. The agents handle cutting, clipping, multi-cam sync, caption generation, and script writing, replacing the manual post-production work that editors and producers do across podcasts, social clips, documentaries, and brand films. A natural-language search layer lets teams find any moment, person, or scene across their entire footage library without manual tagging.

Knowlify converts documents, blog posts, and written ideas into narrated, animated explainer videos, targeting corporate learning and development teams, universities, textbook publishers, and researchers. It replaces manual video production workflows: users either generate videos themselves through an upload-and-chat interface in minutes, or hand the work to Knowlify's in-house production team for a finished video within 72 hours.

Lanesurf builds voice AI agents that handle carrier outreach for freight brokers, conducting rate negotiations across calls, emails, and texts in parallel to fill loads. The agents vet carriers, compare quotes, and execute bookings, replacing the manual phone work done by carrier sales reps. Each booking feeds a shared pricing and lane dataset that sharpens future negotiations across all customers.

Lark builds a billing platform for B2B SaaS and AI API companies that need to support complex, non-standard pricing models: flat-fee, usage-based, credit-based, hybrid, dimensional, and outcome-based. It handles the billing logic that engineering teams would otherwise build and maintain in-house. The founders previously built a usage-based billing system at a major billing provider.

Lilac manages GPU clusters on behalf of data centers that own them and AI companies that lease them, replacing the manual, error-prone work of standing up and running large-scale compute infrastructure. Each cluster goes through pre-billing validation, continuous monitoring, and engineer-backed support, giving AI teams dependable infrastructure without dedicated ops staff. Startups and researchers gain access to vetted GPU capacity, while data centers turn idle hardware into billable revenue.

Liva AI collects targeted audio and video training data for AI model labs. It focuses on domain-specific and realistic speech and video samples that general internet scraping does not capture, replacing ad-hoc internal data collection efforts. The customers are AI labs building speech and video models that need higher-quality, specialized datasets to improve model performance.

Locata builds an AI-powered referral management platform for primary care clinics, automating the administrative steps that back-office staff handle manually: prior authorization submissions, real-time status updates, and patient follow-up. The platform replaces a fragmented, labor-intensive workflow that leaves half of all specialist referrals in the US uncompleted, reducing staff burden and accelerating the time patients reach specialist care.

Lotas builds AI tools for data scientists and statisticians. Its first product, Rao, is a coding assistant embedded in RStudio that reads, writes, and edits R code while using the user's codebase and environment as context. It replaces manual code authoring and debugging for the roughly 5 million R users who work inside that IDE.

Louiza Labs builds an AI-powered intelligence platform that synthesizes real-world patient data for pharmaceutical teams managing post-market drugs. It surfaces the patient-level reasons drugs underperform commercially, specifically lack of effectiveness, poor adherence, and payer resistance, work previously done through manual market research and fragmented data vendors. The company's initial focus is women's health and GLP-1 therapeutics.

Luminal builds a compiler and inference OS for AI models, targeting developers and teams running large-scale model inference. It compiles PyTorch and Hugging Face models ahead of time into optimized GPU and ASIC kernels, replacing interpreted runtime engines with generated native code. The stack covers the full pipeline from graph IR through hardware-aware optimization to codegen, then dynamically schedules and load-balances workloads across heterogeneous compute clusters to maximize throughput and reduce idle GPU cost.

Magnetic builds an AI-powered tax data entry tool for CPA and tax preparation firms. It ingests client documents, handwritten notes, Excel Schedule Cs, and multi-state K-1s, then populates tax software directly, replacing the manual document review and data entry that consumes staff hours during tax season. The product targets firms stretched thin by an industry-wide accountant shortage and unable to take on additional client load.

MangoDesk builds data annotation pipelines for AI teams that need evaluation and post-training datasets created by domain experts. Teams configure a pipeline and MangoDesk staffs it with vetted specialists, replacing the manual work of sourcing annotators, screening for domain fit, and coordinating labeling operations. The product targets the bottleneck between a model team's data requirements and the operational complexity of producing expert-labeled data at volume.

mcp-use builds open-source developer tools and cloud infrastructure for the Model Context Protocol, letting engineering teams build, deploy, and operate MCP servers and apps for AI agents and platforms like ChatGPT and Claude. Its hosted and self-hosted platform handles authentication, access control, and multi-user environments so teams can ship MCP products to production without building that scaffolding themselves. The work it replaces is the manual setup of secure, scalable MCP deployment pipelines.

