Decision page
Legal research with citations stack for Legal
What should a legal team use for internal research that exposes its source passages?
ZBS editorial starting point. Start with Docling for ingestion, pgvector for filtered retrieval, Claude citations for answers and LangGraph for review state.
Researchers receive an answer trail tied to the approved corpus instead of a freestanding response.
Editorial starting point
Retrieval with explicit citations
A concrete starting configuration that keeps source facts, model output and operational authority separate.
Choose this when: The source corpus is controlled and material statements must be traceable.
-
Convert the approved corpus. The public quickstart documents local conversion and structured export.
Limit: Scans, unusual tables, annotations and source metadata require a representative parsing test.
Evidence: source 1
-
Filter sources by authority metadata. It keeps vector search beside relational metadata and access filters in PostgreSQL.
Limit: Approximate retrieval can miss results and needs recall measurement against an exact baseline.
Evidence: source 1
-
Answer with cited passages. Native citations expose which supplied source passages support an answer.
Limit: A correct source link does not prove the interpretation or decision is correct.
Evidence: source 1
-
Record lawyer review. Stateful workflow steps keep model output separate from acceptance and operational action.
Limit: Policy, persistence, access control and recovery remain application responsibilities.
Evidence: source 1
Private / local
Controlled reasoning path
Keep parsing, retrieval or model inference in controlled infrastructure while retaining the same source-of-truth and approval rules.
Choose this when: Sensitive inputs cannot be sent to an external model API and the team can operate the additional infrastructure.
-
Ingestion
Docling
source backed inference
Parse sources locally. The public quickstart documents local conversion and structured export.
Limit: Scans, unusual tables, annotations and source metadata require a representative parsing test.
Evidence: source 1
-
Retrieval
pgvector
source backed inference
Store and filter source chunks. It keeps vector search beside relational metadata and access filters in PostgreSQL.
Limit: Approximate retrieval can miss results and needs recall measurement against an exact baseline.
Evidence: source 1
-
Answer
vLLM
source backed inference
Serve a local answer model. It provides a documented self-operated model-serving layer.
Limit: Serving a model does not prove its task accuracy, safe tool use or secure operation.
Evidence: source 1
-
Require lawyer acceptance. Stateful workflow steps keep model output separate from acceptance and operational action.
Limit: Policy, persistence, access control and recovery remain application responsibilities.
Evidence: source 1
Budget alternative
Lower-cost external model path
Keep the workflow and source integration explicit while evaluating a lower-cost model candidate on the same acceptance set.
Choose this when: External processing is acceptable and measured model spend is a leading constraint.
-
Ingestion
Docling
source backed inference
Keep source preparation fixed. The public quickstart documents local conversion and structured export.
Limit: Scans, unusual tables, annotations and source metadata require a representative parsing test.
Evidence: source 1
-
Retrieval
pgvector
source backed inference
Return the same source passages. It keeps vector search beside relational metadata and access filters in PostgreSQL.
Limit: Approximate retrieval can miss results and needs recall measurement against an exact baseline.
Evidence: source 1
-
Produce a candidate grounded answer. It is a concrete lower-cost external model candidate for the same acceptance set.
Limit: Price alone is not task fitness; output structure, languages, availability and data terms need testing.
Evidence: source 1
-
Capture review and rejection. Stateful workflow steps keep model output separate from acceptance and operational action.
Limit: Policy, persistence, access control and recovery remain application responsibilities.
Evidence: source 1
Trade-offs that change the choice
Implementation path
1. Freeze a dated corpus with authority metadata.
2. Create questions with relevant and misleading sources.
3. Score retrieval separately from answer correctness.
4. Require lawyer verification before client use.
Known limits
The corpus is not assumed complete or current.
Citation does not prove legal correctness or weight.
No product on this page is a universal winner; the configuration still needs a task-specific acceptance test.
EU and US routes stay consolidated with Global until evidence changes the answer.
Validate this stack on your data
A recommendation is a starting point. Practice Lab can test the same workflow on representative inputs, constraints and failure cases.
Request a real-data evaluation
