100 free credits. No credit card required.Start building
Logo
Prism logoPrism API

Run cross-platform Prism queries with one API

Call SocialCrawl Prism with one API key. Get composite social analytics as structured JSON. URL lookup, multi-platform handles, brand intelligence, crisis, GEO, and batch posts or profiles. Reads are priced in credits.

Prism logo
/v1/prism

33 active composite endpoints.

  • GET /v1/prism/lookup
  • GET /v1/prism/comments
  • GET /v1/prism/creator-card
  • GET /v1/prism/handle-audit
  • GET /v1/prism/voice
  • GET /v1/prism/video-intel
  • GET /v1/prism/answers
  • GET /v1/prism/brand-mentions
  • GET /v1/prism/demand-signals
  • GET /v1/prism/campaign
  • GET /v1/prism/ai-visibility
  • GET /v1/prism/crisis-radar
  • GET /v1/prism/crisis-postmortem
  • GET /v1/prism/reputation
  • GET /v1/prism/share-of-voice
  • GET /v1/prism/leads
  • GET /v1/prism/devtool-pulse
  • GET /v1/prism/launch-echo
  • GET /v1/prism/org-radar
  • GET /v1/prism/post-stats
  • GET /v1/prism/profiles
  • GET /v1/prism/comment-lookup

What Prism endpoints does SocialCrawl ship?

Thirty-three live composite endpoints under /v1/prism. Resolve any social URL, fan out handles across platforms, run brand and crisis intelligence, probe AI visibility, and batch post or profile refreshes. Data API only. No posting and no private accounts.

URL lookup

0 credits (pass-through)
/v1/prism/lookup

Paste a social or shopping URL. Prism resolves the platform and resource, then returns the post, profile, video, or product in one envelope. Billing follows the resolved endpoint.

url, include

All comments

metered, min ~2 credits
/v1/prism/comments

Harvest every comment on a TikTok, YouTube, Facebook, Reddit, Hacker News, or Instagram post URL. Pages to completion with nested replies where the platform allows.

url, max, replies, cursor, sort, limit

Creator card

~5 credits
/v1/prism/creator-card

Look up one handle across platforms in parallel. Returns followers, verified status, bio, and ids per platform, or null where the handle is missing.

handle, platforms, include

Handle audit

~5 credits
/v1/prism/handle-audit

Scored pre-flight check for a handle across platforms. Verdict, per-platform scores, activity signals, and a projected credit cost for a full pull.

handle, platforms, sample

/v1/prism/voice

Merge one person's recent posts from X, Threads, Bluesky, and Truth Social into a single time-ordered feed, plus which networks they are active on.

handle, platforms, cursor, include

Video intel

~5 credits
/v1/prism/video-intel

One video URL in: details, stats, comments, optional transcript, and on TikTok or Instagram a few commenter profiles. Covers YouTube, TikTok, Rumble, and Instagram.

url, comments, include

/v1/prism/answers

Ask several AI search engines the same question. Returns each answer verbatim, merged citations, and where the engines agree or disagree.

query, engines, include

Brand mentions

50 credits
/v1/prism/brand-mentions

One keyword mention report: volume over time, positive and negative split, top sources, and recent mention samples in a single call.

keyword, date_from, date_to, date_group, page_type, include

Demand signals

30 credits
/v1/prism/demand-signals

Fuse web mention slope, Reddit velocity, product review levels, and optional app review speed into a transparent consumer demand index.

keyword, google_play_id, app_store_id, signals, amazon_query, country, date_from, date_to, depth

/v1/prism/campaign

Measure a hashtag or phrase campaign: pre, during, and post volume, engagement by platform, and the accounts amplifying it most.

hashtag or phrase, window_start, window_end, pre_days, post_days, include

AI visibility

metered ~2 credits/probe
/v1/prism/ai-visibility

How often answer engines name your brand for a prompt set, plus which domains they cite. For AEO and GEO tracking.

brand, prompts, topic, competitors, engines, runs, preset, include

Crisis radar

15 credits base
/v1/prism/crisis-radar

Compare today mention volume and negative share to a recent daily baseline. Returns calm, watch, alert, or crisis for recurring checks.

brand, sensitivity, confirm, baseline_days, date_to

/v1/prism/crisis-postmortem

Reconstruct a brand crisis timeline across web, Reddit, Hacker News, and social: origin, peak, propagation, and a written summary.

