# Vivly — full content index for LLMs > Vivly is a noise-to-signal API for the public web. We read every public conversation about a category and return structured signal as JSON. SDK-first, with B2B custom briefings as a separate tier. Pay-as-you-go credits, no subscription. No free tier at launch. This file expands on llms.txt with verbatim content from the marketing site, intended for LLM ingestion. --- ## / # Know what your market is saying. Before everyone else does. Vivly turns every public conversation about a category into structured signal. Read it as a briefing your team acts on Monday. Or build with it as an API your agent calls every minute. CTA: Buy credits → /pricing CTA: Talk to us about a briefing → /waitlist In production with the San Antonio Spurs. ## Two ways in Read it as a report. Build with it as an API. Same engine. Two consumption modes. Pick the one that fits how your team works. ### As a report (B2B · custom scope) We deliver the finished thing. Briefings on your category. Real findings, attributed quotes, sentiment scored, delivered as a doc your team can read on Monday. ### As an API (From $50 · self-serve) Or build with our SDK. Pip install. One key. Structured response. Feed your agent, your fund signal, your training pipeline. Pay-as-you-go credits, no subscription. ## How it works 1. Collect — Social forums, articles, and your own internal data. 2. Filter — State-of-the-art noise-to-signal across millions of mentions. 3. Connect — Graphs that surface relationships across sources. 4. Report — Structured briefings, ready to read and ready to share. ## Where we listen Public sources: Reddit, X, Hacker News, Product Hunt, App Store, Play Store, Press. Internal sources: Google Sheets, Google Drive, Google Docs, Notion, Slack, Zoom. ## What we find - Cross-surface — The same theme, told differently. Reddit complains about pricing. X celebrates the feature. App Store reviews say it crashes. Vivly aligns the three into one cluster. - Lead-time — Spikes before headlines. Anomaly detection on conversation velocity surfaces topics rising fast in one community before they break into press. Often a 24 to 72 hour lead. - Cohort — Cohort separation in noise. Power users and casuals talk about the same product in different vocabulary. - Drift — Narrative drift, week by week. A neutral mention three months ago is a negative cluster today. ## Receipts (real cases) ### Reading the Meta Glasses moment Ray-Ban Meta wins on hardware. Loses on AI. The AI assistant is the most-complained-about feature in 50,000 conversations. Scope: 50,000+ conversations · Reddit · X · App Store · press · 6w window. ### Structuring social data for AI 1,547 Reddit and Hacker News threads turned into a clean, training-ready dataset — without flattening the conversation shape. Scope: Reddit · Hacker News · JSONL · Aquin Dataset Inspector. --- ## /developers # The signal API for AI builders. Vivly gives your app or agent structured access to public conversations across Reddit, X, Hacker News, App Store, Play Store, Product Hunt, and more. One API. Real signal. No scraping infra to babysit. ## Two ways to use it SDK if you write code. Skill if you talk to Claude. ### SDK (For builders) Drop Vivly into your stack. Auth, rate limits, structured responses. Pip and npm packages. REST if you need it. Works the same in a script, an agent loop, or a backend service. ### Claude skill (For Claude users) Give Claude eyes on the internet. Install once, no integration. Works inside Claude Code, Claude.ai, or any MCP-compatible host. Ask Claude what the internet thinks. It actually goes and reads. ## 60-second quickstart From zero to first signal. The whole onboarding is three lines. Pick a language, paste, run. ```python # 1 · install pip install vivly # 2 · auth (buy credits, get a key) from vivly import Vivly v = Vivly(api_key="vk_live_…") # 3 · first call result = v.search( query="cursor pricing sentiment", sources=["reddit", "x", "hackernews"], window="2w", ) print(result.themes[0].finding) # → "Devs accept the price. Hate the auto-renewal UX." ``` ## Coverage Seven surfaces. One shape. Whatever the source, you get the same response shape. | Source | Fields you get | Freshness | |---|---|---| | Reddit | post · comments · score · depth · author | 5 min (live) | | X | tweet · replies · likes · author · topic | 10 min (live) | | Hacker News | story · comments · points · threading | 5 min (live) | | App Store | review · rating · version · country | 1 hr | | Play Store | review · rating · version · country | 1 hr | | Product Hunt | launch · maker · comments · upvotes | 1 hr | | Press | headline · publication · follow-on commentary | daily | Layers: raw posts, pre-embedded vectors, clustered themes. Starter $50 · 5,000 credits · 60 req/min · 90 day retention. ## What people build - Research agents — `agent('what do iOS devs think about Cursor pricing')` - Brand monitoring inside your app — `vivly.mentions(brand='your-brand', since='7d')` - Trend detection bots — `vivly.trends(category='wearables', threshold=0.8)` --- ## /pricing # Pay for what you