Skip to Content
TemplatesAgentic

agentic template

Multi-query decomposition over hybrid retrieval — for applications that must answer complex, multi-hop questions.

What you get

project/
├── src/
│   ├── pipeline.py       # Agentic query loop with multi-query decomposition
│   └── config.py         # Pipeline configuration
├── eval/
│   ├── golden_set.json   # Evaluation dataset
│   └── config.yaml       # Evaluation thresholds and CI gate settings
├── pyproject.toml        # Python dependencies and RAG-Forge config
└── README.md

Default configuration

The agentic template enables agent_mode, which activates multi-query decomposition: a complex question is broken into sub-queries, each retrieved independently, and the results are synthesised into a single answer. Retrieval is hybrid (BM25 + vector) with the same alpha = 0.6 default as the hybrid template.

# pyproject.toml [tool.rag-forge] section
[tool.rag-forge]
template = "agentic"
chunk_strategy = "recursive"
chunk_size = 512
overlap_ratio = 0.1
vector_db = "qdrant"
embedding_model = "BAAI/bge-m3"
retrieval_strategy = "hybrid"
retrieval_alpha = 0.6
agent_mode = true

Invoke agentic mode at query time with the --agent-mode flag:

rag-forge query "Compare the refund policies across all product lines" --agent-mode
  1. Index your documents the same way as hybrid: rag-forge index --source ./docs --sparse-index-path .rag-forge/sparse.
  2. Build a golden set that includes multi-hop questions (questions requiring facts from two or more separate documents) so your evaluation reflects the real workload.
  3. Run rag-forge audit --golden-set eval/golden_set.json and review per-question faithfulness scores to identify decomposition failures.

When to upgrade

Move to the enterprise template when you need tenant isolation, PII scrubbing on inputs and outputs, audit logging, or CI/CD quality gates in a regulated environment.

HybridEnterprise