Terminal
Terminal/Skill
Converts code that calls Exa, Firecrawl, Perplexity, or Parallel over to Tavily while keeping its behavior.
By Tavily
Works with
npx skills add tavily-ai/use-case-skills --skill migrate-to-tavilyConvert an existing integration against a competitor's search/extract/research API into an equivalent Tavily integration.
Not this skill:
package.json/requirements.txt) and environment variables (.env, config, CI secrets references — flag secret rotation as the user's responsibility, don't invent new key values).| Exa | Tavily | Notes |
|---|---|---|
exa.search(query, ...) | client.search(query, ...) → tavily-search | Exa's neural/keyword type param has no direct Tavily equivalent — Tavily's search_depth (basic/advanced) is the closest lever |
exa.get_contents(urls) | client.extract(urls) → tavily-extract | |
exa.find_similar(url) | No direct equivalent | Approximate with client.search() using the source page's title/topic as the query |
exa.search_and_contents(...) | client.search(..., include_raw_content=True) | One-call combined pattern, same idea |
| Firecrawl | Tavily | Notes |
|---|---|---|
firecrawl.scrape_url(url) | client.extract(urls=[url]) → tavily-extract | |
firecrawl.crawl_url(url, ...) | client.crawl(url, ...) → tavily-crawl | Firecrawl's limit/maxDepth map to Tavily's limit/max_depth |
firecrawl.map_url(url) | client.map(url) → tavily-map | |
| Firecrawl's markdown/structured extraction modes | Tavily's extract_depth="advanced" + format="markdown" |
| Perplexity | Tavily | Notes |
|---|---|---|
sonar/sonar-pro chat-completions call used for research | client.research(input, model="mini"/"pro") → tavily-research | Perplexity returns research as a chat completion; Tavily's research() is purpose-built, async, and supports output_schema for structured results |
| Inline citations in the completion text | response["sources"] + citation_format | Tavily returns citations as structured data rather than inline markers — update any citation-parsing code accordingly |
| Parallel | Tavily | Notes |
|---|---|---|
parallel-web-search | client.search(...) → tavily-search | |
parallel-web-extract | client.extract(...) → tavily-extract | |
parallel-deep-research | client.research(..., model="pro") → tavily-research | |
parallel-findall | build-entity-list | Same underlying idea — structured list of matching entities |
parallel-data-enrichment | fill-missing-fields | Same underlying idea — fill in fields on an existing list |
parallel-monitor | watch-for-changes | Parallel's is a hosted platform feature; Tavily's equivalent is orchestration-based (see that skill's notes) — call this difference out to the user rather than promising an identical hosted experience |
status / result (run ID lookup) | client.get_research(request_id) / tvly research status/poll | Built into tavily-research rather than separate calls |
parallel-memory | past-research-search | Tavily has no hosted memory store — this is a local logging convention, say so |
find_similar), say so plainly rather than approximating silently and letting behavior quietly change.