Meteor builds an agentic browser for macOS that replaces manual web navigation by letting software agents read pages, click through sites, and complete tasks on a user's behalf. Everyday jobs like booking flights, scheduling calendar events, or shopping are handed off to the agent instead of done by hand. Meteor also offers institutions structured access to public web data through the same underlying agent infrastructure.

Metis builds the post-training and continual-learning layer for enterprise AI agents. Its first product, Insight, uses a company's own data, tools, and environments to continuously train agents, generate reinforcement-learning signals, and validate agent behavior against real tasks before production. This replaces the manual tuning and brittle pilot cycles that prevent autonomous agents from running reliably at scale.

Mimos builds AI agents that help companies rank as cited sources inside AI search tools like ChatGPT. The agents audit a company's site against competitor and customer data to find visibility gaps, generate brand-aligned content, and push technical fixes directly to a CMS. It replaces the manual SEO and content work teams would otherwise do to stay visible in AI-generated answers.

Minimal AI builds AI agents that handle customer support for e-commerce brands, resolving tickets end-to-end across any channel or helpdesk. The agents are trained on a brand's products and business processes, replacing the work of human support staff for the majority of inbound volume. Support shifts from a cost center to a revenue-contributing function through faster, personalized responses.

Modelence builds a batteries-included TypeScript platform that takes a full-stack app from a natural-language prompt to a production deploy, with auth, a MongoDB database, hosting, monitoring, real-time events, cron jobs, email, and AI observability wired in by default. It targets developers and non-technical builders who would otherwise spend weeks stitching together separate services before shipping anything. The platform replaces the stack-assembly work that sits between a working prototype and a production-grade application.

Monarcha builds an AI-powered geospatial platform that lets logistics, real estate, urban planning, and environmental teams run spatial analysis without GIS expertise or specialist software. Users upload maps, vector files, or raster data and issue plain-language prompts to georeference scanned maps, extract structured data from drill logs and assay documents, digitize geological features, and run tools such as buffer analysis, heatmaps, and PostGIS queries. The product replaces desktop GIS workflows and manual data extraction that previously required coding skills and hours of specialist time.

Motives runs AI-powered qualitative research for consumer brands, replacing traditional focus groups and agency-led interviews. A team of AI agents builds the research plan, conducts up to 60-minute video interviews with real consumers, and delivers synthesized insights within 48 hours. The work it replaces: weeks of briefing, recruiting, moderating, and analyzing that insights and marketing teams previously handed to research agencies.

Munify builds a cross-border neobank for people in the Middle East and its diaspora, letting them open a USD account without U.S. residency, hold and spend via virtual cards, and send money internationally. It runs on stablecoin and blockchain rails to settle transfers in real time at low cost, replacing the slow, expensive remittance and banking services this population typically relies on. Businesses also use it to manage global accounts and accept or send payments across borders.

Nautilus builds software for car wash operators, handling membership management, online sales, customer support, marketing automation, and reporting through a single platform that connects to the operator's existing point-of-sale system. It replaces older, costlier tools that operators previously ran separately for each of those functions. The target customer is the multi-site car wash business looking to grow recurring memberships and reduce manual work.

Nexa Labs builds an implantable microchip system and AI platform for cattle health monitoring. Cattle farmers use a subcutaneous sensor, inserted during routine procedures, to continuously collect biometrics that feed an AI engine delivering early disease alerts, reproductive tracking, and GPS location. The system replaces manual observation and fragmented herd records with continuous, data-driven insight across the entire herd.

Nixo builds an operations platform for forward-deployed engineering teams. It centralizes customer context, workload tracking, and code intelligence so FDE teams can manage enterprise onboarding, custom integrations, and complex deployments without manually chasing notes or duplicating work across accounts. It replaces the patchwork of support tools and engineering trackers that FDEs currently stitch together to run customer engagements.

Normal builds applied AI research infrastructure for hardware designers working on rockets, robots, aircraft, satellites, medical devices, and semiconductors. Its tools, centered on data and evaluations, advance frontier AI capabilities for engineers in these fields, replacing manual benchmarking and data curation work that currently slows hardware development cycles.

NOSO LABS builds AI agents for home services technicians, helping them diagnose problems accurately and close sales opportunities on the job. Field technicians are typically strong in one area, whether diagnosis, selling, or paperwork, but rarely all three. NOSO LABS replaces the need for that all-in-one technician by giving any tech an AI-powered co-pilot in the field.

