Enterprise-grade platform for designing, generating, and executing voice-enabled AI agents through visual workflows, LangGraph orchestration, and real-time voice communication.
Quick Start | Installation | Feature Guides | Documentation
GraphForge is a visual voice agent development platform that transforms conversation design from code-first to design-first. Draw your agent's conversation flow, configure voice integrations visually, and generate production-ready LangGraph code—all without writing boilerplate.
Design workflows. Generate code. Deploy voice agents.
GraphForge bridges the gap between conversation designers and AI engineers. We eliminate the friction of building voice agents by providing a visual interface that exports to production-ready code, enabling teams to iterate on conversation flows without rewriting infrastructure.
GraphForge is for teams who need to:
- Conversation Designers — Design complex voice flows visually without writing LangGraph code
- Voice AI Engineers — Accelerate development with code generation from visual specifications
- Product Teams — Prototype and iterate on voice agent experiences rapidly
- Enterprises — Standardize voice agent development across teams with consistent patterns
- Visual flow editor for conversation workflows (12 node types)
- Voice agent configuration (Deepgram STT, ElevenLabs/OpenAI TTS, Silero VAD)
- Code generation from YAML to Python/TypeScript with LangGraph patterns
- YAML import/export for portable workflow specifications
- Auto-layout algorithms (Dagre, ELK) for graph visualization
- Pre-built templates for common voice agent patterns
- Voice runtime with LiveKit integration and human-in-the-loop interrupts
- Visual workflow execution (agents run on generated code, not in the editor)
- Multi-tenant hosted service (self-hosted only)
- Non-voice agent patterns (this is specialized for voice + LangGraph)
Drag-and-drop interface for designing voice agent conversations with 12 node types and real-time visual feedback.
Flow editor with Node Palette showing all 12 node types: Trigger (Start, Voice Input, Interactive, End), Processing (Intent Recognition, Entity Extraction), Response (Text-to-Speech, Action, Message), and Control (Decision, Merge) nodes.
| Node Category | Nodes | Purpose |
|---|---|---|
| Trigger | Start, Voice Input, Interactive, End | Entry/exit points |
| Processing | Intent Recognition, Entity Extraction | Analyze user input |
| Response | Text-to-Speech, Action, Message | Generate responses |
| Control | Decision, Merge | Branching logic |
| Config | Agent Configuration | Global settings |
Flow Editor Guide | Architecture
Configure multi-provider voice services through an intuitive tabbed interface.
Agent Configuration modal with OpenAI tab showing model selection (GPT-5), temperature control, and max tokens configuration.
| Provider | Service | Configuration |
|---|---|---|
| OpenAI | LLM + TTS | Model, temperature, max tokens |
| Deepgram | STT | Model (Flux), language selection |
| ElevenLabs | TTS | Voice, stability, streaming latency |
| Silero | VAD | Silence duration, speech padding |
Generate production-ready Python/TypeScript from visual workflows with customizable Jinja2 templates.
Generated Python code showing async node functions with TypedLivekit integration for real-time voice streaming.
# Generate from YAML specification
graphforge-gen --input workflow.yml
# Output: WorkflowName/graph.py, implementation.py, pipeline.pyAutomatically arrange nodes using professional graph layout engines.
| Engine | Best For | Key Options |
|---|---|---|
| Dagre | Simple hierarchical flows | Rank direction, node/edge separation |
| ELK | Complex multi-branch flows | