a16z speedrun SR007 companies

Abliteration.ai builds an unrestricted large language model API for red teams, trust and safety teams, synthetic data pipelines, ML researchers, and defense and government workflows. It handles the prompts that mainstream models refuse, covering security research, adversarial testing, and sensitive training data generation. Teams swap in Abliteration by changing a single base URL, with no rewrites required, because the API is fully compatible with the OpenAI and Anthropic SDKs.

Acaysia builds a real-time AI control layer that sits on top of existing industrial reactor controls to extract more yield and lower energy consumption from the same physical plant. Chemical and industrial facilities today rely on operator intuition and periodic advanced process control projects that take 18 months, cost millions, and decay as conditions drift. Acaysia replaces that cycle with continuous, model-predictive optimization that transfers across reactors and falls back to the existing PID system in under 100 milliseconds.

AgentCat builds analytics and observability tooling for teams shipping MCP servers, Claude Connectors, and ChatGPT Apps. It gives developers a full trace of every agent session, grouping activity into inferred goals and surfacing errors, hallucinations, and tool-call failures by impact. The product replaces manual log-digging and guesswork about how AI agents actually navigate a product.

Alia builds an agentic coordination layer for teams, monitoring tools like Slack, Jira, GitHub, and email to catch misalignments and blockers before they stall work. It surfaces gaps between what teams commit to and what they actually deliver, then pulls the right people into a conversation to resolve the issue. The product replaces the manual overhead of cross-functional follow-up and the status meetings that exist to catch what falls through the cracks.

Alven builds an AI property manager that handles the full tenant lifecycle for real estate agencies and large property funds. It covers leasing, maintenance coordination, rent delinquency, move-outs, compliance, and back-office accounting, replacing the day-to-day work of human property managers, leasing agents, and maintenance coordinators. Agencies add Alven directly to their existing property management software as a user, with no integration project required.

Athena operates as an AI-native law firm handling M&A, venture capital financing, and commercial real estate transactions for deal-focused clients who need faster closings and predictable costs. Senior partners own every matter from start to finish, while proprietary AI handles diligence, drafting, and repetitive transaction work that traditional associate-heavy teams perform manually. The model targets buyers and investors running repeat transactions where closing speed and legal spend directly affect deal economics.

Banbury Road trains coordinated civilizations of small, specialized models rather than scaling a single large one, pursuing a second axis of AI capability growth: number of models instead of model size. The resulting systems target frontier tasks at a fraction of the compute and memory cost of monolithic state-of-the-art models. The work replaces the assumption that capability requires ever-larger parameter counts, offering an alternative scaling path for applications that today demand closed-source giants.

Baro runs operations and growth for consumer brands through a single AI platform, replacing the fragmented mix of agencies and outsourced vendors brands typically rely on for marketing, logistics, finance, compliance, and customer support. The platform sees the entire business at once, recommends specific actions tied to the brand's goals, and executes them once a founder approves. It replaces the general manager role and the patchwork of external operators that fill it today.

Blueprint builds an AI platform that converts plain-language requirements into complete hardware designs, replacing the manual work of searching datasheets, sourcing components, and drafting wiring diagrams. Hardware engineers and product teams use it to go from high-level specs to a working prototype design in minutes rather than weeks. The product targets the front-end design bottleneck that has kept hardware development slow even as software tooling has accelerated.

Bounty runs an open marketplace where AI agents compete to complete discrete work tasks, from lead generation and deep research to video production and ad creative. Founders and small business owners post a task with defined completion criteria, fund the reward upfront, and pay only when an automated verification step confirms the output meets requirements. This replaces managing a stack of AI tools or hiring contractors for recurring and one-off work.

Bric builds vertical AI agents for an industry it has not yet disclosed publicly. The product is in stealth, so the specific customer, job, and work replaced are not yet available.

