AI application-layer startup radar: July 22–29, 2026

Executive read

The July 22–29 window produced 15 public funding disclosures that fit the application layer: 12 dollar-denominated rounds totaling $238.6 million, plus three euro-denominated rounds totaling €5.4 million. The largest checks went to Freehand ($75 million for autonomous supply-chain spend), AegisAI ($36 million for AI-driven email defense), and Encore AI ($30 million for agents trained on customer interactions). The total is a sum of the current rounds listed below; currencies are kept separate rather than converted at an arbitrary exchange rate.

The clearest cluster was agentic operations. Startups are attaching agents to supply-chain procurement, customer calls, freight forwarding, browser workflows, and car-rental revenue management. A second group is turning difficult datasets into vertical software: clinical-trial imaging, waste streams, e-commerce product discovery, and education. The common diligence question is where the company owns a workflow, data asset, or operating relationship that a general-purpose model vendor cannot copy by adding a feature.

Quick scan

Company	Sector	Window signal	Product positioning	Founding-team background
Freehand	AI agents / supply chain	$75M Series B; July 29	Autonomous agents for complex supply-chain spend and back-office operations	Nitin Jayakrishnan and Abhijeet Manohar previously co-founded Pando, an enterprise logistics software company 1
AegisAI	AI agents / security	$36M Series A led by Battery Ventures; July 23	Agents that detect AI-crafted spear-phishing emails	Cy Khormaee and Ryan Luo are former Google security executives who worked on safe browsing and reCAPTCHA 2
Encore AI	AI agents / customer operations	$30M Series A led by Team8; July 29	Conversation analysis that turns successful customer interactions into agent playbooks	CEO and founder Dvir Ginzburg started the company as Insait IO in 2022, first building recommendation software for financial advisers 3
Greyparrot	Vertical SaaS / waste	$27M Series B led by Omar Mir; July 28	Computer-vision waste intelligence for sorting, recovery, and compliance	Mikela Druckman, Ambarish Mitra, and Nikola Sivacki founded the company; prior affiliations were not disclosed in the opened source 4
Qureight	Vertical SaaS / clinical trials	$20M Series B led by Molten Ventures; July 29	Imaging analytics, disease models, and precision endpoints for clinical trials	Dr Muhunthan Thillai is co-founder and CEO; the opened announcement does not provide a fuller founding-team history 5
telli	AI agents / customer operations	$15M seed led by redalpine; July 23	Charlie builds, deploys, and improves customer-facing agents across voice, chat, SMS, WhatsApp, and email	The funding announcement does not name the founders; the team cites experience scaling Enpal's heat-pump business 6
Hint	Consumer AI / home management	$10M backed by Montauk Capital and others; July 29	AI assistant for maintenance, energy, documents, insurance claims, and home questions	Co-founders include Martha Stewart, Kyle Rush, and Yih-Han Ma; Rush previously worked at Casper and Maisonette and built campaign technology for Obama and Hillary Clinton, while Ma was an SVP and GM at Red Ventures 7
Pangram	Consumer / trust and safety	$9M led by Menlo Ventures; July 29	Detection of AI-assisted text and AI-generated images for consumers, publishers, schools, and APIs	Max Spero and Bradley Emi are Stanford AI and machine-learning graduates who launched Pangram about two years ago 8
Polar	AI agents / knowledge work	$5.7M seed led by Madrona; July 29	AI-first browser for scheduling workflows and assigning agents to logged-in web tasks	CEO Kevin Jiang previously worked at Perplexity on Comet; other founding-team affiliations were not disclosed in the opened source 9
Imagi	Vertical SaaS / education	$4.5M seed; July 23	Guardrailed coding and AI-literacy tools for K–12 students and teachers	Dora Palfi and Beatrice Ionascu launched Imagi in 2018; the opened announcement does not give earlier employer affiliations 10
5U AI	AI agents / freight	$3.2M pre-seed led by Emerge Capital; July 28	Digital workers for quoting, bookings, tracking, invoice reconciliation, and freight administration	Yagiz Abik and Fehmi Şener are Technical University of Munich graduates who founded the company in 2025 11
Scape	AI agents / productivity	$3.2M from Y Combinator, General Catalyst, and FundersClub; July 24	AI-native inbox that drafts replies, creates documents, and prioritizes email	Melvin Hagberg founded the company during Y Combinator's 2024 summer batch; Elis Hodzic joined as co-founder in 2025 after first backing it as an investor 12
Beelzebub	AI agents / security	€3M seed led by United Ventures; July 27	AI-powered hacker traps and an automated analyst for assumed-breach defense	Mario Candela started the work as a Milan research project in 2021 and formally founded Beelzebub in 2025; further affiliations were not disclosed 13
PageMind	Vertical SaaS / e-commerce	€1.2M led by 4Founders Capital; July 23	Product descriptions, guides, FAQs, and analytics optimized for AI-mediated product discovery	Jaume Portell is founder and CEO; the opened source does not provide earlier affiliations 14
Sigvi	Vertical SaaS / mobility	€1.2M pre-seed led by Superhero Capital; July 29	AI revenue agent for pricing, booking, vehicle access, checks, and maintenance in car rental	Founded in 2023 by Vytis Šliažas, Ignas Gibas, and Mindaugas Banaitis; further prior affiliations were not disclosed 15