brand, window_start, window_end, crisis_terms, include

/v1/prism/reputation

Blended reputation from Trustpilot, app stores, Google Business, and web sentiment, with themed pros, cons, and quotes.

brand, sources, country, depth, place, axis, app_store_id, google_play_id, include

Share of voice

~40 credits/brand
/v1/prism/share-of-voice

Engagement-weighted share of voice for two to five brands, split into web and social, with emotion breakdown and per-brand trend.

brands, category_code, market_shares, include, page_type, date_from, date_to

/v1/prism/leads

Ranked public posts and comments where people ask for alternatives to a competitor or say they are switching away.

competitor, product_category, freshness, include, limit

Dev tool pulse

~15-20 credits
/v1/prism/devtool-pulse

Health snapshot for a developer tool: GitHub repo details, Hacker News reaction, Reddit chatter, blog coverage, and a summary label.

query, repo, subreddit, include, date_from, date_to

Launch echo

~15-20 credits
/v1/prism/launch-echo

How a launch was received: top Hacker News threads and comments, blog coverage, and optional GitHub repository context.

query, repo, threads, date_from, date_to, include

Org radar

~26 credits headline
/v1/prism/org-radar

GitHub organization footprint: most starred repos with releases, open issue load, and top feature request and complaint signals.

org, repos, sort, include

Earned media

25 credits
/v1/prism/earned-media

Deduplicated ranked feed of news, tech press, and fresh web articles mentioning a brand, plus which outlets covered it most.

brand, competitor, date_from, date_to, min_domain_rank, include

Truth Social pulse

~15-20 credits
/v1/prism/truthsocial-pulse

One Truth Social account: profile, recent posts, detail on top posts, related news, plus activity and sentiment read.

handle, drill, posts, news_query, include, cursor

Audience overlap

~15-20 credits
/v1/prism/audience-overlap

How much two TikTok creators' commenting audiences overlap: shared people, overlap score, unique counts, and confidence.

handle_a, handle_b, platform, videos_per_creator, depth

Employer brand

30 credits
/v1/prism/employer-brand

What people say about working at a company across Reddit, news, and YouTube, plus LinkedIn post tone when you pass linkedin_url.

company, linkedin_url, surfaces, phrases, timeframe, date_from, date_to

/v1/prism/audience-questions

Real questions people ask about a topic from Reddit and YouTube, grouped by intent such as who, what, why, how, and comparisons.

topic, platforms, max_questions, threads_per_source, timeframe, include

Product reviews

30 credits
/v1/prism/product-reviews

One product across Amazon, Google Shopping, and Trustpilot: rating per source, retailer comparison, and themes moving the score.

query or asin or gid, sources, country, depth, competitors, include

Apps lookup

30 credits
/v1/prism/apps-lookup

One app on Google Play and the App Store side by side: rating, installs, price, store gap, and match confidence.

title or google_play_id or app_store_id, stores, country, language, match_threshold, include

App reviews

~15-20 credits
/v1/prism/app-reviews

Google Play and App Store reviews together: rating summaries, topic clusters, sentiment over time, feature requests, and dev reply rate.

google_play_id, app_store_id, query, country, language, depth, stores, include

Creator vet

50 credits
/v1/prism/creator-vet

Vetting signals for a creator: engagement rate, commenter authenticity, posting frequency, and news or forum controversy.

handle, platform, depth, include

Korea gap

40 full / 15 web-only
/v1/prism/korea-gap

Gap between global and Korean talk about a topic: which Naver surfaces carry it, channel map, and translated example quotes.

query, include, date_from, date_to, display

/v1/prism/review-integrity

A to F grade on whether product ratings look trustworthy across Amazon, Google Shopping, Trustpilot, and the open web. Deterministic, no LLM.

query or asin or gid, sources, country

Batch post stats

metered POST batch
/v1/prism/post-stats

POST up to 100 mixed-platform post URLs. Returns view, like, comment, share, and save counts one row per URL in input order.

body urls (1-100), include

Batch comment lookup

metered POST batch
/v1/prism/comment-lookup

POST up to 25 known comments to re-check current state. Per-item found, not found, errored, or deferred status.

body items (1-25)

Batch profiles

metered POST batch
/v1/prism/profiles

POST up to 50 platform and handle pairs. Returns each profile in input order with per-row status, same shape as single profile calls.

body items (1-50)

Prism API33 endpoints supported
View docs

Returns the data behind any social or shopping link you paste in: the post, profile, video, or product that URL points to, in one response.