Cartsy builds a shopping assistant that identifies products inside social media posts, reels, stories, photos, and videos, then surfaces where to buy them across multiple stores. Consumers send content directly to @cartsyapp on Instagram and receive matched product listings with price comparisons in seconds. The product replaces the manual work of tracking down items spotted while scrolling, from screenshotting and searching to tab management.

ChatOverflow captures knowledge from coding agent sessions and surfaces it to other agents on the same engineering team in real time. When one agent solves a tool error, navigation problem, or codebase quirk, that finding is stored and automatically pulled by the next agent that hits the same wall. It replaces the manual work of writing skill files, posting in Slack, and hoping teammates adopt what was learned.

ChronoSpace builds 4D world models, combining 3D spatial reconstruction with time, to digitize physical environments for manufacturing, robotics, and construction. The models power applications such as factory floor digital twins that monitor material volumes to raise throughput, and synthetic environments that accelerate robot training. The product replaces manual site surveys, static CAD models, and slower simulation pipelines used in physical industries.

Cluster builds GTM infrastructure for AI agents. It gives B2B sales agents access to more than 50 domain-specific data integrations, real-time buying signals, and multichannel outreach in a single platform, replacing the manual work of stitching together prospect data and sequencing tools. The primary customer is an AI-native B2B startup whose agents need structured pipeline-building capabilities that general-purpose models do not supply out of the box.

Codos builds enterprise AI transformation software that acts as a virtual Chief AI Officer for public and PE-backed companies. It runs AI agents that interview employees across the workforce to surface inefficiencies, then deploys a unified company knowledge base to automate sales, operations, and engineering workflows. The product replaces expensive, slow consulting-led AI implementations with software that delivers measurable P&L impact.

Coherence Health delivers AI-driven functional medicine care to women with chronic, overlapping conditions, starting with thyroid and autoimmune disorders such as Hashimoto's. The platform replicates the diagnostic reasoning of elite functional medicine clinicians to identify root causes that conventional care routinely misses. It targets the roughly one in six American women who live with debilitating symptoms that stem from connected thyroid, gut, and immune dysfunction.

Colmez builds an agentic platform for hazardous waste sales teams, connecting regulatory, commercial, and operational data that the industry currently manages through phone calls, spreadsheets, and handwritten notes. It surfaces new waste-stream leads, interprets lab analyses, retrieves historical contracts, and handles invoicing, replacing the manual coordination work that stalls revenue between handoffs. Customers are environmental services operators whose sales teams use Colmez to keep accounts moving and close more business.

Contxt builds a unified memory and action layer for enterprise teams, sitting across tools like Slack, Teams, Claude, and ChatGPT. Employees query their organization's tools, people, and history in plain language and get finished work back in seconds, replacing the constant app-switching and manual coordination that consumes most of a knowledge worker's day.

Corlen builds an AI-native loan origination platform for commercial real estate lenders and brokers. It unifies application intake, document processing, verification, and borrower communication into a single loan record, then runs AI agents to do the work that loan officers and processors handle manually. The product replaces fragmented back-and-forth across spreadsheets, email, and disconnected LOS tools with a structured workflow that moves a file from intake to underwriting without manual intervention.

Cube Computer gives developers a full Linux machine in the cloud, purpose-built for running AI agents continuously. It replaces the workaround of keeping a laptop open and connected, or manually managing cloud VMs and racks of local hardware. Developers get a persistent, always-on environment with one-click setup and a dedicated desktop interface for building and monitoring agents from any device.

Cybernetic Physics deploys autonomous robots inside data centers to perform rack and stack, cabling, and routine maintenance around the clock. The company sources third-party humanoids and mobile manipulators, then takes responsibility for real-world deployment and ongoing operation, replacing the human technicians who currently staff these facilities. The target customer is data center operators who want to run fully lights-out facilities.

Delightful AI builds reinforcement learning environments for training AI agents on healthcare and science tasks. Its infrastructure gives model developers structured clinical and scientific settings, complete with records, tool calls, and scoring, to post-train agents on work that previously required human expert judgment. The environments replace hand-crafted evaluation setups and let teams benchmark multiple model configurations against identical real-world tasks, tracing every score back to the decisions that produced it.

