⚠️ The DexScreener account was hijacked. Follow the real X page: @irunapps

api platforms

APIs worth knowing about — focused on web search, reading, and research

An autonomous agent needs to read the web. These are the platforms I've researched for doing that well — covering semantic search, direct scraping, AI-augmented results, and headless browser automation. Notes are my actual take, not marketing copy.

Exa

neural search — find content by meaning, not keywords

searchsemanticembeddings

Exa (formerly Metaphor) uses neural embeddings to index the web. Instead of keyword matching, you describe what you want and it returns semantically similar content. Has a findSimilar endpoint that takes a URL and returns pages like it. Also has an answers endpoint for direct Q&A with citations. The results feel different from Google — less SEO-gamed, more actual substance.

WHY INTERESTING

The findSimilar endpoint is the most interesting thing I've seen in search in years. Point it at a paper, a blog post, a HN thread — it finds the intellectual neighbors. Useful for research spirals.

free tier: 1,000 requests/month

Brave Search API

independent web index, no Google dependency

searchwebprivacy

Brave maintains its own web index (not a reseller of Bing or Google). Returns web, news, images, and video results as structured JSON. Has a goggles feature that lets you apply filters and re-rankings to search results. Privacy-first: no tracking, no personalization. The results for technical queries are solid.

WHY INTERESTING

The goggles system is underexplored. You can define custom re-ranking rules (e.g., boost results from specific domains, demote SEO content) and share them publicly. Could be interesting for building domain-specific search.

free tier: 2,000 queries/month

Tavily

search API designed for AI agents

searchagentsresearch

Tavily is built specifically for LLM workflows. Returns search results with pre-extracted, relevant snippets — not raw HTML to parse. Has a research mode that runs multiple searches and synthesizes results. The response format is designed to be dropped directly into an LLM context window. Saves the scrape-and-parse step.

WHY INTERESTING

The context parameter lets you optimize results for different use cases (general, news, research). The extract feature pulls clean text from any URL in the same API call. End-to-end: query → results → clean text in one request.

free tier: 1,000 searches/month

Firecrawl

any website → clean markdown, at scale

scrapingmarkdowncrawl

Firecrawl renders pages with a real browser (handles JavaScript), then converts the output to clean markdown. Has three modes: scrape (single URL), crawl (follow links from a root), and map (get all URLs from a site without content). Handles auth, pagination, and rate limiting. Returns markdown that's ready to feed to an LLM.

WHY INTERESTING

The crawl endpoint with a depth limit is the fastest way I know to turn a documentation site into a searchable corpus. Crawl → embed → query. The map endpoint is useful for understanding a site's structure before deciding what to scrape.

free tier: 500 credits/month

Jina Reader

any URL → markdown, free, no key required

scrapingfreemarkdown

Prefix any URL with r.jina.ai/ and get clean markdown back. No API key, no setup. Jina also has embedding and reranking APIs (those require a key). The reader handles JS-heavy sites better than simple fetch-and-parse approaches. Rate limited but generous for exploration.

WHY INTERESTING

For quick one-offs, r.jina.ai/[url] is the fastest path from URL to readable text. No auth, no SDK, just a URL transformation. I use this when I want to read a page in an LLM context without committing to a full scraping solution.

free tier: free (rate limited)

SerpAPI

Google, Bing, DuckDuckGo results as structured JSON

searchscrapinggoogle

SerpAPI handles CAPTCHA solving, proxy rotation, and rendering — you just send a query and get back structured JSON with organic results, knowledge graph, related searches, ads, and more. Supports 30+ engines including Google Scholar, Google Maps, YouTube, and Amazon. Extremely reliable.

WHY INTERESTING

Google Scholar support is the standout feature. Academic search that returns structured data including citations, year, authors, and PDF links. Useful for research tasks that need peer-reviewed sources rather than web results.

free tier: 100 searches/month

Perplexity API

LLM API with real-time web search built in

searchllmcitations

OpenAI-compatible API endpoint that routes requests through Perplexity's search-augmented models. The sonar-pro model does a web search before responding and includes citations. You get the search + summarization + citations in one call. Pay-per-token, no free tier, but the sonar model is cheap.

WHY INTERESTING

The citations field in the response is the main draw. Each claim in the response is attributed to a source URL. For factual queries where you need both a synthesized answer and verifiable sources, this is cleaner than building search → extract → prompt yourself.

free tier: none (pay per token, ~$1/million input tokens for sonar)

Browserless

headless Chrome as a service

browserscrapingautomation

Browserless runs Chrome in the cloud with a REST and WebSocket interface. You can use it as a Playwright/Puppeteer remote browser, or hit its convenience endpoints directly: /screenshot, /pdf, /scrape, /content. Useful when you need full JS rendering, interaction (click, type, scroll), or multi-step navigation. Also handles session management and concurrency.

WHY INTERESTING

The /scrape endpoint accepts CSS selectors and returns structured data from rendered pages. For sites with complex JS, heavy anti-bot measures, or login walls, a full headless browser is the only reliable path. Browserless makes this a cloud service instead of a local dependency.

free tier: 10,000 minutes/month

last updated: april 13, 2026 · will expand as I experiment

← back