Skip to main content
Scira is built on an agentic architecture that breaks down complex research questions, selects the right tools, and delivers grounded answers with verifiable citations.

The Three-Step Process

1

Ask anything

You type a question, upload a PDF, or paste a URL. Pick a search mode or let Scira auto-select the best one for your query.Scira accepts:
  • Natural language questions
  • File attachments (PDF, CSV, DOCX, Excel)
  • URLs from any source (tweets, YouTube, articles, papers)
  • Multi-part research requests
2

Scira plans & retrieves

The agent breaks your question into sub-tasks, searches live sources, and cross-checks evidence. This happens automatically through multi-step workflows.Behind the scenes:
  1. Query decomposition into 3-10 targeted searches
  2. Parallel execution across multiple search providers
  3. Content extraction and verification
  4. Source deduplication and ranking
3

Get cited answers

You receive a grounded answer with inline citations. Click any source to verify it yourself. Every fact is traceable to its origin.

Agentic Planning

Scira uses AI-powered planning to handle complex, multi-step research workflows. Instead of single-query searches, it orchestrates entire research processes.

How Planning Works

When you ask a complex question, Scira’s planning system:
  1. Analyzes the question to identify key aspects and requirements
  2. Generates a research plan with specific sub-tasks and search queries
  3. Executes the plan by calling tools in sequence or parallel
  4. Iterates based on findings to fill gaps and verify information
For a question like “Compare the market position of Tesla vs Rivian in 2026”:
// Generated research plan
{
  plan: [
    {
      title: "Tesla Market Analysis",
      todos: [
        "Search Tesla Q4 2025 earnings report",
        "Find Tesla vehicle delivery numbers 2026",
        "Research Tesla market share electric vehicles"
      ]
    },
    {
      title: "Rivian Market Analysis",
      todos: [
        "Search Rivian production capacity 2026",
        "Find Rivian R1T and R1S sales data",
        "Research Rivian market positioning"
      ]
    },
    {
      title: "Comparative Analysis",
      todos: [
        "Compare Tesla vs Rivian market share",
        "Analyze competitive advantages both companies"
      ]
    }
  ]
}
Scira executes up to 15 actions automatically, using the right tools for each sub-task.

Extreme Search Mode

For maximum depth, Extreme mode uses an autonomous research agent with:
  • Multi-step planning with up to 15 actions
  • Code execution for data analysis and visualizations
  • Cross-source verification
  • Artifact storage in Cloudflare R2
// Extreme search can execute Python code
const research = await extremeSearch(prompt, dataStream, 'exa');

// Available Python libraries:
// pandas, numpy, scipy, keras, seaborn, matplotlib,
// transformers, scikit-learn
Extreme searches use either Exa or Parallel as the content extraction provider, with Firecrawl as fallback.

Tool Selection

Scira has 28 integrated tools. The agent automatically selects the right combination based on your question.

Search & Retrieval Tools

Searches academic papers using Firecrawl with category filtering.When used: Research questions, scientific queries, papersFeatures:
  • PDF parsing and extraction
  • Citation metadata
  • Research paper categorization
  • Up to 20 results per query
await academicSearch({
  queries: [
    "transformer attention mechanisms",
    "LLM reasoning capabilities"
  ],
  maxResults: [20, 20]
});
Searches X (Twitter) posts using xAI Grok with advanced filtering.When used: Social sentiment, breaking news, public opinion, trending topicsFeatures:
  • Date range filtering (start/end dates)
  • Handle filtering (include/exclude specific accounts)
  • Automatic tweet content extraction
  • Citation tracking
await xSearch({
  query: "GPT-5 reactions",
  startDate: "2026-01-01",
  endDate: "2026-03-04",
  xHandles: ["openai", "sama"],
  maxResults: 15
});
Autonomous agent that plans and executes comprehensive research.When used: Complex questions, multi-faceted analysis, in-depth researchFeatures:
  • AI-powered research planning
  • Up to 15 autonomous actions
  • Code execution for analysis
  • Multi-source verification
  • Chart generation
Tools available to extreme search:
  • webSearch — Search with category filtering
  • xSearch — X post search
  • codeRunner — Python execution in Daytona sandbox
  • Reddit Search — Community discussions via Parallel
  • YouTube Search — Video transcripts via Supadata
  • GitHub Search — Repository search via Firecrawl
  • Spotify Search — Music search via Spotify Web API
  • URL Retrieval — Extract content from any URL

Financial & Market Tools

Interactive stock charts with OHLC data, earnings, and news.Providers: Valyu (market data), Tavily + Exa (news)When used: Stock questions, financial analysis, market research
Real-time currency conversion via Valyu.When used: Exchange rate queries, currency comparisons
Cryptocurrency data via CoinGecko:
  • coin_data — Token information and metrics
  • coin_data_by_contract — Contract address lookups
  • coin_ohlc — OHLC charts and price history
When used: Crypto questions, token research, price analysis
Query prediction market data via Valyu with Cohere reranking.When used: Forecasting, probability estimation, market sentiment