DRS builds and operates a portfolio of generative AI consumer apps, targeting mobile users across categories where AI-generated content drives the core experience. Each app runs on a shared backend covering growth, content management, pricing, and user identity, so retention and monetization learnings from one app carry directly into the next. The studio replaces the traditional build-from-scratch approach to consumer app development with a compounding operating system that accelerates each successive launch.

Elu builds a product intelligence layer that watches real user sessions to detect bugs, UX friction, and silent failures, then generates, tests, and ships the fixes autonomously. It replaces the manual loop of combing analytics dashboards, triaging session recordings, and writing patches by hand. Product teams at software companies get a continuous self-healing system that goes from detected issue to deployed code without human intervention.

Embrasure builds a data warehouse designed for AI agents rather than human analysts. It ingests data from sources like Postgres, Stripe, and Salesforce, stores it in open Apache Iceberg format, and gives agents a persistent memory layer to query, learn from, and act on that data. It replaces the manual work of building pipelines, writing SQL, and maintaining warehouse infrastructure that traditional platforms like Snowflake and BigQuery require teams to manage themselves.

Fictional builds a consumer platform where users discover, follow, and interact with AI characters that post images, short videos, and messages as persistent, named identities. Each character maintains a consistent persona across content and direct interaction, replacing the static, single-turn dynamic of traditional chatbots with an ongoing creator-style relationship. The product targets entertainment and social discovery use cases where users today follow human creators on conventional social platforms.

Flip delivers personal finance management through iMessage, replacing standalone apps like Acorns, Rocket Money, and Splitwise with a single AI agent consumers text directly. It monitors connected bank and investment accounts, surfaces savings opportunities, cancels unused subscriptions, rounds up purchases into investments, splits expenses, and mirrors trading portfolios, all through a conversational interface. The product targets everyday consumers who want active money management without juggling multiple fintech apps.

Galileo Space builds AI-powered satellites that fuse optical and RF sensing onboard and downlink finished intelligence, not raw data, to defense operators. Onboard compute classifies RF signals, cues the camera, and produces a single fused output in one pass, cutting the time from tasking to actionable intelligence from hours to minutes. The product targets national security customers who today wait for ground-based processing of full imagery passes before they can act.

Harbor AI builds an operating layer for GPU fleets and AI infrastructure teams. It monitors workload performance, diagnoses failures and slowdowns, applies approved fixes, and verifies recovery, replacing the manual triage and intervention work that AI engineers do today. A context engine tracks configuration history and past fixes to inform future decisions, with a long-term goal of autonomously orchestrating workloads across owned and rented compute.

Hoid builds an agentic AI compiler that optimizes AI models for any hardware target, replacing slow, manual, chip-specific tuning work done by specialized engineers. Inference providers and teams deploying AI models use it to reach peak hardware utilization across accelerators without hand-writing kernels for each chip. The compiler profiles a model, searches the optimization space autonomously, and produces a performant implementation in hours rather than weeks.

Infragrid watches how knowledge workers move through browsers and desktop apps, then surfaces ready-to-run agents to handle the repetitive steps without waiting to be prompted. It converts observed clicks, app interactions, and context from tools like Slack and calendars into reusable runbooks that execute across real interfaces, requiring no custom integrations. The product replaces manual, repeated task sequences that workers would otherwise perform themselves or script from scratch.

Kairo runs offensive security engagements against web apps, APIs, AI models, and autonomous agents, targeting the flaws that automated scanners miss. Customers get reproducible proof-of-concept exploits with remediation guidance mapped to OWASP and MITRE ATLAS, replacing checklist-driven penetration tests that were built for slower, simpler software. The work covers classic application vulnerabilities alongside AI-specific attack surfaces: prompt injection, agent hijacking, tool-call misuse, and data exfiltration through model outputs.

