search-research

search-research

Unified search and research package with local code/knowledge search and web research capabilities

Overview

search-research provides a unified interface for searching across multiple knowledge sources, combining fast local code/knowledge search with comprehensive web research.

Key Features

• Fast Local Search (<1s): Code patterns, documentation, chat history, knowledge base
• Comprehensive Web Search (5-10s): Web providers with HyDE query enhancement
• Intelligent Routing: Auto-detects query intent (local vs web vs mixed)
• Unified API: Single package for both /search and /research commands
• Graceful Degradation: Works without API keys, local always available

Installation

# Basic installation (core backends only)
pip install search-research

# Full installation (all features including web providers)
pip install search-research[all]

# Development installation
pip install -e "P://packages/.claude-marketplace/plugins/search-research[all]"

# Migrating from unified-search?
# See "Migrating from unified-search" section below

Quick Start

Fast Local Search

from core import SearchRouter

router = SearchRouter()
results = router.search("FastAPI patterns", limit=10)
# Returns: <1s, local backends only

Comprehensive Web Search

from core import ResearchRouter

router = ResearchRouter()
results = router.search("FastAPI best practices", limit=20)
# Returns: 5-10s, all backends with HyDE

Full Control

from core import UnifiedRouter, Mode

router = UnifiedRouter(mode=Mode.FAST)
results = router.search(
    "FastAPI patterns",
    limit=10,
    web=True,  # Include web providers
    hyde=False  # Disable HyDE
)

Migrating from unified-search

If you're currently using EnhancedUnifiedSearchRouter from __csf, here's how to migrate:

Installation

# Install the new package
pip install search-research[all]

# Or for development
pip install -e "P://packages/.claude-marketplace/plugins/search-research[all]"

Code Changes

Old code:

from search.unified_router import EnhancedUnifiedSearchRouter

router = EnhancedUnifiedSearchRouter(
    chs_backend=chs_backend,
    cks_backend=cks_backend,
    enable_cache=True,
)
results = router.search("query", limit=10)

New code:

from core import SearchRouter

router = SearchRouter(
    cache_ttl=3600,
    enable_cache=True,
)
results = router.search("query", limit=10)

Key Differences

• Simpler API: No manual backend initialization needed
• Concurrent execution: Parallel backend searches (faster)
• 8 local backends: cds, grep, skills, chs, cks, kg, rlm, persona
• Graceful degradation: Failed backends don't crash search
• Better performance: ThreadPoolExecutor with timeouts

Research Mode

For comprehensive web search, use ResearchRouter:

from core import ResearchRouter

router = ResearchRouter(
    hyde_enabled=True,
    web_providers=["tavily", "brave", "serper"],
)
results = await router.search_async("FastAPI best practices", limit=20)

CLI Commands

Old:

# Using __csf internal command
python -m __csf.src.cli.nip.search_enhanced "query"

New:

# Fast local search
search-research "query" --backend cds grep

# Comprehensive web research
research-research "query" --limit 30 --hyde

Deprecation Timeline

• 2026-03-06: unified-search deprecated
• 2026-06-01: Migration period ends
• 2026-09-01: End of life (Q3 2026)

For detailed migration guide, see MIGRATION.md

Search Modes

FAST Mode (Default for SearchRouter)

Characteristics:

• Backends: Local only (CDS, Grep, CHS, CKS, RLM, Persona, MultiLang, NotebookLM)
• Timeout: 0.5s per backend
• HyDE: Disabled
• Speed: <1s total
• Use case: Quick code search, finding patterns

Example:

router = SearchRouter()
results = router.search("FastAPI route decorator")

COMPREHENSIVE Mode (Default for ResearchRouter)

Characteristics:

• Backends: All (local + web + NotebookLM)
• Timeout: 5s per web backend, 0.5s per local backend
• HyDE: Enabled
• Speed: 5-10s total
• Use case: Learning new topics, best practices, research

Example:

router = ResearchRouter()
results = router.search("FastAPI best practices for authentication")

Backends

Local Backends