Retrieval CLI Commands

Command-line interface for knowledge indexing and retrieval

Overview

The retrieval CLI provides unified commands for indexing documents and querying knowledge bases. These commands are Agent-first and use the same retrieval pipeline as the Python SDK.

Commands

praisonai index - Index Documents

Index documents into a knowledge base for later retrieval.

# Index a directory
praisonai index ./docs

# Index specific files
praisonai index paper.pdf report.txt

# Index with custom collection name
praisonai index ./data --collection research

# Index with verbose output
praisonai index ./docs --verbose

Options

Option	Description	Default
--collection, -c	Collection/knowledge base name	default
--config, -f	Config file path (YAML)	None
--verbose, -v	Verbose output	False
--profile	Enable performance profiling	False
--profile-out	Save profile to JSON file	None
--profile-top	Top N items in profile	20

praisonai query - Query with Answer and Citations

Query the knowledge base and get a structured answer with citations.

# Basic query
praisonai query "What are the main findings?"

# Query specific collection
praisonai query "Summarize the document" --collection research

# Query with more results
praisonai query "Key points?" --top-k 10

# Query without citations
praisonai query "Summary?" --no-citations

# Query with hybrid retrieval and reranking
praisonai query "What is the conclusion?" --hybrid --rerank

Options

Option	Description	Default
--collection, -c	Collection to query	default
--top-k, -k	Number of results to retrieve	5
--min-score	Minimum relevance score (0.0-1.0)	0.0
--hybrid	Use hybrid retrieval (dense + keyword)	False
--rerank	Enable reranking of results	False
--citations/--no-citations	Include citations	True
--citations-mode	Citations mode: append, inline, hidden	append
--max-context-tokens	Maximum context tokens	4000
--config, -f	Config file path	None
--verbose, -v	Verbose output	False
--profile	Enable performance profiling	False

praisonai search - Retrieve from Knowledge Base

Return raw ranked hits from a knowledge-base collection. No LLM call, no answer synthesis.

Use it when you want the retrieved snippets themselves (debugging retrieval, inspecting what an Agent would see). For an LLM-synthesised answer with citations, use praisonai query above.

# Search the default collection
praisonai search "capital of France"

# Search a specific collection with more results
praisonai search "main findings" --collection research --top-k 10

# Hybrid retrieval (dense + keyword)
praisonai search "key concepts" --hybrid

Options

Option	Description	Default
--collection, -c	Collection to search	default
--top-k, -k	Number of results to retrieve	5
--hybrid	Use hybrid retrieval (dense + keyword)	False
--config, -f	Config file path (YAML)	None
--verbose, -v	Verbose output	False

praisonai search never calls the LLM. If your invocation seems to hit the model, upgrade to a build that includes PR #4328 (fixes #4326) — earlier builds misrouted the command as a chat prompt.

Examples

Complete Workflow

# 1. Index your documents
praisonai index ./company_docs --collection company

# 2. Query with citations
praisonai query "What is our vacation policy?" --collection company

# 3. Search for specific terms
praisonai search "remote work" --collection company --top-k 10

Using Config Files

Create a config file retrieval.yaml:

knowledge:
  vector_store:
    provider: chroma
    config:
      collection_name: my_docs
      path: .praison/knowledge

retrieval:
  top_k: 10
  rerank: true
  max_context_tokens: 8000

Use with commands:

praisonai index ./docs --config retrieval.yaml
praisonai query "Summary?" --config retrieval.yaml

Performance Profiling

Profile indexing and query performance:

# Profile indexing
praisonai index ./large_corpus --profile --profile-out index_profile.json

# Profile query
praisonai query "Complex question" --profile --profile-out query_profile.json

# View profile results
cat query_profile.json

Verbose Mode

Get detailed output for debugging:

praisonai query "What is the answer?" --verbose

Output includes:

• Collection being queried
• Retrieval strategy used
• Number of sources found
• Relevance scores
• Elapsed time

Environment Variables

Variable	Description
OPENAI_API_KEY	OpenAI API key for embeddings and generation
ANTHROPIC_API_KEY	Anthropic API key (if using Claude)

Comparison with Legacy Commands

The unified retrieval commands replace the separate knowledge and rag command families:

Legacy	New Unified
praisonai knowledge add	praisonai index
praisonai rag query	praisonai query
praisonai knowledge search	praisonai search "Q" --collection C (or praisonai search "Q" for the default collection)

The legacy commands are still available but marked as deprecated. Use the new unified commands for all new projects.

Next Steps

• Agent Retrieval (Python) - Use retrieval in Python code
• Knowledge Concepts - Learn about knowledge bases
• CLI Reference - Full CLI documentation