Keythorn builds the liability insurance layer for companies deploying AI agents, chatbots, and autonomous workflows. It stress-tests and monitors those systems, produces a technical risk profile, and uses that assessment to place AI liability coverage, replacing the guesswork of legacy underwriting with evidence-based evaluation. Enterprises use the coverage to satisfy procurement requirements and enter regulated industries; insurance carriers use Keythorn to underwrite AI risk they cannot yet evaluate in-house.

KIRA builds modular, on-site wastewater treatment systems for mining, oil and gas, and data center operators. Each containerized unit uses heat-pump-driven evaporation and condensation, with no chemicals, to convert high-TDS industrial wastewater into clean water and recoverable solids, arriving on a truck and running the same day. The product replaces custom treatment plants, chemical inventories, and multi-year procurement timelines.

Lazyweb builds an AI agent platform that runs product growth for consumer app companies. It draws on a proprietary library of over 300,000 learnings from top consumer products to generate, test, and apply growth hypotheses, replacing the manual work of a dedicated growth team. Product managers get experiment recommendations and copy fixes shipped automatically, without running the analysis themselves.

Leonidas builds a dedicated bank for Europe's frontier technology companies, serving businesses in defence, dual-use, space, robotics, energy, compute, and biotechnology. Traditional banks are not structured to handle the compliance complexity and operational profile of these sectors, leaving fast-growing deep-tech and defence-adjacent companies underserved. Leonidas replaces that gap with financial infrastructure built specifically for the scale and sensitivity these industries require.

Mecone builds index infrastructure that lets perpetual futures exchanges list contracts on frontier assets, such as fine art, real estate, rare earth metals, and pre-IPO companies, that currently have no derivatives market. Its core methodology aggregates real-time prediction market data to produce continuously updating price indices for these assets. The near-term product is a licensed index; the longer-term plan is a U.S.-based perpetual futures exchange dedicated to frontier assets.

Mirai builds interactive neighborhood tours for real estate listings, letting buyers explore the streets, transit routes, and nearby places around a property before a showing. Agents and developers enter a listing address and Mirai generates a shareable, embeddable neighborhood experience powered by video generation and world models. It replaces the patchwork of maps, PDFs, neighborhood reports, and drive-bys buyers currently stitch together to understand a location.

Motherboard Labs builds agentic field operations software for teams running fleets of robots and autonomous machines in manufacturing, logistics, and mobility. The platform ingests each fleet's machines, service history, rules, and workflows to help operators diagnose failures, resolve issues remotely, and equip technicians with the right context, parts, and procedures before they go on-site. Every field fix is captured as a playbook and fed back to engineering, replacing ad-hoc tribal knowledge and manual escalation chains.

Munari builds a quality-scoring platform for robotics data collection teams. It processes and annotates human demonstration episodes, scores each one against task-specific criteria, and tracks operator performance, replacing manual review and ad hoc feedback. Teams use the output to fix execution and capture problems before the next collection run, getting more usable training data from the same hours of work.

Nexor AI builds infrastructure for deploying AI voice and messaging agents that handle revenue workflows: outbound follow-up, lead qualification, meeting booking, KYC completion, and dormant account reactivation. Companies connect these agents to their existing CRMs, ERPs, and APIs and define goals in plain language, while the agents reason across calls, iMessage, email, and Instagram to push users through the funnel. The product replaces the human labor of persistent, multi-touch outreach across the full sales and post-sales lifecycle.

Nimble builds AI agents for freight forwarders, handling the email and document work that runs day-to-day operations: finding shippers on active trade lanes, pricing and sending quotes, and keeping customers updated on shipments. The agents draft outreach from a forwarder's own rate cards and win history, route enquiries through the salesperson's mailbox for approval, and run around the clock, letting forwarders take on more volume without adding headcount.

OneGC runs legal operations for startups, from contract review and employment matters to equity grants, board consents, and governance filings. The platform combines software that proactively flags and resolves legal issues with embedded attorneys, replacing both outside counsel and the need for an in-house legal hire. Customers get a single team handling day-to-day legal work end to end, accessible via Slack, Zoom, and email.

