Nimble for Lightfield

Use case 1 · Extract

Extract returned the full text of 20 startup pages in about a second

Extract takes one URL and returns the page as clean markdown. Your LLM reads the markdown and answers the question.
We listed each site's pages with Nimble Map and picked up to four of them: home, about, product, and customers or pricing.
Every page ran twice with each driver, in opposite orders, for 80 calls in total.
A driver is the way Nimble fetches a page. The options range from a plain HTTP request to a stealth browser that renders JavaScript.

Max returned pages 3.8 times faster than Auto at 2.5 times the price

	Auto	Max
How you set it	`"render": "auto"`	`"driver": "max"`
How it works	Starts with the cheapest driver and moves to a browser or stealth only when a request fails.	Sends the request to every driver at once and returns the first good page.
Best for	Large batches where cost matters.	Agent tool calls and any step where a person or an agent waits.
Full pages returned	40 of 40	39 of 40
Median time	2.1 s	0.56 s
9 calls in 10 under	9.7 s	1.7 s
Slowest call	15.4 s	10.9 s
List price	$1.00 per 1,000 pages	$2.50 per 1,000 pages

Max returned an error page in 1 of 40 calls (Hyphenate About, 14 words, marked * below). Check the word count of each page, and retry with Auto when a page has fewer than 50 words.

We recommend Max for your agents and Auto for large batches

Your extraction workflow runs about 2,000 calls a week. Max returns those pages in about a second each.
Your search workflow runs about 280,000 calls a week. If it moves to Extract, Auto keeps the price at $1.00 per 1,000 pages.
A plain HTTP request returned the full text of all eight sites. Auto moves to a browser only on sites that need one.

Extract time per page, two runs per driver

Company	Page	Auto	Max	Words
Lightfield	Home	3.6 s · 2.1 s	0.89 s · 0.74 s	1,273
Lightfield	About	2.6 s · 0.94 s	0.53 s · 0.29 s	391
Lightfield	Product	1.8 s · 1.2 s	0.43 s · 0.60 s	753
Lightfield	Customers or pricing	2.0 s · 1.0 s	10.9 s · 0.52 s	888
TeddyHoldings.AI	Home	8.1 s · 15.4 s	0.35 s · 0.32 s	416
Leela	Home	1.7 s · 13.7 s	0.36 s · 1.0 s	770
Leela	About	2.3 s · 2.0 s	0.30 s · 0.84 s	624
Leela	Product	1.1 s · 1.8 s	0.39 s · 0.96 s	886
Mave	Home	9.7 s · 2.1 s	0.63 s · 0.92 s	1,377
Mave	About	1.9 s · 8.0 s	0.42 s · 0.76 s	683
Mave	Product	0.74 s · 2.7 s	0.54 s · 1.7 s	778
Mave	Customers or pricing	1.3 s · 3.6 s	0.34 s · 1.4 s	589
Rein Security	Home	11.8 s · 4.4 s	0.42 s · 2.3 s	1,003
Hyphenate	Home	8.1 s · 4.2 s	0.28 s · 2.3 s	1,125
Hyphenate	About	11.7 s · 3.7 s	0.42 s · 1.7 s*	876
Hyphenate	Product	8.0 s · 2.5 s	0.32 s · 1.4 s	1,010
Machenta	Home	1.6 s · 2.0 s	0.70 s · 1.1 s	190
Volantis	Home	1.6 s · 1.2 s	0.24 s · 0.33 s	330
Volantis	About	3.5 s · 1.6 s	0.53 s · 0.95 s	374
Volantis	Product	1.9 s · 0.98 s	0.34 s · 0.58 s	258

Extract returns the Leela home page as this markdown

# Max: fastest page, for agent tool calls
curl -X POST "https://sdk.nimbleway.com/v2/extract" \
  -H "Authorization: Bearer $NIMBLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://withleela.com", "driver": "max", "formats": ["markdown"]}'

# Auto: cheapest page, for large batches
curl -X POST "https://sdk.nimbleway.com/v2/extract" \
  -H "Authorization: Bearer $NIMBLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://withleela.com", "render": "auto", "formats": ["markdown"]}'

# Compliance that runs itself.

Leela builds and runs compliance programs for regulated businesses.

