All analysis was generated autonomously, without human review. Scores are analytical opinions drawn from the cited public sources, without hands-on testing. They are not audits, certifications, investment reports, purchasing advice, or evaluations of quality.
HydroX AI sells AI security software to organizations running large language models, and it also sells child-safety software to families and schools and a private meeting assistant. Its blog carries no post after October 8, 2025, no repository push appears after April 2025, and the reviewed sources disclose no financing after the June 2024 angel round of $4 million. HydroX announced in October 2024 that it would help stress-test Anthropic's models under that company's bug-bounty program, and the online-safety nonprofit ROOST lists it among its partners today. No reviewed source names a paying customer, so a buyer has nothing recent to check. Watch whether HydroX publishes a named enterprise customer.
| Description | HydroX AI sells an AI security and safety platform spanning a runtime firewall that blocks jailbreaks and prompt injection, model safety evaluation and red teaming, adversarial testing of customer chatbots, a child safety product, and a local-first AI meeting assistant. | [f1] |
|---|---|---|
| Founded | 2023 | [f2] |
| HQ | San Jose, California, USA | [f3] |
| Latest funding | $4M angel round (June 2024) | [f2] |
| Product | What it does |
|---|---|
| HydroX Firewall | Model-agnostic runtime firewall for LLMs, agents, and AI applications that blocks jailbreaks, prompt injection, and data leaks in real time, listed on Google Cloud Marketplace. |
| EPASS | Evaluation platform for AI safety and security that stress-tests models against adversarial attacks and publishes a public leaderboard comparing model security, privacy, and integrity. |
| Security Analysis | Adversarial security testing for customer-facing chatbots that runs controlled jailbreak, prompt-injection, and coercive conversation tests and returns security insights. |
| Stay Aware AI | Child safety platform for families and schools that uses AI to turn children's conversations and online activity into emotional and behavioral insights and real-time guidance. |
| CaptainNote | Offline-first AI meeting assistant that records, transcribes, and summarizes meetings locally using open-source models, keeping data on the user's device. |
| AI-Driven Scraper | Red-teaming service for defining and validating anti-scraping defenses with AI-agent-driven scraping over 60+ residential IP addresses and LLM-based results processing. |
AI Defense Matrix
| Govern | Identify | Protect | Detect | Respond | Recover | |
|---|---|---|---|---|---|---|
| AI-Workload Platforms Inference servers, training platforms, vector DB platforms, and the model-loading supply chain. | ||||||
| AI Orchestration Tools Agentic orchestration tools, plus their plugins, skills, hooks, system prompts, scaffolding, harnesses, configuration settings, and MCP clients on user devices. | ||||||
| AI-Generated Code Code produced by AI tools, AI-assisted reviews, AI-generated infrastructure-as-code and tests, and vibe-coded apps that bypass CI/CD. | ||||||
| AI Gateways & Routers MCP proxies and gateways, LLM routers, outbound AI-service traffic, shadow AI egress, and model-registry traffic. | ||||||
| AI Model Model weights, fine-tuning checkpoints, model cards, registries, AIBOM, and the third-party LLMs your enterprise consumes. | ||||||
| Training Data Datasets used for training, fine-tuning, and continued learning. | ||||||
| Runtime AI Data User prompts, inference inputs, RAG content, vector DB content, persistent agent memory, and interaction history. | ||||||
| AI Agent Identities AI agents as non-human principals, plus credentials, keys, permission scopes, service accounts, and delegation chains across agents and tools. |
HydroX Firewall intercepts AI inputs and outputs to block jailbreaks and prompt injection and to redact sensitive data from responses, and EPASS red-teams and stress-tests models to uncover vulnerabilities. These capabilities are mapped to the AI Defense Matrix. [f4]
How well the company can compete in its security market, scored across eight dimensions against public evidence.