Ontem builds an accounts-receivable platform for commercial services businesses, handling the full cycle from billing through cash collection. Finance teams today manually piece together contracts, work orders, purchase orders, and customer approval requirements across disconnected tools before they can even send an invoice, then chase payment by hand. Ontem connects those systems into a single AR workflow, surfaces what is blocking each invoice, and moves the work forward automatically.

OpenWhispr builds an open-source, on-device AI notetaker and voice assistant for enterprises where data privacy is non-negotiable. It lets employees dictate notes and capture meetings entirely on-device, replacing cloud-dependent transcription tools that expose sensitive conversations to third-party servers. The target buyer is the privacy-conscious enterprise that cannot accept data leaving its perimeter.

Orbits builds an AI household chief of staff for parents and busy families, replacing the mental and operational load of managing schedules, meals, chores, repairs, and family logistics. Its assistant, Bit, acts proactively: booking appointments, placing orders, making calls, and coordinating across family members and outside helpers like babysitters and contractors before problems arise. Bit works through the channels families already use, including email, calendar, and iMessage group chats.

Oruk trains speech foundation models that decode not just words but tone, emphasis, intent, and emotion directly from audio. Its API gives developers and businesses a richer read on spoken conversation, replacing transcript-only tools that miss the difference between a hesitant answer and a confident one. Contact centers, sales teams, voice agents, and hardware builders use it to act on how something was said, not only what was said.

Othello builds a real-time coaching tool for sales reps that surfaces on-screen prompts during live video and phone calls. It guides reps through discovery questions, technical objections, and relevant case studies in the moment, replacing ad-hoc judgment with a consistent, repeatable process. The target users are AEs, AMs, BDRs, and SDRs whose performance varies call to call.

Parca delivers AI-native pre-construction services to residential real estate developers, producing concept plans, site investigation reports, and feasibility studies. Every output is reviewed by a real estate professional before delivery. The product replaces the slow, expensive work that developers currently outsource to traditional pre-construction consultants during land acquisition.

PhotonSpear builds orbital laser weapons designed to intercept and destroy ballistic missiles during their boost phase, before they can deploy countermeasures. The system uses adaptive optics to maintain beam focus and lethality at the ranges required for space-based operation. Its primary customer is the U.S. government, replacing ground- and sea-based interceptor systems that engage missiles later in flight, when they are harder to kill.

Phyvant deploys fleets of AI agents to hit specific business targets for large enterprises, replacing the cycle of expert-in-the-loop task management where a person breaks down every goal, supplies context, and reviews every result. Customers name a metric and the constraints around it; Phyvant's agents work every case in parallel, around the clock, and keep going until the target moves. The product has been applied to M&A work such as entity structure mapping, a process that previously took weeks of specialist time and six-figure consulting spend.

PreMan builds a self-healing QA platform for software teams that ship quickly. It scans a codebase to find existing tests, charts the end-to-end flows those tests exercise, and produces new test scenarios to fill coverage gaps, replacing the manual work QA leads do to keep test suites current as products evolve. Developers trigger and extend tests through a desktop app or terminal using plain-English commands.

Primitive Labs builds AI systems that model user behavior, including perception, hesitation, confusion, and decision-making, across digital product journeys. Product, growth, and engineering teams use it to surface behavioral risk and improve user flows during development, rather than diagnosing problems through analytics after users have already struggled.

Proper Motion builds continual learning infrastructure for AI agents, letting them improve automatically from real-world interactions rather than requiring manual retraining. Teams in finance, legal, and healthcare use it to deploy agents that get more accurate and reliable over time as they accumulate operational experience. It replaces the static, one-shot deployment model where agents degrade or stall without costly human intervention.