Utility Tools

Execute Python code in a sandboxed Daytona environment.Available libraries: pandas, numpy, scipy, keras, seaborn, matplotlib, transformers, scikit-learnFeatures:
  • Chart generation (line, bar, scatter)
  • Data analysis and processing
  • Mathematical calculations
When used: Data analysis, calculations, visualizations
  • Weather — Current weather, 5-day and 16-day forecasts (OpenWeatherMap, Open-Meteo)
  • Maps & Geocoding — Location search and nearby places (Google Maps)
  • Flight Tracking — Real-time flight status
  • Movie/TV Search — TMDB search and trending lists
  • Text Translation — Multi-language translation with image text support
  • File Query Search — Semantic search over uploaded files (Cohere embeddings)
  • Connectors — Search Google Drive, Notion, OneDrive (Supermemory)
  • Memory — Save and retrieve personal context (Supermemory)
  • Code Context — Programming topic explanations (Exa Context API)
  • Date & Time — Current time with timezone support

Grounded Retrieval

Every answer Scira generates is grounded in verifiable sources. This prevents hallucinations and ensures accuracy.

How Citations Work

  1. Source collection — All tools return structured source data:
    {
      title: string,
      url: string,
      content: string,
      publishedDate?: string,
      author?: string,
      favicon: string
    }
    
  2. Deduplication — Sources are deduplicated by domain and URL to avoid redundancy
  3. Inline citations — The AI model cites sources inline as it generates the answer
  4. Clickable verification — Users can click any citation to view the original source
Scira uses deduplication to ensure you see diverse sources, not multiple results from the same domain.

Content Extraction Strategies

Different tools use different extraction methods: Exa Strategy:
  • Primary: Exa getContents with 3000 char limit
  • Fallback: Firecrawl scraping with markdown parsing
Parallel Strategy:
  • Primary: Parallel extract with full content
  • Fallback: Firecrawl scraping
Firecrawl Strategy:
  • Structured extraction with caching
  • PDF parsing support
  • Proxy rotation for reliability

Search Modes

Search modes determine which tools and models Scira uses. Each mode is optimized for specific question types.
No search tools. Direct conversation with the AI model.Best for: Brainstorming, creative tasks, casual conversation
Uses academic_search tool with Firecrawl research categorization.Best for: Scientific questions, research papers, citations
Uses x_search tool with xAI Grok.Best for: Trending topics, social sentiment, real-time discussions
Uses reddit_search tool via Parallel.Best for: User experiences, opinions, community discussions
Uses github_search via Firecrawl with repo metadata.Best for: Code repositories, open source projects, developer tools
Uses stock_chart tool with Valyu, Tavily, and Exa.Best for: Stock questions, earnings, market data
Uses crypto tools (coin_data, coin_ohlc) via CoinGecko.Best for: Cryptocurrency questions, token data, price analysis
Uses youtube_search via Supadata with transcript extraction.Best for: Video content, tutorials, channel research
Uses extreme_search tool with autonomous agent.Best for: Complex questions, multi-step analysis, comprehensive research
  • Code — Programming questions with code_context tool
  • Spotify — Music search
  • Prediction — Forecast markets via Polymarket/Kalshi
  • Connectors — Search connected knowledge bases (Pro)
  • Memory — Personal memory companion (Pro)
  • Voice — Real-time voice conversation (Pro)
  • XQL — Advanced X query language (Pro)

Architecture Overview

┌─────────────────┐
│  User Question  │
└────────┬────────┘


┌─────────────────┐
│  Mode Selection │ ◄── Auto-select or user choice
└────────┬────────┘


┌─────────────────┐
│ Agentic Planner │ ◄── Break down into sub-tasks
└────────┬────────┘


┌─────────────────────────────────────┐
│       Tool Selection & Execution     │
│  ┌──────┐ ┌──────┐ ┌──────┐         │
│  │ Tool │ │ Tool │ │ Tool │  ...    │
│  │  1   │ │  2   │ │  3   │         │
│  └──────┘ └──────┘ └──────┘         │
└────────┬────────────────────────────┘


┌─────────────────┐
│  Source Fusion  │ ◄── Deduplicate & rank
└────────┬────────┘


┌─────────────────┐
│ Answer Generation│ ◄── With inline citations
└────────┬────────┘


┌─────────────────┐
│  Cited Answer   │
└─────────────────┘

Model Support

Scira supports 100+ AI models across multiple providers:
  • xAI — Grok 3, Grok 4, Grok Code
  • OpenAI — GPT-4.1, GPT-5.x, o3, o4 mini
  • Anthropic — Claude 4.5/4.6 (Haiku, Sonnet, Opus)
  • Google — Gemini 2.5/3 (Flash, Pro)
  • Qwen, Mistral, DeepSeek, Cohere, and more
You can select the model for each conversation based on your needs.

Next Steps

Quickstart

Get started with Scira in under 5 minutes

Search Modes

Explore all 17 specialized search modes

Tool Reference

Complete reference for all 28 tools

Self-Hosting

Deploy Scira on your infrastructure

Build docs developers (and LLMs) love