| Dimension | Score | Rationale |
|---|---|---|
| Problem Clarity How precisely the company defines its problem, with evidence the problem exists at the scale claimed. | 3/5 | HydroX AI names its buyers and its problem concretely, addressing AI and machine-learning developers, AI security teams, and business and risk decision makers who face jailbreaks, prompt injection, toxic output, and data leakage from deployed models. The pain is stated on the vendor's own pages and is generic to the category, and no reviewed source quantifies it for these buyers. [s1, s3, s6] |
| Capability Depth How specific the technical capabilities are, with evidence beyond marketing claims such as docs and third-party validation. | 3/5 | The product page and the Marketplace announcement carry concrete capability detail, covering interception under 100 milliseconds, more than 30 risk types, more than 20 jailbreak scenarios, and redaction of personal data, and the leaderboard ranks 20 models on page one of five against 20 named attack methods plus a no-attack baseline. What a buyer cannot do is inspect it: the Documentation button leads nowhere reachable, hydrox.ai/docs returns 404, docs.hydrox.ai does not resolve, and no independent technical evaluation appears in the reviewed sources. [s3, s6, s4, s20, s24] |
| Market Timing Whether the market is ready for this product, with evidence that buyers are actively seeking solutions. | 3/5 | The enabler is the enterprise move to large language models after the company's 2023 founding, with model builders organizing outside testing around it, shown by the Anthropic bug-bounty role announced in October 2024 and the AI Alliance evaluation work reported in 2024. Buyer-side demand for HydroX AI's own products stays indirect in the reviewed sources, which document ecosystem programs and a marketplace listing rather than budget movement, and none of it is dated within the last twelve months. [s7, s13, s10, s6] |
| Team Credibility Demonstrated domain expertise with public signals such as prior exits, publications, and industry recognition. | 3/5 | Founder Zhuo Li ran the Privacy and Data Protection Office at ByteDance after work at Meta and LinkedIn, senior in-domain roles carried by a wire release and by his own account on the AI Alliance site, and the reviewed sources name no product he previously built and no sustained publication record. The current team page lists five people, with Li holding the chief executive and chief architect titles together, and neither the head of engineering named in the 2024 funding announcement nor the chief operating officer who fronted the 2025 announcements appears on it. [s7, s2, s16, s25] |
| GTM Proof Evidence of actual traction (customers, revenue signals, partnerships) beyond stated intentions. | 3/5 | A Google Cloud Marketplace listing the company announced in July 2025 sits alongside partnerships that non-vendor sources corroborate: ROOST lists HydroX AI among its partners, AI Business reported the AI Alliance model-evaluation work with Meta and IBM, and The American Bazaar reported the RSAC 2025 red-teaming session HydroX would host with Humane Intelligence. No reviewed source names a paying customer or any revenue, so commercial scale is uncorroborated. [s6, s12, s10, s23] |
| Funding Efficiency Whether funding matches go-to-market ambition, with signs of capital-efficient growth. | 3/5 | The $4 million angel round of June 2024 from Vitalbridge Capital, Atom Capital, and Qi Lu is proportional to a seed-stage motion, and shipping followed it. AI Business reported independently that the company built an evaluation platform businesses use to test their language models for safety and security, and the firewall reached Google Cloud Marketplace in July 2025. The round is now more than two years old with no follow-on on the record, and no reviewed source discloses revenue, margin, or customer growth, so output per dollar stays unconfirmed. [s7, s10, s6] |
| Category Clarity Whether the company creates or fits a recognizable category that buyers can quickly place in their stack. | 2/5 | The firewall and the model-evaluation work fit the forming AI-security category, and the same brand also sells a child-safety product to families and schools, an offline meeting assistant, and an anti-scraping red-team service, while the homepage offers to architect and build a customer's AI product outright. A buyer cannot say which budget line HydroX AI fills without the company explaining it. [s3, s1] |
| Incumbent Defensibility How vulnerable the core value proposition is to absorption as a feature by a platform vendor. | 3/5 | The firewall's inline position on live AI traffic and its Google Cloud Marketplace deployment channel put some friction in the way of absorption. The published differentiators are feature-level, covering real-time interception, redaction of personal data, and jailbreak-pattern coverage, with no accumulated data asset or certification behind them in the reviewed sources, so none of it is a structural barrier. [s6, s3, s19] |
HydroX AI sells against the risks organizations take on when they run large language models in production: jailbreaks, prompt injection, toxic or unsafe output, and leakage of sensitive data. The homepage frames the buyer broadly, naming organizations deploying generative AI and also children, families, and schools, and it names compliance with evolving AI regulation as the reason to act now.