Use it when you have a URL but do not know which platform endpoint fits, so you can skip writing your own link parsing and routing.

0 credits

GET/v1/prism/lookup?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DdQw4w9WgXcQ

url · Absolute http(s) URL of a single post, video, product or repo. Profile, channel and storefront URLs are NOT accepted: use that platform's own `/profile` endpoint for those.

$ curl https://www.socialcrawl.dev/v1/prism/lookup?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DdQw4w9WgXcQ \
    -H "x-api-key: sc_YOUR_API_KEY"
idle
// Running this live needs your own API key. Hit "Try it" to see an example response

How the Prism API works

Prism is SocialCrawl's composite layer under /v1/prism. You call GET or POST with one API key, spend credits on live work, and get the shared JSON envelope. No second SDK and no per-platform OAuth for these composites.

Authenticate every call

Send your key in the x-api-key header. The same SocialCrawl key works for every /v1/prism recipe and the rest of the platform catalog.

GET for recipes, POST for batches

Most Prism routes are GET with query params. post-stats, comment-lookup, and profiles are POST with JSON bodies. We validate before charging.

Pay in credits, not seats

Composites bill by recipe: flat tiers, per-probe meters, or per-item batch costs. Cache hits cost 0. Empty or hard failures refund. Bad params return 400 with no charge.

Read one JSON envelope

Every response is the same shape: success, data, credits_used, credits_remaining, request_id, cached. Many composites also surface legs[] or per-item status for audit.

The usual integration chain

Most products start with a URL or handle, deepen into creator or voice context, then run brand or crisis intelligence when monitoring is the job.

01Resolve
GET /v1/prism/lookup?url=…

lookup: resolved platform path plus post or profile payload

You have a link but not the platform endpoint. Prism routes and bills the resolved leg.

02Creator
GET /v1/prism/creator-card?handle=…&platforms=…

creator-card: one author card per platform for a handle

Confirm presence and follower counts before a deeper pull or partnership.

03Audit
GET /v1/prism/handle-audit?handle=…

handle-audit: scored verdict plus projected pull cost

Pre-flight risk and credit estimate before you spend on full collection.

04Brand
GET /v1/prism/brand-mentions?keyword=…&date_from=…

brand-mentions: volume timeline, sentiment, sources

Keyword monitoring after the identity is known.

request
GET /v1/prism/lookup
  ?url=https://www.tiktok.com/@…/video/…
Host: www.socialcrawl.dev
x-api-key: sc_your_api_key_here

# multi-platform handle
GET /v1/prism/creator-card?handle=mrbeast&platforms=tiktok,instagram,youtube
GET /v1/prism/handle-audit?handle=mrbeast
response envelope
{
  "success": true,
  "data": {
    "resolved": {
      "platform": "tiktok",
      "resource": "post",
      "path": "/v1/tiktok/post"
    },
    "result": {
      "id": "7123456789",
      "url": "https://www.tiktok.com/@…/video/…",
      "content": { "text": "Example caption" }
    }
  },
  "credits_used": 1,
  "credits_remaining": 9999,
  "request_id": "req_…",
  "cached": false
}

What lands in data

Prism returns finished composite objects, not raw multi-scraper glue. Field names stay close to SocialCrawl archetypes where a social leaf applies.

Resolve / Creatorlookup, creator-card, handle-audit

lookup resolved path + result; creator-card and handle-audit per-platform cards, scores, and nulls where missing

Brand / Campaign / Reputationbrand-mentions, campaign, reputation

volume timelines, sentiment splits, share metrics, campaign lift windows, blended reputation with themes and quotes

Answers / AI visibilityanswers, ai-visibility

engine answers with citations, agreement maps, brand mention rates in AI replies, cited domain ranks

Batch POSTpost-stats, profiles, comment-lookup

post-stats, profiles, comment-lookup: per-item rows in input order with status and engagement or profile leaves

Inside the gateway

Same request lifecycle as every other /v1 surface. Prism is composite orchestration on top of registry endpoints, not a separate product auth path.