Notte builds a browser automation platform for developers and enterprise teams that need AI agents to operate reliably on the web. It packages browser sessions, AI-driven agents, and serverless browser functions into a single API, replacing hand-written scraping scripts, fragile selectors, and separately managed browser infrastructure. The target customer is any engineering team deploying web automation at scale, from one-off task agents to high-concurrency production workflows.

Novaflow builds an AI data analyst for biology labs, letting life scientists upload experimental data and ask questions in plain English to get publication-ready plots and analysis. It replaces manual bioinformatics work, including writing and running computational pipelines, that traditionally takes months. Researchers at academic medical centers and universities use it to go from raw biological data to interpretable results in minutes.

Nox Metals runs software-driven metals processing factories that supply custom-cut aluminum and stainless steel to manufacturers across North America. Its platform handles quoting, scheduling, and production operations, replacing the slow, manual workflows typical of traditional metal service centers. Buyers get cut-to-size material with instant online pricing, cutting out the lead time and friction of conventional supply chains.

Nozomio builds tools that solve the context problem in agentic coding workflows. Its product Nia automatically researches and indexes external codebases and documentation, then feeds the right context to coding agents via MCP or API. This replaces the manual work developers do to curate and maintain the inputs that keep AI coding agents accurate and on-task.

Nuntius builds alignment infrastructure for AI models deployed in production B2B settings, where rule-following is a hard requirement rather than a best-effort goal. The product addresses the gap between a capable AI model and one that reliably stays within defined constraints, replacing manual guardrail engineering and ad-hoc prompt patching that teams otherwise do themselves.

Okibi generates a command-line interface from a software product's codebase, giving AI agents a reliable way to operate that product without constructing raw API calls. It researches existing tools, designs a command set, and verifies every command in a cloud sandbox. The target customer is a software company that wants agents to use its product correctly and completely, replacing manual CLI authoring and the fragile tool configurations common with MCP servers.

Omnara provides an open-source platform for deploying and running AI agents in production. Developers use a single stateful API to ship agents into product dashboards, Slack workspaces, CLI tools, and custom workflows, on any model, machine, or cloud. It replaces the bespoke infrastructure teams otherwise build themselves to handle durability, retries, multi-provider support, and shared visibility.

OnDeck AI builds a vision analysis platform that lets enterprises search, understand, and report on any footage without training a custom computer vision model. Customers in defense, robotics, public safety, and industrial monitoring use it to detect objects, behaviors, and events across large video archives, replacing months of data collection, model training, and deployment work. The platform runs in cloud or on-premises environments and adapts to each customer's workflow through agentic self-improvement.

Opennote built an AI-powered note-taking platform for students and individual learners whose study material was fragmented across multiple apps. The product connected personal notes and web sources into a single workspace, then surfaced proactive edits and feedback tailored to each user's writing and learning patterns. Opennote was acquired by Reducto in May 2026 and the app has since been shut down.

Orange Slice builds an AI-powered spreadsheet for sales and revenue operations teams where each column runs TypeScript code, generated from plain-English instructions. Teams describe the data they need, such as finding a CEO's contact info or tracking hiring signals, and the tool writes the code, creates the columns, and executes them across thousands of rows. It replaces manual prospecting, CRM enrichment, and point-and-click workflow builders.

Outrove builds an AI recruiting agent that conducts video and audio interviews over Google Meet, presenting as a human interviewer. It targets B2B hiring teams and replaces the early-stage screening work that human recruiters or coordinators would otherwise handle. The agent handles real-time candidate conversations end-to-end, cutting the manual effort out of initial interview rounds.

Palace builds AI agents that run back-office operations for trucking carriers, handling load planning, dispatch, order entry, and invoicing. The agents plug into a carrier's existing systems and take over the repetitive decisions that otherwise occupy planning, dispatch, and customer-service teams. Carriers get a leaner operation and more revenue per driver without replacing their current software stack.

Pally builds an AI personal assistant that consolidates messages, tasks, and notifications from across a user's apps, including Gmail, WhatsApp, Slack, iMessage, and calendar tools, into a single text-based conversation. Users text Pally as they would a friend, and it tracks action items, drafts replies, books travel, creates tickets, and surfaces what needs attention, replacing the manual work of checking and managing multiple inboxes and tools.

