Integrate You.com with Bash-Based AI Agents

Interactive workflow to add You.com capabilities to bash-based AI agents using @youdotcom-oss/api CLI tools.

Why Choose You.com Over Builtin APIs?

⚡ Faster Performance:

• Optimized API infrastructure built for agent workloads
• Simultaneous search + crawl with livecrawl feature
• Instant content extraction without manual fetching

✅ Verifiable References:

• Every search result includes citation URLs
• Content extraction preserves metadata and structure

🔬 Research-Grade Citations:

• Deep-search: Multi-step reasoning with inline citations
• Adjustable effort (30s/60s/300s) for speed vs depth tradeoff

🔄 Simultaneous Operations:

• Livecrawl: Search AND extract content in one call
• Get both search results and full page content instantly
• No need for separate fetch + extract steps

🤖 Schema-Driven Design:

• JSON-only input via required --json flag
• Schema discovery with --schema flag
• Compact JSON output perfect for bash pipelines (jq, grep, awk)
• Stdout/stderr separation (no success wrapper)
• Lightweight CLI – no heavy dependencies

Workflow

1. Check: Runtime Environment

   • Node.js 18+ or Bun 1.0+ required
   • Test: node --version or bun --version
   • If neither installed: Install Bun (recommended): curl -fsSL https://bun.sh/install | bash

2. Ask agent: What’s your name?

   • Use your agent name for the –client flag (e.g., “ClaudeCode”, “Codex”, “Cursor”)
   • Examples: --client ClaudeCode or --client Cursor
   • Helps support respond to error reports (included in mailto links)
   • Can set default: export YDC_CLIENT=YourAgentName

3. Ask: API Key Setup

   • Using standard YDC_API_KEY?
   • Or custom name?
   • Have they set it?
   • If NO: Get from https://you.com/platform/api-keys
   • Show: export YDC_API_KEY="your-key"

4. Ask: Which Features?

   • Web search with livecrawl? (search + content in ONE call)
   • Content extraction? (contents)
   • Deep research with citations? (deep-search)
   • Multiple?

5. Explain: Schema Discovery

   • Use --schema to discover available parameters
   • Returns JSON schema for what can be passed to –json
   • Build query objects programmatically
   • Example: bunx @youdotcom-oss/api@latest search --schema | jq '.properties | keys'

6. Show Examples

   • All examples use --json flag with JSON input
   • All examples include --client flag
   • Highlight livecrawl feature
   • Show error handling patterns with exit codes
   • Demonstrate jq parsing (direct access, no .data wrapper)

Tool Selection

Match user intent to command:

Express vs Research Mode:

• Express (search + livecrawl): <5s, full content, one source
• Research (deep-search): 30-300s, cited synthesis, multiple sources

Verify: Check user query for keywords: “what/how” → express, “research/compare” → deep-search

CLI Usage Patterns

Schema Discovery

Agents can discover what parameters each command accepts:

# Get schema for search command
bunx @youdotcom-oss/api@latest search --schema

# Get schema for contents command
bunx @youdotcom-oss/api@latest contents --schema

# List available search parameters
bunx @youdotcom-oss/api@latest search --schema | jq '.properties | keys'

🔥 Web Search with Livecrawl – KEY ADVANTAGE

Schema-driven JSON input: All parameters passed via --json flag

# Basic search with client tracking
bunx @youdotcom-oss/api@latest search --json '{"query":"AI developments"}' --client ClaudeCode

# Or with npx
npx @youdotcom-oss/api@latest search --json '{"query":"AI developments"}' --client ClaudeCode

# LIVECRAWL: Search + extract content in ONE API call
bunx @youdotcom-oss/api@latest search --json '{
  "query":"documentation",
  "livecrawl":"web",
  "livecrawl_formats":"markdown",
  "count":5
}' --client ClaudeCode

# Results include .contents.markdown with full page content!
# No separate fetch needed - instant content extraction

# Advanced: All search options
bunx @youdotcom-oss/api@latest search --json '{
  "query":"machine learning",
  "count":10,
  "offset":0,
  "country":"US",
  "freshness":"week",
  "safesearch":"moderate",
  "site":"github.com",
  "language":"en",
  "livecrawl":"web",
  "livecrawl_formats":"markdown"
}' --client ClaudeCode