The demand evidence is strongest on the model-builder side. HydroX announced in October 2024 that it would help stress-test Anthropic's models under that company's bug-bounty program, and AI Business reported in July 2024 that the company was evaluating generative AI models alongside Meta and IBM through the AI Alliance. It is thinnest on the enterprise-deployer side, where the company sells its products and the reviewed sources quantify nothing about the buyer's pain. [s1, s3, s13, s10]
The security platform has three pillars. A runtime firewall intercepts model inputs and outputs to block jailbreaks and prompt injection and to redact personally identifiable information, which the vendor describes as covering more than 30 risk types and more than 20 jailbreak scenarios in under 100 milliseconds. An evaluation service stress-tests models before and after deployment, and a Security Analysis service runs controlled jailbreak, prompt-injection, and coercive conversation tests against a customer's chatbots and returns security insights.
The public leaderboard ranks 20 models on page one of five, scored against 20 named attack methods alongside a no-attack baseline, and two of those 20 are HydroX's own models. On GitHub the organization holds 26 public repositories, 20 of them forks of other projects including implementations of published jailbreak methods such as TAP, and its own PII Masker carries 178 stars. No repository in that list has been pushed since April 2025.
The portfolio extends well past enterprise AI security, adding Stay Aware AI for families and schools, the offline meeting assistant CaptainNote, and an anti-scraping red-team service the vendor says operates more than 60 residential IP addresses. Documentation is the gap: the product page's Documentation button leads nowhere reachable, hydrox.ai/docs returns 404, and docs.hydrox.ai does not resolve, so a buyer cannot read how any of it works. [s3, s6, s4, s17, s21, s20, s24]
HydroX AI competes in the runtime-guardrail and AI red-teaming space. The comparable vendors named on this page are selected here as analyst comparisons rather than from any cited market study, and the reviewed sources establish no cross-vendor feature baseline, so nothing here states what rivals do or do not ship. The firewall's pitch, model-agnostic, one line of code, sub-second interception, is the vendor's own description.
What the record does show is who works with the company. HydroX announced the Anthropic bug-bounty role in October 2024 and a Google Cloud Marketplace listing in July 2025, says it graduated from the Google Cloud AI Accelerator in June 2025 as one of 15 startups, and appears today among the partners the online-safety nonprofit ROOST publishes. The consumer-facing lines cut against a single category story, because the same brand also sells to families and schools. [s6, s13, s14, s12, s3]
The visible go-to-market motion is partnership-led. HydroX announced in October 2024 that it would help stress-test Anthropic's models under that company's bug-bounty program. AI Business reported in 2024 that it was evaluating models alongside Meta and IBM through the AI Alliance, HydroX says Google chose it as one of 15 startups for the 2025 Cloud AI Accelerator, and the company announced its Firewall on Google Cloud Marketplace on July 1, 2025. The American Bazaar reported in April 2025 that HydroX and Humane Intelligence would host a two-hour red-teaming session at RSAC 2025.
Commercial proof has not followed. No reviewed source names a paying customer or any revenue, and disclosed funding remains the $4 million angel round of June 2024 from Vitalbridge Capital, Atom Capital, and Qi Lu. The homepage does carry a ticker headed Trusted by the World's Top Enterprises, and the page names no company in its text, so a reader cannot tell whether any of them buys anything.
The company's own announcement channel has been silent. Its blog index carries no post newer than October 8, 2025, nearly eleven months without a new proof point, and its GitHub organization has pushed no repository since April 2025. [s13, s10, s14, s6, s23, s7, s1, s5, s21]
Founder and CEO Zhuo Li came up through privacy engineering, running ByteDance's Privacy and Data Protection Office after work at Meta and LinkedIn. He describes himself on the AI Alliance site as a cybersecurity veteran of more than 15 years, the same background the 2024 funding announcement carries, and both accounts trace to the company rather than to an independent check.
The team page now lists five people, and Li holds the chief executive and chief architect titles together. The head of engineering named in the 2024 funding announcement does not appear on it, nor does the chief operating officer who led the 2025 RSAC session and was quoted in the Marketplace announcement. The chief information security officer is listed only as Bamboo, with a description and no surname. [s2, s7, s25, s16, s6]
No SOC 2, ISO 27001, or comparable attestation appears in the reviewed sources or on the probed trust surfaces. For a company whose homepage promises to enforce safety, governance, and compliance for customers, that gap means nothing in the compliance record stands between HydroX AI and a replacement.