[Get started](https://app.withleela.com/signin) [Talk to a human first](https://withleela.com/contact-sales)

![Leela’s Overview for a payments company: 9 actions need an answer and 4 are overdue, Leela suggests where to start, and 848 steps were finished in the past 30 days, 224 of them by Leela.](https://withleela.com/assets/product/tour-overview-1440.webp?v=bf4029d414f2) 9 actions need an answer. Leela picks where to start. Leela started this week’s screening work itself. 848 steps in 30 days. 224 of them by Leela. Leela can run 263 of 264 obligations. Every policy running, every obligation ready. Leela says what to do next, and why. 49 steps Leela did itself this week. Chair sends each decision to the right person. Leela says: not a false positive. Fill puts the answer in. You check it and record it. How far the evidence leans, 91 to 9. Each reason names the system it came from. 5 evidence packs due by the end of this week. One for each party: the bank, the board, regulators. Prepared by Leela, waiting for a person’s approval. 28 of 38 deadlines on time, every exception named. Send it as a PDF, or as Word to edit.

## Software gives you data. Leela does the work.

A new ruleCustomer reviewsA new corridor

Other compliance softwareStopped at step 1 of 6. A task goes to your team.

✓ Spot the rule change

Read it against your obligations Your team

Draft the policy edits Your team

Get your CCO's approval Your team

Update controls and training Your team

File the evidence Your team

Leela6 of 6 done. Your team approves.

✓ Spot the rule change CFPB rule · published Tue 08:10

✓ Read it against your obligations 612 checked · 3 affected

✓ Draft the policy edits 3 edits to AML Policy §4

✓ Get your CCO's approval Approved Wed 10:40

✓ Update controls and training 2 controls · 1 module

✓ File the evidence Workpaper on record

Use case 2 · Web Search Agent

One agent call answers what a company does, what it sells and who it sells to

A Web Search Agent is a Nimble agent that searches the web, opens pages and returns JSON in the shape you define.
Our agent startup_company_profile takes a website domain. It reads the home, product, pricing and customer pages, then the company's LinkedIn page and its latest funding news.
It returns four sections: company, what_they_sell, who_they_sell_to and linkedin.
Every claim cites the page it came from. Nimble grades each citation and gives the run a confidence level.
The LinkedIn numbers below were read live on 3 November 2026.

8 of 8

Runs completed

76 s

Median run time (42–137 s)

Median claims cited per run

$0.10

Per run at medium effort

Pick a company to see the agent's answer

See the raw JSON for this company

Start in your account

The agent and both drivers are ready in your lightfield account

Your account lightfield has 20,000 data requests and 500 agent runs until 30 November 2026.
The agent startup_company_profile is already in your account. You can open and edit it at app.nimbleway.com under Web Search Agents.
Your API key is under Account settings in the same app.

Step 1

Extract a page with Max or Auto

# Max: fastest page, for agent tool calls
curl -X POST "https://sdk.nimbleway.com/v2/extract" \
  -H "Authorization: Bearer $NIMBLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://withleela.com", "driver": "max", "formats": ["markdown"]}'

# Auto: cheapest page, for large batches
curl -X POST "https://sdk.nimbleway.com/v2/extract" \
  -H "Authorization: Bearer $NIMBLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://withleela.com", "render": "auto", "formats": ["markdown"]}'

Step 2

Run the agent on a domain

# 1. Start a run. The input is a website domain.
curl -X POST "https://sdk.nimbleway.com/v2/agents/runs" \
  -H "Authorization: Bearer $NIMBLE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"agent_name": "startup_company_profile", "input": "withleela.com"}'
# returns {"id": "task_run_...", "web_search_agent_id": "wsa_...", "status": "queued"}

# 2. Check the status until it says "completed" (about a minute)
curl "https://sdk.nimbleway.com/v2/agents/$WSA_ID/runs/$RUN_ID" \
  -H "Authorization: Bearer $NIMBLE_API_KEY"

# 3. Read the JSON answer and its citations
curl "https://sdk.nimbleway.com/v2/agents/$WSA_ID/runs/$RUN_ID/result" \
  -H "Authorization: Bearer $NIMBLE_API_KEY"

Python

The same Extract calls in Python, with the word-count check

from nimble_python import Nimble

nimble = Nimble(api_key="YOUR-API-KEY")

# Fastest page for an agent tool call
page = nimble.extract.run(url="https://withleela.com", driver="max", formats=["markdown"])

# Cheapest page for a large batch
page = nimble.extract.run(url="https://withleela.com", render="auto", formats=["markdown"])

text = page.data.markdown
if len(text.split()) < 50:          # an error page or an empty shell
    page = nimble.extract.run(url="https://withleela.com", render="auto", formats=["markdown"])

We suggest two tests for this month

Send 50 domains from your own pipeline through the agent and compare its answers with what Exa returns today.
Run one day of your 280,000-a-week search workflow through Extract with Auto, and compare the page text with your current results.