Paloma builds AI-native operational software for companies that make physical products, giving their operations teams intelligence, custom internal tools, and workflow automation in place of manual, engineer-dependent workarounds. It connects ERPs, spreadsheets, emails, and other data sources into a unified layer that understands how a specific business runs, then answers operational questions, enforces compliance, and moves data between systems automatically. The work it replaces includes hand-keyed data transfers, ad-hoc reporting, and bespoke internal tools that ops teams could never get engineering time to build.

Panels builds and sells curated audio datasets to voice AI teams, covering multilingual speaker-separated recordings, single-speaker scripted audio, and turn-taking evaluation sets. It sources material from vetted, diverse contributors and works directly with frontier voice labs and early-stage startups to match data to each team's training and evaluation specs. The product replaces the effort of collecting, cleaning, and structuring raw audio that model teams would otherwise have to run themselves.

Pangolin builds an open-source, identity-aware remote access platform built on WireGuard, letting IT teams connect employees, AI agents, and infrastructure to private and public resources without legacy VPN complexity. Users authenticate through their existing identity provider, and access is granted per application under zero-trust rules, replacing traditional network-wide VPN tunnels. An AI Gateway layer extends the same identity and policy controls to AI provider APIs and self-hosted models, eliminating shared API keys.

Parachute builds governance infrastructure that helps hospitals select, evaluate, and monitor clinical AI tools. It automates vendor scoring for accuracy, safety, and bias, tracks deployed models for drift and compliance failures, and generates an immutable audit trail for regulators. The product replaces manual committee reviews, hand-written test scripts, and the ongoing compliance work that hospital IT and clinical informatics teams otherwise do themselves.

PARES AI builds an end-to-end operating platform for commercial real estate brokers, investors, and property managers. It consolidates CRM, AI-assisted underwriting, marketing material generation, deal distribution, and property management into one system, replacing the patchwork of tools like Yardi and Appfolio that teams typically juggle across a deal lifecycle. Brokers use it to source leads, run underwriting, produce offering memoranda and BOVs, and manage assets after close.

Perseus Defense builds man-portable, self-guided missile systems that intercept and destroy Group 1 and Group 2 attack drones for the U.S. Department of Defense and Department of Homeland Security. Its Harpe interceptor is designed for mass production at under $10,000 per engagement, replacing current kinetic counter-UAS solutions that are far more expensive, one-to-one in capability, and restricted from use on American soil. The system deploys in under 60 seconds, operates autonomously, and can network across multiple launchers to cover a wider battlespace.

Perspectives Health builds an AI platform for behavioral health clinics, automating the administrative work that pulls clinicians away from patient care. The platform runs nightly audits across every chart, flags compliance issues, generates utilization review packets, writes clinical notes and assessments, and assembles audit defense documentation. Work that previously required staff hours chasing documentation gaps, insurance requirements, and payer audits is handled by the software inside the systems clinics already use.

Phases builds AI agents that automate clinical trial operations, beginning with patient recruitment. The agents scan medical records and conduct voice interviews to screen patients for trial eligibility, replacing work that trial coordinators and site staff do manually today. The long-term goal is to run the full clinical trial back-office through agents, cutting the time and cost required to bring drugs to market.

Pingo AI builds a conversational language-learning app that teaches spoken fluency through back-and-forth dialogue with an AI tutor. Learners set their goals, practice real-life scenarios such as travel, job interviews, or daily routines, and the tutor corrects them in the moment and tracks their progress across sessions. The product replaces structured coursework and grammar drills with ongoing spoken practice that adapts to each learner's level and goals.

Playgent builds high-fidelity sandbox environments for enterprise AI agents, targeting engineering teams that need to reproduce production failures. Each environment is delivered as a single MCP URL with mocked tool calls, abstracted authentication, and test data seeded via natural language or JSON, replacing the manual, brittle process of reconstructing the exact user state and context that caused an agent to break. Runs export as OpenTelemetry traces for direct use in observability tools or reinforcement learning pipelines.

Pleom builds an agentic business intelligence platform that automatically surfaces insights across a company's data without requiring users to learn a BI tool. It targets business teams that need fast answers from their data, replacing the manual work of building dashboards, writing queries, and configuring visualizations.

Pond builds an AI-native desktop messaging app for professionals who manage high volumes of text-based customer conversations. It works across existing services like iMessage, surfacing unanswered messages, enabling conversation snoozing, and drafting replies in the user's own voice. Pond replaces the manual work of triaging and responding to messages across fragmented inboxes.