  1. 01

    Edge receives the call

    Next.js catch-all routes into the Hono social API. We mint a request_id, authenticate the key, then enforce rate limit and concurrency for the key.

  2. 02

    Validate, then debit

    Registry lookup finds the prism resource. Required params and body shape checks run first. Invalid input returns 400 with no charge. Valid calls debit the recipe cost (or hold for metered batches) before fan-out.

  3. 03

    Fan out or cache

    A deterministic cache key is built from platform, resource, and params. Hit: return with credits_used = 0. Miss: parallel platform legs, content analysis, or AI probes run under the recipe, with retries and circuit breakers where policy allows.

  4. 04

    Fuse and return

    Legs merge into one Analytics or batch result, validate against the response contract, wrap the success envelope, attach leg receipts when the recipe bills per leg, and log for billing audit.

Billing rules that matter in production

  • lookup: 0 at the composite; pass-through bills the resolved endpoint
  • Simple composites (creator-card, handle-audit, voice, video-intel): ~5 credits base
  • Intelligence recipes: fixed tiers from about 15 to 50 credits
  • ai-visibility: metered ~2 credits per probe
  • share-of-voice: ~40 credits per brand
  • POST batches: metered per successful item; failed items refund
  • Cache hit: 0 credits
  • Empty / hard fail: auto-refund
  • Bad params: 400, never charged; no balance: 402, never charged

How we get the data

Prism is a composite layer over SocialCrawl social reads, content analysis, app stores, marketplaces, and AI answer engines. One key and one envelope instead of many scrapers.

What Prism is for this API

Finished composite jobs: resolve a URL, audit a handle, monitor a brand, grade reviews, probe AI visibility, or refresh a batch of posts. Read-only. No posting and no private inbox access.

How SocialCrawl reaches it

Each recipe fans out to registered platform endpoints and analysis legs in parallel, then fuses and scores in code. Deterministic grades where the contract says so. LLM only where the recipe documents narrative or clustering.

What leaves our edge

The unified JSON envelope: success, data, credits_used, request_id, cached. Composite data objects plus optional legs[] receipts so you can audit which sources contributed.

What we do not ship live

No write endpoints. No per-platform profile/full under /v1/prism (those live on each platform hub). No guaranteed private or authenticated social content.

field map sketch
URL or handleresolved platform legsURL or handle becomes resolved platform legs
fan-out social readsmerged Analytics / Post / Authorfan-out social reads merge into Analytics, Post, or Author leaves
batch POST bodiesper-item results + creditsbatch POST bodies become per-item results and credit rows

What this API is used for

The jobs this API is most often used for.

33active Prism endpoints in the registry
0-50crcredit range across Prism recipes

Cross-platform resolve, brand intelligence, and creator pre-flight

Callers concentrate on lookup, creator-card, handle-audit, brand-mentions, and the flagship intelligence recipes (share-of-voice, review-integrity, ai-visibility, crisis). Batch post-stats shows up in monitoring loops. One composite layer for URL routing, multi-platform handles, brand and crisis intelligence, GEO probes, and metered batch refreshes under a single API key.

Simple resolve and handle cards are typically a few seconds on live miss. Multi-source intelligence recipes are slower because they fan out and fuse several legs.

Use case examples

Common ways teams put this data to work, and the stack each one tends to run.

Brand and PR monitors

Python, cron, Slack

Run brand-mentions, crisis-radar, and earned-media on a fixed cadence. Escalate to crisis-postmortem when status leaves calm.

Creator and partnership ops

Node, sheets, Explorer

creator-card and handle-audit first, then creator-vet or audience-overlap before spend. Same key as the rest of SocialCrawl.

Backend product jobs

Go, workers, warehouses

lookup for URL routing, post-stats and profiles for batch refresh, intelligence recipes when a finished grade or SOV number is the product output.

Call it in two lines

Start with lookup or creator-card. Cache hits are free. Intelligence recipes and POST batches use the higher rungs documented per endpoint.

curl "https://www.socialcrawl.dev/v1/prism/lookup?url=https://www.tiktok.com/@mrbeast/video/7123456789" \
  -H "x-api-key: sc_your_api_key_here"
curl "https://www.socialcrawl.dev/v1/prism/creator-card?handle=mrbeast&platforms=tiktok,instagram,youtube" \
  -H "x-api-key: sc_your_api_key_here"
Prism logoPrism on SocialCrawl

Same key as the rest of the catalog

Endpoints

What data does the Prism API return?

