Use case · AI assistants

Live web answers for assistants, with their sources

Ground replies in pages fetched at question time. The Tavily-compatible endpoint returns ranked results plus an answer quoted from them, so your model cites instead of guessing.

Try it free, no key Get a key

Assistantchat

How does daylight saving time work in the EU?

POST /tavily/search3.1 s

{ "query": "how does daylight saving time work in the EU", "include_answer": true, "max_results": 5 }

answer · extractivequoted, no LLM

In the European Union, Summer Time begins and ends at 1:00 a.m. Universal Time (Greenwich Mean Time). It begins the last Sunday in March and ends the last Sunday in October. In the EU, all time zones change at the same moment.

[1] en.wikipedia.org[2] en.wikipedia.org[3] webexhibits.org[4] en.wikipedia.org

Recorded keyless call to api.contextfetch.dev/tavily/search on 17 Oct 2026. The answer passage, sources and timing are the real response.

Compatible

Keep tavily-python, change one argument

Point the client at ContextFetch and your existing search calls keep working, including answers, raw page content and domain filters.

Start with no key: pass "keyless" as the API key for 25 free requests a day.

Compare with Tavily

assistant.py

  from tavily import TavilyClient  - tavily = TavilyClient(api_key="tvly-...")+ tavily = TavilyClient(api_key="YOUR_API_KEY", api_base_url="https://api.contextfetch.dev/tavily")

Extractive answers

Answers quoted from the sources, not generated

The answer is up to three of the best-scoring passages across the results, quoted word for word. Every sentence your assistant repeats can be traced to a URL.

If nothing scores high enough, the answer is null, which is a better signal than a confident guess.

results · "how does daylight saving time work in the EU"

Daylight saving timeen.wikipedia.org0.995
Daylight saving time in Asiaen.wikipedia.org0.993
Daylight Saving Time - When do we change our clocks? - Webexhibitswebexhibits.org0.990
Daylight saving time by countryen.wikipedia.org0.970
Analysis of daylight saving timeen.wikipedia.org0.762

Relevance scores from the recorded response, 0 to 1.

Keyless start

Free to start, one link when it is time to pay

Every keyless response tells your assistant how many free requests are left today. When they run out, the error carries a one-click checkout link it can show the user.

Credits are prepaid and never expire. Auto-reload is optional and capped per month.

The 402 loop

notice, from the recorded response

Free keyless tier (10/25 remaining today). Get 15,000 full-markdown searches for $15 (Apple Pay / Google Pay).

HTTP/2 402 Payment Required

{
  "error": "keyless_limit",
  "action": {
    "type": "payment_link",
    "url": "https://api.contextfetch.dev/buy/instant"
  }
}

Code

Answers and sources in one request

Ask for include_answer and read the quoted answer next to the ranked results, each with its URL, title, content and score.

include_raw_content returns each page as Markdown
include_domains and exclude_domains, up to 100 each
Basic and advanced depth both cost one credit

Tavily-compatible docs

# pip install tavily-pythonfrom tavily import TavilyClient tavily = TavilyClient(    api_key="YOUR_API_KEY",  # or "keyless" for the free tier    api_base_url="https://api.contextfetch.dev/tavily",) res = tavily.search(    "how does daylight saving time work in the EU",    include_answer=True,    max_results=5,)print(res["answer"])for r in res["results"]:    print(r["url"], r["score"])

// npm i @tavily/coreimport { tavily } from "@tavily/core"; const tvly = tavily({  apiKey: "YOUR_API_KEY",  apiBaseURL: "https://api.contextfetch.dev/tavily",}); const res = await tvly.search("vector database benchmarks 2026", { searchDepth: "basic", maxResults: 5 });for (const r of res.results) console.log(r.url, r.score);

curl -s https://api.contextfetch.dev/tavily/search \  -H "Content-Type: application/json" \  -d '{"query":"how does daylight saving time work in the EU","include_answer":true,"max_results":5}'

Pricing

Tavily-compatible search

$1.00per 1,000 searches

Answers, raw content and domain filters are in the same credit. 25 searches a day are free with no key. One-time payment. Credits never expire. Full Markdown and pruning included.

Buy 15,000 for $15 All pricing

Common questions

Does tavily-python work with ContextFetch?

Yes. Set api_base_url to https://api.contextfetch.dev/tavily and keep your code. search() accepts query, search_depth, max_results (up to 25), include_answer, include_raw_content, include_domains and exclude_domains. Errors use Tavily's shape, so the SDK raises the exceptions it already knows.

Is the answer written by a language model?

No. No language model is called anywhere in the API. The answer is extractive: up to three of the best-scoring passages across the results, quoted verbatim. When no passage scores at least 0.1, answer is null, so your assistant can say it found nothing instead of guessing.

How fresh are the results?

Pages are fetched when the question is asked. The same query within 15 minutes is answered from a short query cache, and a fetched page can be reused for up to 48 hours.

What happens when the free tier runs out?

Keyless callers get 25 requests a day per IP. After that the API answers with a payment error that carries a one-click link to buy credits (on the Tavily route it is HTTP 432, as tavily-python expects). Your assistant can show that link to its user.

How much does ContextFetch cost for an assistant?

One credit per search, basic or advanced depth: $1.00 per 1,000 on the $15 pack and less on bigger packs. Credits are a one-time purchase and never expire, so there is no per-seat or monthly plan.

Do you keep the questions my users ask?

A query and its results stay in a 15-minute cache that is not tied to who asked. We keep no request logs of queries and do not train models on them. The privacy page lists everything we store.

Related use cases

Give your assistant the web, with receipts

Start with no key: 25 free searches a day.

Try it free Get a key