Proliferate builds the operational backbone that lets AI coding agents deploy software autonomously, handling isolated test environments, event-based triggers, and multi-step validation so agents can take code from a prompt all the way to production without human hand-holding at each stage. The product targets engineering teams that want to run coding agents beyond code generation, replacing the manual DevOps scaffolding those teams would otherwise build themselves.

Reacher builds an AI platform that handles creator marketing operations for brands selling on TikTok Shop, Instagram, and YouTube. It replaces manual work across creator discovery, outreach, campaign management, and performance reporting, letting marketing teams run affiliate programs at scale without growing headcount. The product centralizes creator relationships, sample logistics, content strategy, and GMV tracking in one place.

RealRoots matches women with compatible friend groups and organizes in-person social experiences for them, replacing the effort of finding, vetting, and planning outings with strangers. A short compatibility survey feeds an AI matchmaker that pairs users, books venues, and structures a six-week series of guided meetups in their city, led by a human guide. The product targets adult women who want real-life friendships but lack a practical way to build them.

Relling builds reinforcement learning and data infrastructure for robotic workcells deployed on American factory floors. Its platform handles the full stack: on-premise AI inference, line integration with MES and PLC systems, fleet management, and operator enablement, replacing the patchwork of manual programming and fixed automation that breaks down under high-mix production. Customers span automotive, food and beverage, electronics, battery and EV, aerospace, metals, plastics, agriculture, textile, and recycling.

Rid handles the entire process of selling secondhand items for consumers who never get around to listing them. A seller texts Rid what they want to sell, and Rid researches the item, lists it across marketplaces, fields buyer questions, negotiates, and arranges delivery. It replaces the effort that keeps most people from selling things they already own.

Risely AI builds AI agents that handle administrative work across university departments, from enrollment and financial aid to advising, compliance, and advancement. The platform connects directly to existing university systems, such as SIS, LMS, and CRM tools, and deploys specialized agents that execute tasks those systems require human staff to do manually. Universities get broader staff capacity without adding headcount, and students get faster, consistent responses across every office.

RowFlow converts web forms into AI-driven text conversations, targeting B2B teams that collect structured data from clients, registrants, or respondents. Instead of static forms that yield low response rates and messy submissions, RowFlow conducts the exchange automatically over text, Slack, or embedded chat, and returns clean, structured data. It replaces form-building and manual follow-up for intake, compliance, feedback, and survey workflows.

Ruvo builds a multi-currency account for Brazilians who need to receive, hold, convert, and send money between U.S. dollars and Brazilian reais. It connects Pix, Visa, and crypto in a single app, letting users get U.S. account details, convert between BRL, USD, and digital dollars at real market rates, spend abroad with a dollar card, and earn yield on dollar balances. It replaces the combination of Brazilian bank wires, IOF-heavy cards, and slow SWIFT transfers that individuals and businesses currently piece together for cross-border payments.

Sagekit builds an AI workspace for research and writing, aimed at knowledge workers who move from scattered notes and sources to finished documents. The product works alongside users in real time, replacing the manual effort of organizing research and drafting content into cohesive written output.

Serafis builds a search and discovery engine that turns podcast transcripts, interviews, and independent media into structured market intelligence for institutional investors and operators. Its AI knowledge graph indexes conversations from public company CEOs, founders, and asset managers, content that traditional financial terminals do not cover. Investors use it to research themes, companies, and people without manually sifting through long-form audio and video.

Shor builds an Employer-of-Record platform for startups and growing remote teams that need to hire and pay international employees and contractors. It uses AI agents to handle contracts, worker classification, compliance, and payroll end to end, and settles payments via stablecoins in seconds rather than days. Shor targets companies priced out of enterprise-grade platforms like Deel and Rippling, replacing a slow, fee-heavy process with one founders can set up in minutes.

SigmanticAI builds an AI-native chip design assistant for hardware engineering teams, automating RTL generation, verification IP development, and debugging inside a VSCode-based environment. It replaces manual HDL coding and testbench authoring by taking designs from natural language specifications through synthesizable Verilog and SystemVerilog, iterating against real compiler and simulator feedback until the design compiles and passes synthesis. A second autonomous engine, Polaris, targets fully end-to-end chip generation from specification to verified implementation.