Every endpoint returns structured JSON in a unified envelope. Computed fields such as engagement rate and content category are included only where the endpoint supports them and the required source inputs are present.

LookupUniversal post/product URL dispatcher: any post, video, product or repo link → the right detail endpoint's unified response.CommentsEvery comment on a post, replies nested, server-paginated to completion.Brand MentionsBrand mention volume time-series, sentiment split, top sources, and recent mentions for one keyword.Demand SignalsConsumer-demand nowcast: app-review velocity, web mention slope, Reddit velocity, and commerce review levels, fused into a published demand index.CampaignCampaign tracker: pre/during/post volume lift, cross-platform engagement, and ranked top amplifiers for a hashtag or phrase.Ai VisibilityAI Share-of-Voice / GEO monitoring: prompt set x reruns to per-brand appearance-% per AI engine plus a cited-domain ranking.Crisis PostmortemCrisis post-mortem: a who-said-what-first timeline across web, Reddit, Hacker News, and social, with an origin, peak, propagation sequence, and a grounded narrative.Crisis RadarStateless crisis breach check: a z-score on daily mention volume and negative share, with on-breach confirmation and a severity grade.Devtool PulseDeveloper-brand health: a devtool's repo dossier + Hacker News reaction + Reddit chatter + dev-blog echo, in one call.LeadsRanked feed of public conversations where people seek alternatives to or are switching from a competitor.Earned MediaA brand's earned-media footprint: news + tech-press + fresh-web clips, deduped and ranked, with an outlet-coverage rollup.Truthsocial PulseA Truth Social handle's pulse: profile, recent posts, per-post detail drill, and the news echo, in one call.Launch EchoHow a launch landed: the Hacker News reaction (top threads + comments), the dev-blog echo, and an optional repo dossier.Audience OverlapHow much two TikTok creators' commenter audiences overlap. Jaccard, shared-fan count, and a confidence label.ReputationA brand's cross-source reputation. Trustpilot + app stores + Google Business + web sentiment, blended into one weighted score with themed pros/cons.Employer BrandA company's employer brand: what people say about working there across Reddit, the web, YouTube, Naver, and the company's own LinkedIn voice.Audience QuestionsThe real questions a topic's audience asks: harvested from Reddit + YouTube threads and clustered by intent (who/what/why/how/vs).Product ReviewsA product's reviews across Amazon + Google Shopping + Trustpilot, folded into a cross-marketplace rating + themed pros/cons report.Apps LookupOne app across Google Play + the App Store: resolved, title-matched, and compared into a cross-store rating + listing report.Org RadarA GitHub org's footprint: its top repos each expanded into a full dossier (releases, issue load, top request/complaint), rolled up.Creator VetVet a creator before partnering: engagement quality, commenter authenticity, posting cadence, and controversy signals, optionally across platforms.Korea GapWhat the world is talking about that Korea isn't (and vice versa): the global vs Korean (Naver) conversation gap for a brand/topic.Share Of VoiceEngagement-weighted Share of Voice across 2-5 brands, with web+social split, emotion overlay, and ESOV.Review IntegrityCross-source review integrity verdict (statistical, deterministic).AnswersMulti-engine AI consensus: one question → Perplexity + Grok + Tavily answers verbatim, merged citations, and an agreement matrix.Video IntelOne video URL → detail + stats + transcript + top comments + commenter sample, across YouTube/TikTok/Rumble/Instagram.VoiceOne person's public posts across X, Threads, Bluesky, and Truth Social, time-merged.App ReviewsCross-store app review intelligence (Google Play + App Store): translated, clustered, sentiment-scored.Creator CardOne handle, unified author cards across TikTok, Instagram, YouTube, X (and more).Handle AuditShould you pull this handle? One call scores a handle across platforms, ranks the best ones, and projects the data volume + credit cost to pull it.Post StatsUp to 100 mixed-platform post URLs → current engagement per URL, failed URLs refunded.Comment LookupRe-check up to 25 known comments in one call: per-item results, failed items refunded.ProfilesUp to 50 (platform, handle) pairs → one canonical Author per row, failed handles refunded.
Comparison

How does SocialCrawl compare to Calling each endpoint by hand for Prism data?