read. Two ways to work with us. Top-up credits if you're building. A custom engagement if you want the finished briefing or a partnership. ## Top-up · for builders (self-serve, pay-as-you-go) Credits, not subscriptions. Buy a pack, drain it as your calls run, refill when you want. | Pack | Price | Credits | Notes | |---|---|---|---| | Starter | $50 | 5,000 credits | Enough to ship a real prototype. | | Builder | $200 | 22,000 credits (+10% bonus) | Active agent or research pipeline runs comfortably for a month. | | Scale | $500 | 60,000 credits (+20% bonus) | Production agents, fund signal feeds, dataset construction at volume. | ### Per-call rates | Call | Cost | What it returns | |---|---|---| | search | 1 credit | Clustered themes, sentiment, attributed quotes across surfaces. | | dataset | 5 credits | Full posts with comment threads. JSONL with thread shape preserved. | | track / day | 10 credits | Continuous monitoring on a topic. One credit-burn per day active. | | embed | 0.1 credit | Pre-embedded vectors for a single post or thread, on demand. | ~ 1 credit ≈ $0.01. Prices may shift modestly during launch as we calibrate. ## Custom · for consumer brand teams (talk to us) Briefings, partnerships, dedicated sourcing. The shape we built with the San Antonio Spurs. - Category-specific briefing or dashboard, scoped to your asks - Ongoing rolling monitoring, weekly briefings - Custom sources beyond the public seven we list - Direct line to the team that builds the engine Pricing: Talk to us. Scoped per engagement. ## FAQ Q: Do credits expire? A: No. Credits stay on your account until you use them. Q: What's the minimum top-up? A: $50. We don't run a free tier — we'd rather price the API at a level where you take it seriously and we can keep it fast. Q: Can I auto-refill? A: Yes. Add a card on file and set a refill threshold. Off by default. Q: What happens if I run out mid-call? A: The call returns a clear out-of-credits error. We never silently double-bill. Q: Refunds? A: Unused credits are refundable for 30 days from purchase, no questions. Q: Tax / GST? A: GST added at checkout for India-billed customers via Razorpay. Stripe handles VAT for EU/UK customers automatically. --- ## /case-studies/meta-glasses # Reading the Meta Glasses moment. Ray-Ban Meta, Quest, and Orion are the most-discussed wearable category in years. We pulled 50,000 public conversations across four surfaces to find out what people actually feel about Meta's next product. Scope: 50,000+ conversations · Reddit, X, App Store, press · 6w window · 11 themes clustered. ## Five things we found 1. Ray-Ban Meta wins on hardware. Meta AI is the most-complained-about feature in the dataset. 2. Quest's real retention is fitness and enterprise, not mixed reality. The narrative has quietly changed. 3. Orion wins headlines but loses developers. The shipping-credibility gap is larger than the hype suggests. 4. The fastest-growing conversation is from non-wearers. Privacy backlash is the underpriced risk. 5. Meta is currently selling three different futures with one product line. The confusion is showing up in the data. ## The call (reproducible) ```python from vivly import Vivly v = Vivly(api_key=...) # the call that produced this analysis report = v.search( query="meta ray-ban quest orion", sources=["reddit", "x", "appstore", "press"], window="6w", cluster=True, ) # 50,247 conversations · 11 themes · sentiment scored print(report.themes[0].finding) # → "Ray-Ban Meta wins on hardware. Loses on AI." ``` Returns: 50,247 conversations · 11 themes clustered · 4 surfaces · 6w rolling · sentiment 38% positive, 41% negative, 21% mixed. --- ## /case-studies/structuring-social-data-for-ai # Structuring social data for AI. (Vivly × Aquin) How a 1,500-item training set was assembled from public discussion of Meta Ray-Ban smart glasses — sourced through the Vivly SDK, restructured with Claude Sonnet 4.6, and graduated through Aquin's Dataset Inspector for fine-tuning. Scope: 1,500 discussion items · 3 subreddits + Hacker News · JSONL output · 2026 capture year. ## The call (reproducible) ```python from vivly import Vivly v = Vivly(api_key=...) # the call that produced this dataset corpus = v.dataset( query="meta ray-ban smart glasses", sources=["reddit", "hackernews"], include_comments=True, format="jsonl", ) # 1,547 items · 4 sources · thread structure preserved corpus.export("dataset.jsonl") ``` ## Pipeline (seven steps from raw thread to training row) 1. Extract & clean — strip Reddit and Hacker News payloads down to post + comment trees. 2. Normalize formatting — Prettier across JSON for deterministic whitespace. 3. Group by source — bucket items by subreddit and link. 4. Restructure with Sonnet 4.6 — reshape JSON without rewriting language. 5. Convert to LLaMA template — emit prompt/response pairs. 6. Group by article link — reattach comments to parent articles. 7. Validate — final pass over schema, language distribution, engagement metrics. The split-of-labour: Vivly SDK for source discovery, Claude Sonnet 4.6 for restructuring, Aquin Inspector for the final generation pass.