Sira builds an AI-native workforce operations platform for businesses that run hourly, deskless teams. It handles time tracking, scheduling, payroll, and HR by deploying AI agents that text and call workers directly: chasing missed clock-ins, onboarding new hires over SMS, filling open shifts, and flagging timesheet anomalies. This replaces the manual back-office coordination that managers and HR staff otherwise do themselves.

Slashy builds an AI-powered email tool for professionals who need to move through high volumes of inbox work quickly. It brings keyboard-driven, AI-assisted editing to email, letting users draft, triage, and act on messages without the slow back-and-forth of manual composition. The product replaces the time spent reading, writing, and organizing email by hand.

Socratix AI builds AI agents that work alongside fraud and risk teams at fintechs and banks, handling alert investigation and evidence gathering that analysts currently do by hand. Each agent pulls in transaction data, device signals, identity checks, and other sources into a single workspace, then produces an explainable decision without requiring a human to stitch context together across tools. The product targets the repetitive, high-volume review work that slows teams down and drives operational headcount growth.

Solva builds AI software that automates insurance claims processing for property and casualty carriers and third-party administrators. It handles complex claims end-to-end, replacing manual review, adjudication, and quality-assurance audits that claims teams currently do by hand. The system is designed to catch incorrect payouts and accelerate resolution cycles on high-complexity claims.

Spotlight Realty runs a sell-side residential rental brokerage in New York City where an AI agent handles tenant inquiries, screening, and showing scheduling from listing through closing. It targets landlords who currently pay traditional broker commissions for that same end-to-end leasing work. The service covers MLS syndication, photography, staging, offer comparison, and digital document signing at one-third the standard commission rate.

stagewise builds an open-source, browser-native frontend coding agent that lets backend developers, designers, and product managers make UI changes directly inside a running localhost app. Users select elements and describe changes in plain language, and stagewise edits the local codebase without switching to an external tool. It works across frameworks and can be dropped into existing projects, replacing the manual back-and-forth between design intent and code.

Stellon Labs researches and builds compact AI models designed to run directly on edge devices such as smartphones, wearables, and embedded systems. The work targets the constraint that frontier-scale model capabilities have traditionally required cloud or data-center hardware. Its first release, KittenTTS, is an open-source text-to-speech model that replaces heavier, server-dependent TTS pipelines with a model small enough to run on-device.

Stockline builds an AI-native ERP for food wholesalers, using AI agents to handle repetitive back-office work. The starting point is order entry, a task currently done by hand or through outdated software. Stockline replaces legacy ERP systems with a platform where AI agents do the operational labor directly.

Stormy AI builds an AI platform for field-service businesses, targeting trades and service companies where most of the workforce operates away from a desk. It acts as a trainable AI employee across roles like front desk, dispatch, and back office, handling calls, texts, scheduling, and invoicing. It replaces the work a new hire would do across those functions, trained by the business's own team through calls and messages rather than formal software onboarding.

Synthetic Society builds AI-simulated user research tools for product teams and developers. It generates synthetic user personas that stand in for real user testing, replacing the time and logistics of recruiting, scheduling, and running participant studies. Teams use it to get behavioral and attitudinal signal on their products without running live sessions.

TectoAI builds a governance platform for companies deploying agentic AI tools across their workforce. It helps business operators identify which roles can be handed to AI agents, onboard those agents, and then monitor their behavior and regulatory compliance in production. The platform replaces ad hoc, manual oversight of AI deployments with a structured, continuous management layer.

Tesora builds AI agents for insurance actuaries, handling complex recurring work such as rate filings, reserve opinions, and pricing analyses that previously consumed weeks of manual effort each month. Every output carries a full audit trail with source citations, and each agent result clears a series of automated checks before it reaches the actuary's desk. The product is built to actuarial professional standards and enterprise security requirements, running in single-tenant environments with no shared infrastructure.

The Prompting Company helps B2B software products get discovered and used by AI assistants such as ChatGPT, Perplexity, and Claude. It identifies the questions target users ask AI, produces and publishes content optimized for AI citation, and routes AI-driven traffic to dedicated pages, replacing the manual work of SEO agencies and content teams. For developer-facing products with APIs, CLIs, or MCP servers, it also audits and fixes the friction points that prevent AI agents from successfully completing tasks with the product.

Topological builds physics-based AI models that automate topology optimization for hardware engineering teams. Its first model, UToP-v1, takes a mechanical problem's physical requirements and generates structurally efficient, manufacturable CAD designs, replacing slow iterative simulation workflows that today can take days. The target customer is hardware teams doing complex structural design work where geometry, physics, and manufacturing constraints must all be satisfied simultaneously.