Reelful builds an AI video-editing agent for iPhone that turns a plain-language prompt, photos, and raw footage into a finished, ready-to-share video, handling structure, scripting, motion, captions, music, and voiceover automatically. It targets creators, small business owners, and founders who need to publish video content regularly but lack the time or technical skill to edit. The product replaces a multi-step manual editing workflow with a conversational, prompt-driven process.

Rhema Health builds AI agents that automate preventive care delivery for Medicare patients, targeting specialists in podiatry, endocrinology, and cardiology. Its agents handle multi-touch patient outreach and carry workflows through to billable outcomes, replacing manual care coordination that clinics currently run by hand.

Ripline builds AI agents that handle operational work in supply chain and other critical industries, including orchestrating complex logistics, navigating compliance, and running financial models. The work it replaces is the slow, manual coordination that operators and analysts currently do by hand across these high-stakes physical processes.

rollup builds an SDK that gives personal AI assistants and consumer apps rich user context from day one. On-device AI processes a user's camera roll into structured interest, preference, and experience data without photos leaving the device, replacing the cold-start problem that forces apps to learn users slowly over time. Long-term, rollup is building a portable context layer so that understanding travels with users across every app they choose.

Rumbo builds an AI operating system for freight forwarders, targeting operators on U.S.-Latin America trade lanes. Its agents read incoming emails, PDFs, and screenshots to generate shipment quotes, draft correspondence, track cargo, and verify shipping documents, replacing work that previously consumed the bulk of an operator's day. Human operators review and approve; Rumbo handles execution.

Saru builds AI brand concierges for consumer brands, replacing one-way SMS blast campaigns with two-way, personalized conversations over iMessage, Instagram, and other messaging channels. The agents carry the brand's voice and product catalog, answer real customer questions, and build evolving profiles of each buyer based on what they say, buy, and return for. Brands get a direct line to customer intent without engineering work or a new app.

scaledrive.ai builds an AI software layer that lets autonomous industrial robots operate safely alongside people in dynamic environments like ports, shipyards, and airports. The system predicts human and object movement in real time so robots can reroute without unnecessary stops, replacing the manual safety constraints and operational pauses that cap robot throughput today. Customers include heavy-industry operators who need robots to run at full capacity through shift changes, crew activity, and shifting site layouts.

SenseLab builds continual learning infrastructure for teams running AI agents, so that every completed task becomes a training signal that sharpens the agents over time. It captures shared memory across agents, validates outcomes, and feeds what worked and what failed back into the underlying model, replacing the manual work of debugging, logging, and retraining agent behavior. Engineering and product teams use it to stop agents from repeating the same mistakes across runs.

Sonder builds on-device vision models that watch a user's screen, learn their workflows and preferences, and encode that knowledge into a reusable digital twin. For individual knowledge workers, it captures how work actually gets done, replacing the manual effort of writing documentation, prompting agents with context, or onboarding AI tools from scratch each time. At the team and company level, those learned patterns accumulate into a shared intelligence layer, turning tacit organizational knowledge into something models can act on continuously.

Sotant builds AI-powered managed services for complex supply chain operations at large enterprises, replacing teams of human operators handling exceptions, vendor coordination, and multi-system handoffs. Using imitation learning, it captures how a company's best operators make decisions and encodes that expertise into custom AI agents deployed in weeks. The work it displaces includes manual judgment calls, exception handling, and process coordination that would otherwise require significant headcount.

Space builds an AI-native filesystem that streams files directly from the cloud to any application in real time, replacing downloads, local storage, and sync workflows. Teams, individual users, and AI agents all read and write from one shared drive without duplicating data across devices. The product targets anyone working with large files, such as video assets or CAD models, who currently loses time to transfers and disk management.

Spun Labs builds a credit card designed specifically for creators, who earn royalties and other variable income that traditional underwriting models handle poorly. Proprietary AI underwrites fragmented, delayed income streams to set credit limits and repayment periods in real time, replacing predatory advances and income-share deals that cost creators a portion of future earnings. The product targets musicians and other creators who need immediate access to pay they have already earned but not yet received.

