The search API
that thinks.
One API call returns search + extracted content + citations — drop it into your RAG pipeline or AI agent. No scrapers, no plumbing.
Why developers choose Fetchium
A search API with federated backends, neural ranking, and cross-session learning built into a single retrieval pipeline.
See comparisonBuilt for every AI workflow
From real-time search to complex research agents, Fetchium provides the foundation for reliable AI output.
Capabilities no other search API has
Fetchium combines federation, ranking, extraction, and research primitives into one API surface for AI search workloads.
11-Backend Search Federation
Fan out a single query across SearXNG, Brave, GitHub, Reddit, HackerNews, StackOverflow, YouTube, Wikipedia, ArXiv, Bing, and DuckDuckGo in parallel. Adaptive Backend Selector picks the right mix per query intent.
HyperFusion Neural Ranking
8-signal ranking engine: BM25 lexical match, semantic similarity, temporal decay, domain authority, evidence density, source diversity, content depth, and cross-source consensus.
5-Layer CEP Content Extraction
Content Extraction Protocol: CSS selectors, Readability, headless JS rendering, PDF parsing, and screenshot OCR. Designed to recover structured text from a wide range of pages, including SPAs and PDFs.
QATBE Token Budget Control
Query-Aware Token-Budgeted Extraction scores every content segment with BM25 then solves a greedy knapsack to pack maximum relevance into your exact LLM context window. You always get the most useful content — not just the first N characters.
Deep Research Pipeline
AMRS multi-agent research swarm: 4 specialist agent types communicate over async channels, synthesize findings, and generate evidence graphs with claims traced to sources. Citation export supports APA, IEEE, BibTeX, and Chicago.
Production Resilience
Circuit breakers, bulkhead isolation, adaptive rate limiting, and backend-aware timeouts. Automatic failover and retry-and-refine validation help stabilize retrieval across heterogeneous sources.
YouTube & Social Intelligence
VideoFusion ranking for YouTube with transcript extraction and channel/video analysis. Native Reddit, HackerNews, and StackOverflow backends add community-signal retrieval alongside web search.
PIE Cross-Session Learning
Persistent Intelligence Engine tracks source trust, failure patterns, and query predictions across sessions via SQLite for deployments that enable persistence.
MCP Protocol Native
First-class Model Context Protocol support. Fetchium exposes 12 MCP tools spanning search, fetch, estimate, research, YouTube, and social workflows.
What developers build with Fetchium
From quick RAG prototypes to production research pipelines — one API handles every retrieval workload.
RAG Pipelines & AI Agents
Drop Fetchium into custom RAG stacks and AI agents over HTTP or MCP. One call returns search results, extracted content, and citations ready for downstream retrieval or prompting.
Deep Research Reports
The AMRS pipeline spawns 4 agent types in parallel, searches different angles, cross-validates findings, and assembles an evidence graph.
Content Monitoring & Diffs
Track pages, domains, or topics over time and compare fetched content across runs for change detection workflows.
YouTube Intelligence
Search, extract, and analyze YouTube content at scale. Get transcripts, metadata, engagement signals, and semantic summaries — all through the same unified API.
Social Intelligence
Pull structured data from Reddit and Hacker News alongside broader web retrieval to understand community signals around a topic.
How it works
Six stages. Search, extraction, ranking, and evidence tracing in one pipeline with an evidence graph.
Query Analysis
Stage 1Your query is fingerprinted, classified by intent, scored for complexity, and expanded with semantic variants. The system chooses the optimal backend mix before a single network call is made.
Multi-Backend Federation
Stage 2The Adaptive Backend Selector fans your query across up to 17 sources in parallel — SearXNG, Brave, GitHub, Reddit, StackOverflow, YouTube, and more. Circuit breakers handle backend failures invisibly.
HyperFusion Ranking
Stage 3Results are scored on 8 signals: BM25 lexical match, semantic similarity, temporal freshness, domain authority, evidence density, source diversity, content depth, and cross-source consensus.
CEP Content Extraction
Stage 4Top-ranked URLs are deep-extracted via the Content Extraction Protocol: CSS selectors, Readability, headless JS rendering, PDF parsing, and screenshot OCR.
Token Budget Control
Stage 5Extracted content is segmented, BM25-scored for query relevance, then packed into your token budget via greedy knapsack. You always get the most relevant content that fits your LLM context window.
AI-Ready Response
Stage 6The final response includes ranked results, extracted content within your budget, an evidence graph tracing every claim to a source, and auto-generated citations in APA, IEEE, BibTeX, or Chicago format.
