Legal Dictation

Specialized pipeline for legal document transcription and formatting.

Overview

The legal-dictation template is optimized for legal professionals, with citation detection, proper formatting, and compliance features.

Configuration

template: legal-dictation

stt:
  model: whisper-1
  language: en
  word_timestamps: true

llm:
  model: gpt-5.2
  system_prompt: |
    You are a legal document formatter.

    Tasks:
    - Format dictation as proper legal documents
    - Identify and format case citations (e.g., Smith v. Jones, 123 U.S. 456)
    - Identify and format statute citations
    - Apply proper legal paragraph numbering
    - Maintain formal legal language
    - Correct obvious dictation errors while preserving meaning
  max_tokens: 2000
  temperature: 0.3

formatting:
  style: legal
  citations: detect_and_format
  paragraphs: numbered
  output_format: docx

compliance:
  audit_log: true
  data_classification: confidential

Features

Citation Detection

Automatically identifies and formats:

• Case citations: Smith v. Jones, 123 U.S. 456 (1990)
• Statute citations: 42 U.S.C. § 1983
• Regulatory citations: 17 C.F.R. § 240.10b-5
• Court rules: Fed. R. Civ. P. 12(b)(6)

Legal Formatting

• Numbered paragraphs
• Proper indentation
• Block quotes for lengthy citations
• Signature blocks
• Caption formatting

Usage

Basic Transcription

from gateflow_mcp import MCPClient
import base64

client = MCPClient(agent_id="agent_legal", api_key="gf-agent-...")

with open("deposition_clip.mp3", "rb") as f:
    audio_b64 = base64.b64encode(f.read()).decode()

result = client.call_tool(
    name="voice/pipeline",
    arguments={
        "audio": audio_b64,
        "template": "legal-dictation"
    }
)

print("Formatted Document:")
print(result["formatted_document"])
print(f"\nCitations found: {result['citations']}")

Memo Dictation

result = client.call_tool(
    name="voice/pipeline",
    arguments={
        "audio": audio_b64,
        "template": "legal-dictation",
        "overrides": {
            "llm": {
                "system_prompt": """Format as a legal memorandum with:
- TO/FROM/DATE/RE header
- Issue section
- Brief Answer
- Facts
- Analysis
- Conclusion"""
            },
            "formatting": {
                "document_type": "memorandum"
            }
        }
    }
)

Contract Clause Dictation

result = client.call_tool(
    name="voice/pipeline",
    arguments={
        "audio": audio_b64,
        "template": "legal-dictation",
        "context": "Formatting contract language for a software licensing agreement.",
        "overrides": {
            "formatting": {
                "document_type": "contract",
                "numbering_style": "1.1.1"
            }
        }
    }
)

Response Format

{
  "transcription": "The defendant in Smith versus Jones argued that under 42 USC section 1983...",
  "formatted_document": "1. The defendant in *Smith v. Jones*, 123 U.S. 456 (1990), argued that under 42 U.S.C. § 1983...",
  "citations": [
    {
      "type": "case",
      "text": "Smith v. Jones, 123 U.S. 456 (1990)",
      "formatted": "*Smith v. Jones*, 123 U.S. 456 (1990)"
    },
    {
      "type": "statute",
      "text": "42 USC section 1983",
      "formatted": "42 U.S.C. § 1983"
    }
  ],
  "document_structure": {
    "paragraphs": 5,
    "citations_count": 3,
    "word_count": 450
  },
  "compliance": {
    "audit_id": "audit_xyz789",
    "classification": "confidential"
  },
  "usage": {
    "stt_seconds": 120,
    "llm_tokens": {"prompt": 500, "completion": 800}
  }
}

Document Types

Type	Format	Best For
memorandum	Legal memo	Research memos
brief	Court brief	Motion practice
contract	Agreement	Transactional
letter	Correspondence	Client letters
deposition	Q&A format	Depositions
pleading	Court filing	Complaints, answers

Deposition Format

result = client.call_tool(
    name="voice/pipeline",
    arguments={
        "audio": deposition_audio,
        "template": "legal-dictation",
        "overrides": {
            "formatting": {
                "document_type": "deposition",
                "speaker_labels": True,
                "line_numbers": True,
                "timestamp": True
            }
        }
    }
)

Output:

DEPOSITION OF JOHN DOE
February 16, 2026

Page 1
1    Q.  (By Mr. Smith) Please state your name for the record.
2    A.  John Doe.
3    Q.  And where do you currently reside?
4    A.  123 Main Street, Boston, Massachusetts.

Compliance Features

Audit Trail

Every transcription is logged:

{
  "audit_id": "audit_xyz789",
  "timestamp": "2026-02-16T10:00:00Z",
  "agent_id": "agent_legal",
  "template": "legal-dictation",
  "audio_duration_seconds": 120,
  "output_hash": "sha256:abc123...",
  "classification": "confidential"
}

Data Classification

result = client.call_tool(
    name="voice/pipeline",
    arguments={
        "audio": audio_b64,
        "template": "legal-dictation",
        "overrides": {
            "compliance": {
                "classification": "attorney_client_privilege",
                "matter_id": "CASE-2026-001",
                "client_id": "client_xyz"
            }
        }
    }
)

Best For

• Depositions - Transcription with speaker labels
• Legal memos - Research memoranda
• Contract drafting - Clause dictation
• Court filings - Pleadings and motions
• Client correspondence - Formal letters

Permissions Required

permissions:
  tools:
    - voice/pipeline
  models:
    - whisper-1
    - gpt-5.2
  pipelines:
    - legal-dictation
  data_classification:
    - confidential
    - attorney_client_privilege

Next Steps

• Ambient Scribe - Medical transcription
• Custom Templates - Build your own
• Compliance - Compliance features