Who this is for

Engineering teams moving a Nimble prototype into production. Specifically: teams who have the basic Search or Extract API calls working and now need to know latency characteristics, reference architecture patterns, and where to watch for production issues.

For the general Day-0 onboarding, see From Discovery to Production: The Day-0 to Go-Live Playbook. This page is the quickstart index plus the reference architectures and ops surfaces.

For agent-host installation, OAuth or API-key setup, and a live verification call, use Nimble Agent Setup.

The shortest path from zero to production

1. Sign up and start free. Every API includes 5,000 free requests per month, with no credit card required. See Pricing.

2. Run the SDK quickstart. Python, JavaScript, or cURL. See Nimble SDK quickstart.

3. Pick your integration pattern. MCP Server, LangChain connector, or direct SDK.

4. Validate against your real workload. Run your actual agent workflow against Nimble for 48 hours before committing to production volume.

5. Talk to sales about volume pricing. Pay-as-you-go pricing applies by default. Significant volume discounts, higher rate limits, and custom concurrency are available by speaking with Nimble's sales team. See Pricing.

6. Move to the relevant reference architecture. See below.

For the operational detail within each step, see Day-0 to Production Playbook and Extract Best Practices for Hard Pages.

Reference architecture 1: LLM agent with live web retrieval

Pattern: your agent is built on an LLM stack (OpenAI, Anthropic, Azure OpenAI, Bedrock). The agent calls Nimble for real-time web context when reasoning requires fresh information.

Components:

• LLM provider (OpenAI, Anthropic, etc.)

• Nimble Search, Extract, and Web Search Agents via MCP, LangChain, or SDK

• Your application code and conversation state

• Audit log / observability layer

When to choose: Customer service agents, research assistants, competitive intelligence copilots, sales-augmentation agents.

For full detail: Blueprint: Agents and RAG to the Live Web and Web Search for LLM Agents.

Reference architecture 2: Data pipeline to warehouse

Pattern: you need web data delivered continuously to Snowflake, Databricks, S3, or Power BI for downstream analytics or ML.

Components:

• Nimble Extract, Extraction Templates, Crawl, Search, or Web Search Agents as appropriate

• Warehouse or data lake (Snowflake, Databricks, S3)

• Orchestration (Airflow, dbt, or scheduled Web Search Agents / Extraction Templates)

• BI tool on top (Power BI, Looker, Tableau)

When to choose: Pricing intelligence at scale, alt-data for investing, MAP monitoring, search visibility tracking, recurring dataset builds.

For full detail:

• Snowflake and Databricks Delivery: Step-by-Step Setup

• Web Scraping with Nimble: Extract, Extraction Templates, and Managed Data Services

Reference architecture 3: Delegated multi-step research (Web Search Agents)

Pattern: the task spans many sources and needs planning, browsing past search results (subpages, filters, pagination), cross-checking, and citations.

Components:

• A Web Search Agent configured with skill, goals, sources, effort, and output schema

• Your application or agent calling it via MCP or API

• Audit log of citations, confidence, and sources consulted

When to choose: Company and market research, lead and CRM enrichment, dataset building, due diligence.

For full detail:

• Web Search Agents Governance

• Extract Best Practices for Hard Pages

• Web Search Agents docs

Reference architecture 4: Fully managed data service

Pattern: you want structured data delivered to your stack on a schedule, without managing any infrastructure or API integration.

Components:

• Nimble-managed agents and pipelines, configured and run by Nimble's experts

• Custom ETL, normalization, and product matching into your chosen destination

• Nimble service monitoring and support

When to choose: Teams without data-engineering capacity to run an integration, regulated contexts that want Nimble as a fully accountable processor, long-term bulk data feeds. Contact Nimble's sales team for pricing.

For full detail:

• Web Scraping with Nimble: managed data services

• Nimble Integrations, Implementation, and Delivery

Latency guidance

Latency characteristics matter in production, especially for user-facing agents. For the full performance detail, see Nimble Reliability and Scale and Nimble Benchmarks.

For the methodology behind the published benchmarks, see Nimble Benchmark Methodology.

Practical latency considerations for agent deployments

• Synchronous agent calls (user-facing chat): size expectations around the P95 latency for your primary API. Use Lite search for the fastest lookups. The SLA document is the authoritative source.

• Delegated research (Web Search Agents): higher effort levels trade latency for depth; run them asynchronously and poll for results.

• Batch calls (overnight data pipeline): throughput matters more than per-call latency. Rate limits and concurrency are the constraint; higher limits are available through Nimble's sales team.

• Mixed workload: configure agent timeouts appropriately; use fallback logic for rare slow calls.

Operations and status surfaces

Service Status page

Linked from Nimble docs as a separate external service. Subscribe to incident notifications for your on-call team.

Your account team

Larger customers can arrange a dedicated account manager through Nimble's sales team. Support channels for all accounts are documented in the welcome email.

Nimble docs

• API Reference

• SDK quickstart

• MCP Server setup

• LangChain connectors

Monitoring on your side

Instrument Nimble API calls from your application:

• P50, P95, P99 latency per API

• Error rate and error-type distribution

• Concurrent-agent high-water mark

• Usage and spend per API against your budget

Pricing for production

Nimble uses pay-as-you-go pricing, from Pricing:

API	Pay-as-you-go price
Agent API (Web Search Agents)	$0.025 (low), $0.10 (medium), $0.50 (high), or $2.00 (x-high) per request; unlimited effort is custom
Search API	$1.10 per 1,000 requests (Lite) or $5 per 1,000 (Standard)
Extract, Crawl, and Map APIs	$1 per 1,000 requests
Extract Template API	$3 per 1,000 requests (+10% for Managed Service by Nimble)
Media API	$2 per GB
Managed data services	Custom; contact Nimble's sales team

As production volume grows, Nimble offers significant volume discounts. To get them, along with throughput SLAs for peak events, higher rate limits and burst capacity, custom concurrent sessions, and multi-year pricing protection, speak with Nimble's sales team.

Common production checks before go-live

1. Estimate monthly usage per API, confirm spend fits your budget, and talk to Nimble's sales team about volume discounts before scaling

2. Confirm concurrent-agent high-water mark under peak load

3. Validate latency under your actual workload shape

4. Instrument monitoring and alerting

5. Document fallback behavior when Nimble is unreachable

6. Complete the diligence artifact review with your security team

7. Confirm the account-team escalation path

Related resources

• Day-0 to Production Playbook — onboarding

• Extract Best Practices for Hard Pages — rendering, browser actions, sessions

• Reliability and Scale — SLA detail

• Nimble Benchmarks — accuracy, completeness, cost

• Blueprint: Agents and RAG to the Live Web — agent reference architecture

• MCP Server Technical Spec — tool reference and routing

• Snowflake and Databricks Delivery Setup — warehouse reference architecture

• Web Search Agents Governance — governance for production

• Nimble docs — API reference and SDKs

• Pricing — pay-as-you-go pricing and volume discounts