First result in 60 seconds
Install the SDK, paste your key, ship. Real multi-source search with zero boilerplate.
1import { Fetchium } from "@fetchium/sdk";23const fetchium = new Fetchium({4 apiKey: process.env.FETCHIUM_API_KEY!,5 baseUrl: "https://api.fetchium.com",6});78// Multi-source federated search9const results = await fetchium.search("rust async programming", {10 backends: ["searxng", "brave", "github", "stackoverflow"],11 maxResults: 10,12 tier: "summary", // key_facts | summary | detailed | complete13 tokenBudget: 2000, // QATBE greedy-knapsack packing14 ranking: "hyperfusion" // 8-signal neural ranking15});1617console.log(results.items[0].title);18console.log(results.meta.tokensUsed); // always within budget19console.log(results.evidenceGraph); // citations + trust scores2021// Deep-extract any URL22const page = await fetchium.extract("https://docs.rs/tokio", {23 format: "markdown",24 tokenBudget: 4096,25 layer: "readability", // css | readability | headless | pdf | ocr26});
The async book covers futures, async/await syntax, Tokio runtime, and concurrent task management...
A runtime for writing reliable, asynchronous, and slim applications. Stars: 28k...
Accepted answer (1.2k votes): Rust's async/await desugars into state machines at compile time...
Fetchium capability shape at a glance.
This view focuses on first-party Fetchium capabilities and broad product-shape differences. It intentionally avoids hard benchmark and pricing claims for third-party services that can change independently of this repo.
| Feature | Best Value Fetchium | Tavily | Exa | SerpAPI | Firecrawl |
|---|---|---|---|---|---|
Multi-source federation 17+ simultaneous backends | |||||
Token budget control (QATBE) | |||||
5-layer content extraction (CEP) | |||||
8-signal neural ranking | |||||
Evidence graphs + citations | |||||
Cross-session learning (PIE) | |||||
Deep research pipeline (AMRS) | |||||
YouTube & social search | |||||
Real-time monitoring + diffs | |||||
MCP protocol support | |||||
Independent backend mix | |||||
Free tier (renewing) | |||||
Plan-based rate limits From current API auth configuration | 60-2000/min | varies | varies | varies | varies |
Free tier available 1,000 requests/month in current API auth configuration |
Fetchium values in this table are tied to the current codebase and auth configuration. Non-Fetchium entries are shown as broad capability comparisons only.
Full pipeline means search, extraction, ranking, citations, and research workflows in one product surface.
Plans synced to the current auth configuration
Free tier limits are sourced from the API auth layer. Paid plan names and request ceilings below reflect the current codebase; contact sales for commercial pricing details.
Explore the API, build a prototype, or run personal projects.
- 1,000 API requests per month
- All 17 search backends
- 5-layer CEP content extraction
- HyperFusion 8-signal ranking
- Token budget management
- Evidence graphs + citations
- MCP protocol tools
- Community support (Discord)
First paid tier in the current API configuration.
- 25,000 API requests per month
- Everything in Free
- YouTube intelligence API
- Social media research
- Async jobs + usage tracking
- Admin key management
- Usage dashboard
source: Current auth limits: 200 req/min and 25,000 req/month
Higher-volume production usage in the current API configuration.
- 250,000 API requests per month
- Everything in Starter
- Deep research pipeline (AMRS)
- Cross-session learning (PIE)
- Adversarial content shield
- Higher monthly quota
- Higher per-minute limits
- Best fit for production workloads
source: Current auth limits: 500 req/min and 250,000 req/month
Custom volume and support for teams that need enterprise handling.
- Unlimited API requests
- Everything in Pro
- Custom commercial terms
- Priority onboarding
- Dedicated support channel
- Deployment guidance
- Security review on request
What is verified here
This section is intentionally conservative. Request quotas, rate limits, and free-tier availability are taken from the current API auth code. If you need a signed commercial quote, use the contact flow.
Source: current API auth configuration in the Fetchium codebase.
Questions about pricing? Talk to us or see the full pricing FAQ. All plans include all 17 algorithms, all backends, and evidence graphs — only scale differs.
Frequently asked questions
Everything you need to decide if Fetchium is right for your project.
Still have questions? Contact us or join our Discord community.
Start building with Fetchium today
Join the open beta. 1,000 free API requests per month, all features included. Upgrade when you need more with higher-volume Starter, Pro, and Enterprise tiers.