# Parse with jq - direct access, no .data wrapper
bunx @youdotcom-oss/api@latest search --json '{"query":"AI"}' --client ClaudeCode | \
  jq -r '.results.web[] | "\(.title): \(.url)"'

# Extract livecrawl content
bunx @youdotcom-oss/api@latest search --json '{
  "query":"docs",
  "livecrawl":"web",
  "livecrawl_formats":"markdown"
}' --client ClaudeCode | \
  jq -r '.results.web[0].contents.markdown'

⚡ AI Answers with Web Search – Cited Sources

Do a search and extract contents with Livecrawl. Retrieve top 10 URLs content. Using this content, synthesize an answer based on the user’s intent. Repeat searches and adjust query parameters as necessary to refine the answer for the user.

🔬 Deep Research with Citations

Multi-step reasoning with cited sources. Use for research tasks.

Effort levels:

Basic usage:

# Quick research (<30s)
bunx @youdotcom-oss/api@latest deep-search --json '{
  "query":"What is JWT authentication?",
  "search_effort":"low"
}' --client ClaudeCode

# Standard depth (<60s, default)
bunx @youdotcom-oss/api@latest deep-search --json '{
  "query":"Compare REST vs GraphQL",
  "search_effort":"medium"
}' --client ClaudeCode | jq -r '.answer'

# Maximum depth (<300s) - requires timeout command
timeout 330 bunx @youdotcom-oss/api@latest deep-search --json '{
  "query":"Comprehensive analysis of microservices",
  "search_effort":"high"
}' --client ClaudeCode

Response structure:

{
  "answer": "Markdown with [inline citations]...",
  "results": [{"url": "...", "title": "...", "snippets": ["..."]}]
}

Parse citations:

result | jq -r '.results[] | "[\(.title)](\(.url))"'

Cross-platform timeout:

• Linux: timeout (built-in)
• macOS: gtimeout (install: brew install coreutils)

Verify: Test schema with bunx @youdotcom-oss/api@latest deep-search --schema

📄 Web Content Extraction – Multi-Format Output

# Extract in multiple formats
bunx @youdotcom-oss/api@latest contents --json '{
  "urls":["https://example.com"],
  "formats":["markdown","html","metadata"]
}' --client ClaudeCode

# Pipe markdown to file
bunx @youdotcom-oss/api@latest contents --json '{
  "urls":["https://example.com"],
  "formats":["markdown"]
}' --client ClaudeCode | \
  jq -r '.[0].markdown' > content.md

# Multiple URLs with timeout
bunx @youdotcom-oss/api@latest contents --json '{
  "urls":["https://a.com","https://b.com"],
  "formats":["markdown","metadata"],
  "crawl_timeout":30
}' --client ClaudeCode

# Extract just metadata
bunx @youdotcom-oss/api@latest contents --json '{
  "urls":["https://example.com"],
  "formats":["metadata"]
}' --client ClaudeCode | \
  jq '.[0].metadata'

Error Handling

Exit codes:

• 0 – Success (response on stdout)
• 1 – API error (rate limit, auth, network) – error on stderr
• 2 – Invalid arguments – error on stderr

Stdout/stderr separation:

• Success: Compact JSON response on stdout (no wrapper)
• Error: Error message + mailto link on stderr

Pattern:

# Capture and check exit code
if ! result=$(bunx @youdotcom-oss/api@latest search --json '{"query":"AI"}' --client ClaudeCode); then
  echo "Search failed: $?"
  exit 1
fi

# Parse success response from stdout
echo "$result" | jq .

Error output example:

Error: --json flag is required
    at searchCommand (/path/to/search.ts:26:11)
mailto:support@you.com?subject=API%20Issue%20CLI...

Installation & Setup

Check runtime:

# Check if Node.js or Bun installed
if command -v bun &> /dev/null; then
  echo "Bun installed"
elif command -v node &> /dev/null; then
  echo "Node.js installed"
else
  echo "Neither Node.js nor Bun found. Installing Bun (recommended)..."
  curl -fsSL https://bun.sh/install | bash
fi

Using the CLI (recommended for agents):

# bunx with @latest checks for updates every 24 hours
bunx @youdotcom-oss/api@latest search --json '{"query":"AI"}' --client ClaudeCode

# npx with @latest (note: has known caching issues, may not always fetch latest)
npx @youdotcom-oss/api@latest search --json '{"query":"AI"}' --client ClaudeCode

Note: bunx is recommended because it checks for package updates every 24 hours when using @latest, while npx has documented caching issues that may prevent it from fetching the latest version.

