Providers

seek is a thin, uniform layer over several provider APIs. Commands are grouped by intent (search, research, scrape); within a command you pick the provider whose trade-offs fit your query. Every provider requires its own API key — see Configuration.

Provider	Commands	Best for
Parallel	web, research, scrape	Dense, LLM-ready context; strict-JSON deep research; extraction
Exa	web, scrape	Neural semantic search; bulk clean page text
Brave Search	web brave search	Real-time facts from an independent global index
Brave Answers	web brave answers, research brave	AI answers with verifiable sources; deep research
Perplexity	web, research	Short cited answers; polished long-form briefings
Google/Gemini	research google	Long-horizon, Google-grounded reports
Firecrawl	scrape firecrawl	Clean Markdown/HTML/JSON/screenshots from any live URL

Parallel

The most versatile provider — used by all three commands.

• web — returns relevant excerpts optimized for LLMs, replacing multiple keyword searches with a single call for broad or complex queries.
• research — Parallel Deep Research with tunable processor depth (core-fast … ultra) and optional strict JSON output via a schema file. Best for B2B and data-ops work where output must be auditable. Supports resumable --async submission plus later research status and research get calls.
• scrape — the /v1/extract endpoint returns clean markdown excerpts (and optional full content) plus publish dates, and can authenticate through logins and paywalls.

Exa

Neural semantic search that finds pages by meaning rather than keywords.

• web — best for conceptual queries; supports search types (auto … deep-reasoning), category filters, and date windows.
• scrape — bulk page parsing that returns stripped text optimized for LLM token windows.

Brave

Brave exposes two separate APIs, provisioned and keyed independently.

• Brave Search (web brave search) — clean, structured snippets from an independent global web index; best for real-time consumer facts, with optional freshness and country filters.
• Brave Answers (web brave answers, research brave) — AI-generated answers backed by verifiable sources. The same endpoint powers deep research when the research flag is enabled. Research remains synchronous because the streaming endpoint does not expose a resumable job id.

Perplexity

Synthesizes the web into cited answers.

• web — short, conversational answers with citations; model, recency, search mode (web/academic), and domain filters are configurable.
• research — Sonar Deep Research (async job) for polished long-form briefings with inline citations and a tunable reasoning-effort level. Supports resumable submission and later status/result retrieval.

Google (Gemini)

• research google — the Gemini Deep Research Agent autonomously plans, executes, and synthesizes multi-step research into a detailed, cited report. Runs can take several minutes; resumable submission and an interactive plan-approval mode are available. With --interactive --async, the plan is approved before the provider job is submitted.

Resumable provider jobs

Google, Parallel, and Perplexity expose durable provider jobs. seek research <provider> <query> --async returns a provider-qualified locator without polling. seek research status <provider> <job-id> and seek research get <provider> <job-id> reconnect directly to that remote job; no local daemon or job-history database is involved. See Research for command examples and the JSON lifecycle contract.

Firecrawl

• scrape firecrawl — turns any live URL into clean Markdown, HTML, raw HTML, links, a summary, a screenshot, or LLM-extracted JSON, with controls for JS rendering, mobile viewport, ad blocking, proxy modes, and caching.

Choosing a provider

• Not sure which to use? Start with Parallel for web and research, and Firecrawl for scrape.
• Running a command without the matching key prints the missing-key error plus a ranked list of sibling providers, marking which ones you already have keys for.