Funding profiles

Freehand puts agents inside procurement, not beside it

Freehand raised a $75 million Series B co-led by Battery Ventures and NewRoad Capital Partners, with Nexus Venture Partners and Penny Pritzker participating. Its agents manage non-standard supply-chain spend and back-office work inside existing enterprise systems, including invoice checks, contract and policy review, and supplier negotiation. The company was founded by Nitin Jayakrishnan and Abhijeet Manohar after their earlier company, Pando, was acquired in 2026. 1

This is a familiar wedge with a better-defined buyer than a general-purpose agent. The diligence test is whether Freehand can turn its access to procurement systems and exception handling into repeatable software, rather than a series of high-touch deployments.

AegisAI is selling judgment at the email boundary

AegisAI raised a $36 million Series A led by Battery Ventures, with Accel and Foundation Capital participating. Its agents inspect email context and anomalies to catch spear-phishing attempts, including malicious PDF attachments that appear legitimate to rule-based filters. The company says dozens of customers, including Mesh, LangChain, and Lokker, have adopted the product. 2

The product is application-layer security because the value sits in the investigation and enforcement workflow, not in a new foundation model. The open question is whether its Google security pedigree translates into durable telemetry and a lower false-positive rate than the vendors it is trying to replace.

Encore and telli take different routes to the same customer interaction

Encore AI raised a $30 million Series A led by Team8, with Planven, Lukatz, Garage, and financial institutions participating. It mines calls, emails, text messages, and CRM data to identify the playbooks behind successful outcomes, then uses them to train voice and text agents. Encore says it has more than 40 enterprise customers, mostly financial institutions, and that ARR has grown more than fivefold since its seed round; those figures are company-reported. 3

telli is earlier and broader at the channel layer. Its Charlie product helps B2C companies build, test, deploy, and improve agents across voice, chat, SMS, WhatsApp, and email. The company says its agents have handled millions of conversations for customers including Sky, Enpal, and Vaillant; that is also company-reported. 6

The comparison matters. Encore is betting that interaction history is the proprietary training asset; telli is betting that the customer-operations control plane should span every channel. Both have to prove that the agent improves conversion, resolution, or retention rather than only reducing labor cost.

Qureight and Greyparrot monetize hard-to-collect data

Qureight secured a $20 million Series B led by Molten Ventures. Its end-to-end clinical-trial imaging platform combines enterprise imaging services, AI lung biomarkers, synthetic control arms, and a 3D chest-imaging foundation model. The company says the platform is already used by biopharma partners and will expand into additional lung and heart conditions. 5

Greyparrot raised $27 million in Series B led by technology investor Omar Mir. Its camera systems monitor waste on sorting lines, while its Deepnest product helps brands understand packaging after disposal. Greyparrot says the platform has detected more than one trillion waste objects and is used by operators and brands including Veolia, Unilever, L'Oréal, and Kenvue. These are company-reported deployment and volume figures. 4

Neither business is just a model wrapped in a dashboard. The asset is a difficult stream of domain data tied to a regulated or operational decision. The diligence burden is correspondingly specific: data rights, measurement error, deployment economics, and evidence that the output changes a customer's decision.

Hint turns home maintenance into a consumer system of record

Hint emerged with $10 million from Montauk Capital, Brian Kelly, Slow Ventures, Tusk Venture Partners, Energy Impact Partners, Amplo VC, and Hannah Grey. The app combines property records, maintenance schedules, energy management, insurance claims, document storage, and question answering. Co-founders include Martha Stewart, Kyle Rush, and Yih-Han Ma. 7

The consumer-AI angle is more concrete than a generic home chatbot: Hint is trying to accumulate the documents and recurring tasks that make a property legible over time. The risk is distribution. Homeowners may value the information, but the product still needs a reason to be opened before a repair, claim, or purchase creates urgency.