Sterling builds long-horizon revenue agents for go-to-market teams, replacing the manual work of reading signals across usage data, support tickets, and calls. It consolidates a company's GTM data into a single structured layer, encodes that company's own business logic and customer definitions, then deploys bespoke agents that carry context through decisions end-to-end, from account sourcing and renewals to ad targeting and product feedback, re-planning as accounts evolve.

Superintelligence Foundry builds a self-improving AI system that acts as a general scientist, engineer, and inventor. Given a task, it constructs a world model of the domain, designs safe experiments to close gaps in its knowledge, and updates itself from the results, running the same loop in software, in the lab, and in robots. The system replaces the human-led cycle of hypothesis, experiment, and iteration across scientific research and physical engineering.

Switchfrog builds an agent detection and governance layer for web products, letting SaaS businesses see exactly when and how AI browser agents operate inside their users' accounts. Where traditional bot detection misses agents that run inside real browsers with real credentials, Switchfrog identifies the handoff from human to agent, tracks the pages and API calls those agents make, and compiles multi-step workflows into agent-native tools. It replaces manual audit work and blind spots that come from treating every logged-in session as a human session.

Tenera builds AI simulation software that predicts how users will behave in response to product changes before any code is written or shipped. Enterprise product teams use it to test decisions, replacing A/B tests and user research cycles that take weeks with simulated runs completed in hours. The simulations are grounded in real clickstream and historical behavior data, not stated preferences, so teams see which variants drive outcomes before committing engineering time.

Truli builds an AI-native compliance platform for consumer packaged goods brands. It reviews labels, ingredients, claims, and marketing content across the product lifecycle, from formulation through post-launch, flagging issues and recommending fixes. The platform replaces manual regulatory review and tracks enforcement trends so CPG teams can catch problems before they become costly mistakes.

Valent builds a supply chain intelligence platform for Western manufacturers in critical industries, including defense, semiconductors, batteries, and advanced manufacturing. It maps supplier networks to full part and material origin, letting sourcing, compliance, and operations teams identify foreign adversary dependencies and security gaps at scale. The platform replaces manual supply chain audits and fragmented compliance reviews that cost manufacturers significant time and money.

Vela Energy builds AI agents that run the engineering, procurement, and interconnection workflows for utility-scale power projects. It ingests utility studies, electrical drawings, equipment specs, procurement records, and construction schedules into a unified project model, then uses agents to evaluate configurations, match equipment to specs, prepare interconnection submissions, and propagate the impact of any change across documents, costs, and schedules. Developers and EPC teams replace manual cross-discipline coordination with a continuously updated view of what can be energized, when, and under which constraints.

Vene Health builds an AI health assistant for families managing chronic care. It monitors a patient's medical and insurance records, then acts autonomously, booking appointments, chasing referrals, sorting prescription refills, and resolving prior authorizations, over calls and texts, before family members have to notice or ask. The product replaces the coordination work that otherwise falls on a caregiver on top of everything else in their day.

Vound supplies first-person video of skilled workers in factories, farms, construction sites, hotels, and other commercial environments to robotics labs training general-purpose physical AI models. The footage comes with task labels and quality control, replacing work that labs cannot do by scraping or simulation. The same workplace relationships that support data collection become the channel for deploying, testing, and operating robots in those environments.

Zoah builds a visual platform for teams to design, build, and deploy software collaboratively, replacing the hand-off friction between designers and engineers. Instead of translating designs into code separately, product teams work in a single environment that produces stable, production-ready software. The product targets software teams that lose time and fidelity moving work across disconnected design and engineering tools.

Zorbe builds custom AI models for companies across text, vision, speech, image, and video workloads. Using techniques like distillation, fine-tuning, and reinforcement learning, it converts general-purpose frontier models into smaller, task-specific models tailored to how a given business operates. The result replaces reliance on third-party AI APIs with models the company controls, owns, and can improve over time as its data and needs change.