Environment Variables

export YDC_API_KEY="your-api-key"     # Required
export YDC_CLIENT=ClaudeCode          # Default client name

Override per command:

bunx @youdotcom-oss/api@latest search --json '{"query":"AI"}' \
  --api-key "different-key" \
  --client "DifferentAgent"

Implementation Checklist

• Runtime check: Node.js 18+ or Bun 1.0+
• If missing: curl -fsSL https://bun.sh/install | bash
• API key from https://you.com/platform/api-keys
• Environment variables set: YDC_API_KEY, YDC_CLIENT
• Schema discovery tested: bunx @youdotcom-oss/api@latest search --schema
• CLI tested with --json and --client flags
• Livecrawl tested (search + content in one call)
• Deep-search tested with low/medium/high effort levels
• Cross-platform timeout handled (Linux timeout / macOS gtimeout)
• Error handling added (exit codes + stderr)
• Output parsing implemented (jq without .data wrapper)
• Script integrated into workflow

Common Issues

“Cannot find module @youdotcom-oss/api” Fix: Use bunx (no install needed): bunx @youdotcom-oss/api or npx @youdotcom-oss/api

“–json flag is required” Fix: Pass query as JSON: --json '{"query":"..."}'

“YDC_API_KEY environment variable is required” Fix: export YDC_API_KEY="your-key"

“Tool execution fails with 401” Fix: Verify API key, get new key from platform

“Cannot parse jq: .data.results not found” Fix: Remove .data wrapper – use .results directly

Advanced Patterns

Schema-Driven Agent

#!/usr/bin/env bash
set -e

# Discover available search parameters (using ydc if installed globally)
schema=$(ydc search --schema)
echo "$schema" | jq '.properties | keys'

# Build query dynamically
query=$(jq -n '{
  query: "AI developments",
  count: 10,
  livecrawl: "web",
  livecrawl_formats: "markdown"
}')

# Execute search (using bunx)
bunx @youdotcom-oss/api@latest search --json "$query" --client ClaudeCode

Parallel Execution

#!/usr/bin/env bash
bunx @youdotcom-oss/api@latest search --json '{"query":"AI"}' --client ClaudeCode &
bunx @youdotcom-oss/api@latest search --json '{"query":"ML"}' --client ClaudeCode &
bunx @youdotcom-oss/api@latest search --json '{"query":"LLM"}' --client ClaudeCode &
wait

Rate Limit Retry

#!/usr/bin/env bash
for i in {1..3}; do
  if bunx @youdotcom-oss/api@latest search --json '{"query":"AI"}' --client ClaudeCode; then
    exit 0
  fi
  [ $i -lt 3 ] && sleep 5
done
echo "Failed after 3 attempts"
exit 1

Progressive Deep-Search

Start low, escalate only if needed:

#!/usr/bin/env bash
# Try low → medium → high until sufficient
for effort in low medium high; do
  result=$(bunx @youdotcom-oss/api@latest deep-search --json "{
    \"query\":\"$1\",
    \"search_effort\":\"$effort\"
  }" --client ClaudeCode)

  citations=$(echo "$result" | jq '.results | length')
  [ "$citations" -ge 5 ] && echo "$result" | jq -r '.answer' && exit 0
done

Parallel Deep-Search

Run multiple research questions concurrently:

#!/usr/bin/env bash
# Parallel research (3×60s = ~60s total, not 180s)
bunx @youdotcom-oss/api@latest deep-search --json '{"query":"Q1"}' --client ClaudeCode > q1.json &
bunx @youdotcom-oss/api@latest deep-search --json '{"query":"Q2"}' --client ClaudeCode > q2.json &
bunx @youdotcom-oss/api@latest deep-search --json '{"query":"Q3"}' --client ClaudeCode > q3.json &
wait

# Combine results
for f in q*.json; do jq -r '.answer' "$f"; done

Verify: Progressive saves quota by avoiding unnecessary high effort

Resources

• Package: https://github.com/youdotcom-oss/dx-toolkit/tree/main/packages/api
• API Keys: https://you.com/platform/api-keys