Trace builds a workflow automation platform for mid-sized enterprise teams that splits processes into discrete steps, sends repetitive work to AI agents, and routes judgment-dependent tasks to the right people with permission controls and a full audit log. It replaces ad-hoc tool sprawl and manual handoffs across departments, acting as a coordination layer between AI agents and human reviewers. Teams use it to deploy AI across operations without months of integration work.

TraceRoot.AI builds an open-source AI agent that watches live production systems and repairs bugs without human intervention. It pulls in telemetry signals, including traces, logs, and metrics, alongside the codebase, to diagnose failures and apply fixes autonomously. Engineering teams use it to cut the manual triage and patching work that typically follows a production incident.

truthsystems builds an AI governance platform for law firms, giving them real-time control over how attorneys and AI agents use generative AI across matters and clients. Two products sit at the center: Charter intercepts every prompt at the browser or desktop level, checking it against firm policy and client-specific outside counsel guidelines before it reaches any AI tool, while Alexandria acts as an MCP gateway that connects agents to firm data with full audit trails. The platform replaces manual compliance review and ad-hoc access controls with automated, per-matter enforcement.

Uplift AI builds foundational voice models for regional languages, starting with Urdu, and exposes them via APIs, SDKs, a text-to-speech studio, and an AI calling product. Developers use these tools to add voice interfaces to services such as e-commerce, banking, and tutoring, reaching speakers who do not engage with text-based digital products. The company owns its full stack: data collection, labelling tooling, model training, and voice infrastructure.

Veritus builds an AI-powered communications platform for consumer lenders, handling loan servicing and collections across voice, SMS, email, and chat. Its agents conduct inbound and outbound conversations, negotiate repayment plans, and follow up with borrowers end to end, replacing the human agents and manual outreach that lenders traditionally use to manage delinquencies and recoveries.

Verne Robotics builds AI models that control robot arms and teach them new skills in hours. The robots are deployed on factory floors and in back-of-house operations to handle tasks such as kitting, assembly, and inspection, replacing manual labor for customers including biotech firms and apparel brands.

VibeFlow generates full-stack web applications from natural-language prompts, combining frontend scaffolding with a visual, editable backend workflow builder and built-in integrations. It targets developers and builders who use tools like Lovable or v0 and hit a wall when backend logic becomes too opaque to iterate on through prompting. VibeFlow replaces that closed black-box backend with a node-based visual layer that users can inspect and modify directly.

Vulcan Technologies builds AI agents for regulatory drafting and legal compliance, serving government agencies and businesses that would otherwise commission large consulting engagements. Its JUSTINIAN agent gives government knowledge workers instant access to a structured database of U.S. laws, regulations, court cases, and enforcement records to draft regulations, respond to public comments, and produce cost-benefit analyses. Its TRAJAN agent sits on government websites to help businesses and citizens navigate permits, licenses, and regulatory questions without staff intervention.

Wafer builds a continual inference platform that automatically profiles AI workloads and optimizes the GPU serving stack in real time, replacing the manual tuning and static configurations that teams otherwise maintain themselves. It targets AI companies and cloud providers running large-scale LLM inference, delivering lower latency and higher throughput as traffic patterns shift. The platform replaces both bespoke kernel engineering and prefetching infrastructure that engineers build to keep inference fast under load.

Wayline builds an AI phone and messaging assistant for property managers, handling inbound calls and tenant communications around the clock. It replaces call centers, robo-menus, and manual front-desk staff by answering, triaging, and resolving resident requests, scheduling appointments, and routing issues to the right team members. The system connects to existing property management software and is trained on real estate workflows, covering everything from routine maintenance requests to after-hours emergencies.

Wedge builds and operates AI agents that automate back-office work for large health systems, replacing manual staff workflows in areas like revenue cycle and administrative operations. The engagement starts with an onsite audit that maps automatable tasks, then forward-deployed engineers build and maintain tailored agents tied to measurable outcomes. Wedge handles development, integration, and ongoing maintenance so health system operators do not manage the AI stack themselves.

ZeroEval builds an automatic optimization layer for AI agents, targeting companies that deploy agents in production. It replaces manual experimentation by running calibrated LLM-based judges and automated evaluations, cutting the iteration cycle companies face when tuning agent reliability. The product addresses the reliability gap between a working prototype and a dependable production agent.