The trust story the company does tell is engineering-led. CaptainNote processes meetings locally so recordings do not leave the device, the firewall redacts sensitive data in real time, and some tooling is open-sourced for outsiders to inspect. Those are self-described product behaviors, and no third party has attested to them in the reviewed sources. [s18, s1, s3, s17]
| Company | Relationship | Note | Compare |
|---|---|---|---|
| Lakera | competes with | Competes for the same buyer, an organization that wants runtime guardrails on its language-model traffic. | N/AWe captured the evidence for these companies under different evidence-model versions (v1 vs v2), so the totals were scored under different conditions and are not directly comparable. |
| Prompt Security | competes with | Competes for the same enterprise buyer of runtime protection for generative AI use. | N/AWe captured the evidence for these companies under different evidence-model versions (v1 vs v2), so the totals were scored under different conditions and are not directly comparable. |
| CalypsoAI | competes with | Competes on the same pairing of model evaluation with runtime guardrails. | N/AWe scored these companies at different scopes, so the totals measure different things. |
| Guardion AI | competes with | Competes for the same inline gateway position on language-model and agent traffic. | N/AWe captured the evidence for these companies under different evidence-model versions (v1 vs v2), so the totals were scored under different conditions and are not directly comparable. |
| General Analysis | competes with | Competes on the same pairing of adversarial testing with runtime protection. | N/AWe captured the evidence for these companies under different evidence-model versions (v1 vs v2), so the totals were scored under different conditions and are not directly comparable. |
Add analyzed competitors to compare them side by side with HydroX AI.
A closer look at the company's product strategy, measuring how defensible it is against market forces and examining the eight areas behind it.
reinforce or reposition
The HydroX Firewall runs inline on live AI traffic, the closest thing to lock-in the record shows. The company's own announcement says integration takes one line of code with minimal engineering overhead, and nothing cited sizes what leaving would cost. HydroX now sells expertise alongside software, offering to architect and build a customer's AI product and to run adversarial tests against a customer's chatbots. The data and model assets the record names are published: an open-source detector for personal data, 13 models on a public hub, and a public leaderboard. No security attestation appears in the reviewed sources, so nothing in the compliance record stands between the company and a replacement. Adversarial testing skill is a head start here, not yet a durable lead.
| Dimension | Score | Rationale |
|---|---|---|
| Value Delivery Does the product sell software as the product, or judgment, trust, or accountability with software as the delivery mechanism. | 2/3 | HydroX AI sells software the customer wires into its AI stack and also sells the expertise around it, with the homepage offering to architect, build, and deploy a customer's AI product and the Security Analysis service running adversarial tests against a customer's chatbots and returning security insights. Code and expertise blend here, and the reviewed sources show no layer that takes accountability for the customer's outcomes. |
| Switching Cost How expensive leaving is for a customer: data portability, integrations, learned workflows, network effects, regulatory data residency. | 2/3 | The firewall runs inline on live AI traffic, and the vendor describes configurable mitigation actions and visual dashboards layered on top of it, which is meaningful integration friction. The cited record does not size the migration, and the company's own announcement says integration takes one line of code with minimal engineering overhead, so nothing documents what an exit would cost in effort. |
| Compliance Moat Whether certifications, liability acceptance, or audit trails block an easy replacement. | 1/3 | No SOC 2, ISO 27001, or comparable attestation appears in the reviewed sources or on the probed trust surfaces, so a funded competitor faces no compliance requirement it would have to obtain before replacing HydroX AI. |
| Problem Complexity Whether the product requires ML, optimization, real-time systems, or years of specialized expertise. | 3/3 | Filtering adversarial prompts inline at the vendor's claimed sub-100-millisecond latency is real-time systems work, and the company publishes 13 models on a public model hub, including a detector for personal data built on DeBERTa-v3, which is machine-learning engineering rather than integration. Its leaderboard scores models against 20 named attack methods, the kind of adversarial evaluation that takes specialized expertise to build. |
| Buyer Profile Whether buyers are SMB operators, mid-market IT teams, or regulated enterprises and governments with procurement gates. | 2/3 | The firewall and evaluation products address enterprise AI and security teams, a budget-holding buyer with some governance, and the reviewed sources name no customer at all, so that buyer class is addressed rather than evidenced. The child-safety and meeting-assistant lines pull toward families and individuals. |
| Layer Whether the product is an end-user application, a platform with application features, or infrastructure other applications depend on. | 2/3 | The firewall sits between applications and the models they call, a real traffic position, and the catalog records it as covering runtime AI data as a primary asset. Evaluation, adversarial testing, and the consumer products do not hold that position, so most of what HydroX AI sells is not something other applications depend on. |
| Proprietary Data, Content, or IP Whether the product accumulates datasets, content licenses, or IP that a rival cannot recreate from scratch. | 1/3 | Every data and model asset the record names is published: PII Masker is open source, the 13 models it publishes sit on a public model hub, and the safety leaderboard is readable by anyone. The reviewed sources name no retained corpus and state nothing about what such a corpus would accumulate from, and 20 of its 26 repositories are forks of other projects. |
HydroX AI addresses at least three distinct buyer groups at once. The homepage speaks to AI and machine-learning developers, AI security teams, and business and risk decision makers for the firewall and evaluation products, while the product page's Stay Aware AI targets families and schools and CaptainNote targets people who want private, on-device meeting notes.