Same Prism data, two very different paths. Auth, rate limits, schema, and cost, side by side.

Platform detection

SocialCrawl
Sniffs the platform and endpoint from the URL automatically
Calling each endpoint by hand
You map every URL pattern to the right endpoint yourself

Calls to learn

SocialCrawl
One request shape for any of 15 supported URL types
Calling each endpoint by hand
A different request to learn and wire up per platform

Pricing

SocialCrawl
Pass-through: the underlying endpoint's price, no markup
Calling each endpoint by hand
Same per-endpoint cost, plus the routing logic you maintain

Response shape

SocialCrawl
Unified JSON with a resolved{} routing block and legs[] transparency
Calling each endpoint by hand
Each platform hands back its own raw payload to normalize

Maintenance

SocialCrawl
Routing stays current as endpoints and URL shapes change
Calling each endpoint by hand
Your URL parsing breaks when a platform changes its links
FAQ

Have a question? We got answers

Find answers to frequently asked questions about SocialCrawl's API, pricing, and capabilities.

Contact us
What is the SocialCrawl social media URL API?
GET /v1/prism/lookup is SocialCrawl's social media URL API: it takes any supported social or commerce URL, detects the platform and the right detail endpoint, and returns that endpoint's unified JSON in one call. You hand it a link and get back structured data. No need to know which of SocialCrawl's endpoints to call.
Which URLs does the lookup API support?
Fifteen URL types today: TikTok videos, Instagram posts and reels, YouTube watch links, X/Twitter status URLs, Facebook posts, LinkedIn posts, Bluesky posts, Pinterest pins, Reddit threads, Threads posts, Rumble videos, Truth Social posts, GitHub repos, Amazon product pages, and Spotify tracks. Unsupported hosts return a 400 with the accepted list and cost nothing.
How much does a URL lookup cost?
There is no routing surcharge: you pay the underlying endpoint's normal price. A social URL costs 1 credit; an Amazon product URL costs 5 credits because it dispatches the advanced commerce endpoint. The credits_used field always reports the resolved cost.
What does the response look like?
You get SocialCrawl's standard envelope with a resolved{} block naming the platform, endpoint, and archetype it routed to, the dispatched endpoint's unified data spread in, and a legs[] array showing the leg's status, credits, and latency. The archetype tells you whether to expect a Post, Author, Product, or CommentList shape.
Why use one URL endpoint instead of calling each platform directly?
You stop maintaining URL-to-endpoint routing yourself. Prism sniffs the platform from the link, handles per-platform quirks like Truth Social's URL shape and Amazon's ASIN extraction, and keeps the routing table current as endpoints change. One request shape covers every supported link.
Do I need to send the platform name with the URL?
No. The url parameter is the only required input. Prism infers the platform and endpoint from the host and path. That makes it the ideal first call for AI agents and integrations that hold a link but don't know which platform it belongs to.
Does SocialCrawl offer a cross-platform social data API?
Yes, SocialCrawl is that cross-platform data API, and Prism is its one-call front door. Hand GET /v1/prism/lookup any supported social or commerce URL and it detects the platform, routes to the right detail endpoint, and returns unified JSON, so you read a TikTok video, an Instagram reel, a Reddit thread, a GitHub repo, or an Amazon product page through one request shape. There is no routing surcharge: you pay the underlying endpoint's normal price.
Is scraping Prism data legal?
SocialCrawl returns publicly available Prism data and does not access private or login-gated content. Whether scraping fits your project depends on your use case and jurisdiction. You are responsible for complying with Prism's terms of service and applicable data-protection laws such as GDPR and CCPA. This is general information, not legal advice.
Prism scraping API vs the official Prism API: what's the difference?
With SocialCrawl there is no app review or approval queue. Sign up and call Prism endpoints immediately with a single x-api-key. Responses share one unified schema with every other SocialCrawl platform, and credit-based pricing replaces per-platform quotas. Official APIs are still the right choice for posting and other write actions: SocialCrawl is read-only data.

Ask AI about SocialCrawl

Read the Prism API reference in the docs

🤖 AI agent or LLM? Read this page as markdown