Scape and Polar are competing for the logged-in web

Scape emerged from stealth with $3.2 million from Y Combinator, General Catalyst, and FundersClub. Its inbox drafts responses, creates or edits attachments, and hides low-priority messages after learning from email and meeting context. The company says it had five people in Stockholm and was opening early access at scape.app. 12

Polar raised a $5.7 million seed led by Madrona. It packages browser-based workflows for knowledge workers, allowing users to schedule tasks and assign agents to work across logged-in tabs. CEO Kevin Jiang came from Perplexity's Comet team. 9

Both products make the browser or inbox the control surface. The diligence question is whether users grant enough permissions for automation to become habitual, while still trusting the product with sensitive email, documents, and authenticated web sessions.

Product-launch checkpoints

Product Hunt's daily leaderboards supplied positioning signals from July 23–29. These are launch snapshots, not evidence of funding, adoption, or complete founding histories. The repeated themes were agent control, voice interfaces, and vertical workflow packaging.

Date	Product	Launch positioning
July 23	Teable 3.0; Quaso	AI spreadsheet for business; an automation agent that works across apps and browsers 16
July 24	Pushary; Fluree AI	Approve AI requests from a lock screen; give every AI agent trusted context 17
July 25	OpenComputer; Health in ChatGPT	Deploy a managed agent; a personal health companion powered by ChatGPT 18
July 26	Athena by Shoplazza; Yoggi	An orchestrator agent for a commerce stack; safe AI chat for kids 19
July 27	Estera; Comms	AI receptionist for calls and WhatsApp; launch iMessage agents in seconds 20
July 28	Cekura; Conduit	A self-improvement loop for voice agents; AI agents purpose-built for hospitality 21
July 29	Prelint; Vela	Prevent product drift in AI-written code; an AI recruiting coordinator 22

What changed this week

1. Agent software is moving from answering to operating

The funded companies are increasingly responsible for actions: negotiating supply-chain spend, reconciling freight invoices, booking appointments, defending email, learning from calls, or executing work in authenticated browser sessions. The interface is still often a chat box, but the product boundary is the system that lets an agent act safely and prove what it did.

That makes permissions, exception handling, audit trails, and liability more important than a generic autonomy claim. The Product Hunt snapshots reinforce the same direction, with lock-screen approvals, agent context, managed deployment, and agent self-improvement appearing as standalone product positions.

2. Vertical AI is being built around evidence, not only prompts

Qureight works with clinical images, Greyparrot with waste streams, PageMind with product catalogs and search intent, 5U AI with freight decisions, and Sigvi with vehicle operations. In each case, the model is one part of a larger operating system. The harder question is whether the company has the right to collect the data, the workflow to act on it, and the customer relationship to keep improving it.

This is a healthier diligence frame than asking whether a startup has a proprietary model. A domain-specific data loop can be defensible, but only if the output is used in a decision that someone pays to improve.

3. Founder evidence is strongest where the workflow is hardest

The public record is unusually clear for several teams: Freehand's founders previously built Pando, AegisAI's founders worked on Google security, Encore began as a financial-adviser recommendation company, 5U AI's founders are Munich technical-university graduates, and Hint combines consumer-brand reach with software operators. Other entries disclose only names or a founding date. That is a limitation on public diligence, not a negative signal about the company.

Diligence questions

For agentic operations companies, which permissions and exception cases are handled in software rather than by a services team?
For security products, what is the false-positive rate against a current incumbent, and who carries liability when an agent blocks a legitimate action?
For vertical AI, is the data licensed, customer-contributed, or collected through the workflow? Can the company show that the output changes a measurable decision?
For browser and inbox agents, how much access do users grant, how often do they return, and what happens when a website changes its interface?
For company-reported customer, ARR, or volume figures, what is the denominator and which figures have independent confirmation?
For teams with limited public background, which founder and operator claims still need to be checked in a data room or reference call?

Coverage note

This radar covers public funding and product disclosures published from July 22 through July 29, 2026, with a cutoff of 5:00 PM Pacific Time on July 29. Each funding entry was included only when the opened detail source provided a date inside the window. Model-layer companies such as Fish Audio, robotics and physical-AI infrastructure, and reported or prospective raises without a confirmed close were excluded. Product Hunt entries are launch-positioning checkpoints only; they do not establish financing, traction, or complete founding histories.

Where an opened source did not disclose prior affiliations or a complete founding team, the gap is stated rather than filled from an unverified database.