Serving enterprise security buyers and consumer households from one seed-stage company is a segmentation choice the reviewed sources never explain. Each group hears a different promise from the same brand, and the public team page lists five people, one of whom carries revenue and partnerships. The homepage lead, protecting children and organizations in one sentence, captures the tension.
The core technical claims are concrete. The firewall intercepts model inputs and outputs in real time to block jailbreaks and adversarial prompts, with the vendor citing interception in under 100 milliseconds across more than 30 risk types and more than 20 jailbreak scenarios, and the evaluation service stress-tests models before and after deployment. The public leaderboard ranks 20 models on page one of five, scored against 20 named attack methods alongside a no-attack baseline, two of those 20 being HydroX's own models.
The engineering evidence sits on GitHub and Hugging Face. The organization holds 26 public repositories, 20 of them forks of other projects including implementations of published jailbreak methods such as TAP, and the model hub holds 13 public models. PII Masker, a detector for personal data the company open-sourced, carries 178 stars.
Two limits bound all of it. No repository in the org listing has been pushed since April 2025, so the published engineering record runs no later than that month. The product page's Documentation button leads nowhere reachable, hydrox.ai/docs returns 404, docs.hydrox.ai does not resolve, and no independent technical evaluation of the products appears in the reviewed sources.
HydroX AI's visible go-to-market motion runs through the AI ecosystem. It announced in October 2024 that it would help stress-test Anthropic's models under that company's bug-bounty program, AI Business reported in 2024 that the AI Alliance work placed it beside Meta and IBM evaluating models for industries such as health care and finance, it says Google chose it for the 2025 Cloud AI Accelerator, and it announced Marketplace availability for the Firewall in July 2025. ROOST lists the company among its partners today.
That motion has produced visibility without published commercial proof. The reviewed sources name no paying customer and no revenue, and they show no funding event after the June 2024 angel round. The homepage carries a ticker headed Trusted by the World's Top Enterprises and names no company in its text.
The company's own announcement channel has gone quiet. The blog index carries nothing newer than October 8, 2025, so nearly eleven months have passed with no new proof point on the channel that carried its announcements steadily through 2024 and 2025.
No pricing appears on the pages reviewed here. Their calls to action are contact forms and a book-a-demo flow. The Google Cloud Marketplace listing the company announced adds what that announcement calls fast procurement and billing for enterprises that buy that way.
Hidden pricing at this stage reveals little by itself. Combined with the absence of any named customer in the reviewed sources, it means the public record offers a buyer no anchor for what the products cost or what unit the company charges by.
The firewall is delivered as a service the customer wires into AI applications, described as model-agnostic and integrating with one line of code with minimal engineering overhead, and the AI Defense Matrix catalog records the product's deployment model as SaaS. Evaluation and red-teaming are described as stress-testing a customer's models before and after deployment.
The homepage now also offers delivery by engagement. One of its two streams has HydroX architecting, building, and deploying the customer's AI product outright, and the Security Analysis service runs adversarial tests against a customer's chatbots and hands back security insights. That is expertise sold with the software rather than software alone.
CaptainNote inverts the delivery model again, running locally on open-source models so meeting data does not leave the device. Operating an inline latency-sensitive filter, an evaluation platform, a testing service, a consumer app, and a scraping service is a wide operational footprint for the five-person roster the about page shows.
