🎥 Watch me develop this live! Follow along as I build Chatfield in real-time: YouTube Development Streams
Transform data collection from rigid forms into natural conversations powered by LLMs.
Chatfield provides implementations in both Python and TypeScript/JavaScript, allowing you to create conversational interfaces for gathering structured data in any application.
- Natural Conversations: Replace traditional forms with engaging dialogues
- LLM-Powered Validation: Smart validation and guidance through conversation
- Type Safety: Full type support in both Python and TypeScript
- Rich Transformations: Convert responses into any data type with casts
- Framework Integration: React components, CopilotKit support, and more
- Flexible APIs: Multiple API styles to suit different preferences
from chatfield import chatfield, Interviewer
# Build your interview using the fluent builder API
bug_report = (chatfield()
.type("Bug Report")
.desc("Collecting bug report details")
.field("steps")
.desc("Steps to reproduce")
.field("expected")
.desc("What should happen")
.field("actual")
.desc("What actually happens")
.build())
# Conduct the interview
interviewer = Interviewer(bug_report)
user_input = None
while not bug_report._done:
message = interviewer.go(user_input)
print(message)
user_input = input("Your input > ")
# Access collected data
print("Steps:", bug_report.steps)
print("Expected:", bug_report.expected)
print("Actual:", bug_report.actual)import { chatfield } from '@chatfield/core'
const BugReport = chatfield()
.field('steps', 'Steps to reproduce')
.must('be detailed')
.field('expected', 'What should happen')
.field('actual', 'What actually happens')
.build()
const result = await BugReport.gather()
console.log(result.steps, result.expected, result.actual)Transform conversational responses into any data type you need:
from chatfield import chatfield
project_estimate = (chatfield()
.type("Project Estimate")
.desc("Project estimation")
.field("team_size")
.desc("Number of developers")
.as_int() # Parse to integer
.must("between 1 and 50")
.field("confidence")
.desc("Confidence in timeline")
.as_percent() # Parse to 0.0-1.0
.field("technologies")
.desc("Tech stack you'll use")
.as_list() # Parse to list
.field("needs_design")
.desc("Need design resources?")
.as_bool() # Parse to boolean
.build())
# After collection:
project_estimate.team_size.as_int # 5 (int)
project_estimate.confidence.as_percent # 0.75 (float)
project_estimate.technologies.as_list # ["Python", "React", "PostgreSQL"]
project_estimate.needs_design.as_bool # True (bool)Casts can have multiple transformations with custom parameters:
feature_request = (chatfield()
.type("Feature Request")
.field("affected_users")
.desc("Number of users impacted")
.as_int() # Base integer cast
.as_lang('fr') # Translate to French
.as_lang('es') # Translate to Spanish
.as_bool('even', 'True if even number') # Custom boolean
.as_bool('critical', 'True if > 100 users') # Another boolean
.as_str('uppercase', 'In all caps') # String format
.as_set('factors', 'Prime factors') # Set operation
.build())
# After collection, if user said "two hundred":
feature_request.affected_users # "200" (base string)
feature_request.affected_users.as_int # 200
feature_request.affected_users.as_lang_fr # "deux cents"
feature_request.affected_users.as_lang_es # "doscientos"
feature_request.affected_users.as_bool_even # True
feature_request.affected_users.as_bool_critical # True
feature_request.affected_users.as_str_uppercase # "TWO HUNDRED"
feature_request.affected_users.as_set_factors # {2, 5}Control how many options can be selected:
task_assignment = (chatfield()
.type("Task Assignment")
.field("task_type")
.desc("Type of task")
.as_one('selection', 'bug', 'feature', 'enhancement')
.field("priority")
.desc("Priority level (if known)")
.as_maybe('level', 'low', 'medium', 'high', 'critical')
.field("components")
.desc("Affected components")
.as_multi('systems', 'frontend', 'backend', 'database')
.field("reviewers")
.desc("Optional reviewers")
.as_any('people', 'Alice', 'Bob', 'Charlie', 'Diana')
.build())
# After collection:
task_assignment.task_type.as_one_selection # "bug"
task_assignment.priority.as_maybe_level # "high" or None
task_assignment.components.as_multi_systems # {"frontend", "backend"}
task_assignment.reviewers.as_any_people # {"Alice", "Diana"} or set()This repository contains two parallel implementations:
- Decorator-based API with LangGraph orchestration
- OpenAI integration with extensible LLM support
- Rich validation and transformation system
- Full async support and type hints
- NPM package
@chatfield/core - React hooks and components
- CopilotKit integration
- Multiple API styles (builder, decorator, schema)
cd chatfield-py
pip install .cd chatfield-js
npm install- Python Documentation - Full API reference and advanced features
- TypeScript/JavaScript Documentation - React integration and builder patterns
- Python Examples - Working examples with all features
- TypeScript Examples - Framework integration examples
Both implementations share the same core concepts but are tailored to their respective ecosystems:
- Python: Uses decorators, LangGraph for orchestration, and async/await patterns
- TypeScript: Provides React integration, builder patterns, and full type safety
We welcome contributions to either implementation! Please see the individual project directories for specific development setup instructions.
# Python tests
cd chatfield-py && python -m pytest
# JavaScript tests
cd chatfield-js && npm testApache License 2.0 - See LICENSE for details.
Both implementations require an OpenAI API key:
export OPENAI_API_KEY=your-api-key- Python Details: See chatfield-py/CLAUDE.md for implementation details
- TypeScript Details: See chatfield-js/CLAUDE.md for implementation details
Both implementations are in active development with feature parity as a goal. The Python implementation currently has more complete decorator support, while the TypeScript implementation offers better framework integrations.