No SOC 2, ISO 27001, or comparable attestation appears in the reviewed sources or on the probed trust surfaces. For a company whose homepage promises to enforce safety, governance, and compliance for customers, and whose child-safety product ingests children's conversations and online activity, nothing in the compliance record stands between HydroX AI and a replacement.
The trust story the company does tell is engineering-led, covering local-first processing in CaptainNote, real-time redaction of sensitive data in the firewall, and open-sourced tooling outsiders can inspect. Those are self-described product behaviors, and none of them is an audited attestation.
Ecosystem position is the company's strongest card. It sits inside the AI Alliance with Meta and IBM, announced an Anthropic bug-bounty role for jailbreak stress-testing in October 2024, says it graduated from Google's accelerator in June 2025, announced Marketplace availability in July 2025, announced ROOST sponsorship and tool pilots in March 2025, and runs a public community with capture-the-flag challenges alongside an open GitHub presence.
ROOST's own site lists Hydrox AI among its partners today, which is confirmation from outside the company. None of these relationships is evidenced as exclusive or revenue-bearing in the reviewed sources, and each partner could end or internalize the work.
Founder and CEO Zhuo Li came up through privacy engineering, running ByteDance's Privacy and Data Protection Office after work at Meta and LinkedIn. On the AI Alliance site he describes himself as a cybersecurity veteran of more than 15 years, and both that account and the 2024 funding announcement trace to the company.
The current team page lists five people, Li now holds the chief architect title alongside chief executive, and the head of engineering named in the 2024 funding announcement is absent, as is the chief operating officer who led the RSAC 2025 session and was quoted in the Marketplace announcement. The chief information security officer is listed only as Bamboo, with a description and no surname.
The public roster names five people against six product lines spanning enterprise, consumer, and family buyers, the reviewed sources state no headcount, and they do not explain the omissions.
| Id | Source | Tier | Accessed |
|---|---|---|---|
| f1 | HydroX AI product page | official | 2026-08-30 |
| f2 | PR Newswire: HydroX AI Announces Closing of $4 Million Angel Funding Round and Launches EPASS | press | 2026-08-30 |
| f3 | HydroX AI homepage footer | official | 2026-08-30 |
| f4 | HydroX AI (AI Defense Matrix Catalog), mapping aligned to the catalog | other | 2026-08-30 |
This site is an experimental research aid created by Zeltser Security Corp. All its data gathering and analysis was performed autonomously without human review, and it can contain errors of fact, interpretation, and judgment that a human reviewer might catch.
The analyses are statements of opinion, not statements of fact. Machine analysis produced the scores, summaries, and matrix placements by weighing the public sources each page cites, and reasonable people can weigh the same sources differently. Where a page states a fact, it cites the public source and the date it was checked, and the statement is only as accurate as that source. Unless a profile expressly says otherwise, the analysis involves no hands-on testing and no independent validation of any company's products or services.
Nothing here is professional, security, legal, financial, investment, or purchasing advice, and nothing here is a recommendation to invest in, do business with, or avoid any company. Inclusion of a company is not an endorsement, and absence of a company is not a judgment about it. Reading this site creates no advisory or client relationship. Verify any detail you plan to act on against the vendor's current materials.
The content is provided "as is" and "as available," with all warranties disclaimed, express or implied, including merchantability, fitness for a particular purpose, accuracy, and non-infringement. No entry is warranted to be complete, current, or correct. Companies change, vendors update their claims, sources can be wrong, and automated analysis can misread them.
To the fullest extent permitted by law, the operator, Zeltser Security Corp, is not liable for any damages that arise from using this site or relying on its content, including direct, indirect, incidental, special, and consequential damages and lost profits, even if advised that such damages were possible. If you are dissatisfied with the site or disagree with these terms, your remedy is to stop using it.
Entries link to vendor pages, press coverage, and other external sites that Zeltser Security Corp does not control and is not responsible for. A link is not an affiliation with the destination or an endorsement of it. Product and company names and trademarks are the property of their owners, used here nominatively to identify the companies described. Short quotations from cited sources appear for identification and commentary.
Use, quotation, automated retrieval, and redistribution of the content are governed by the Terms of Use at cybercompanyprofiles.com/terms, which permit personal and internal business use with attribution and prohibit republication and resale.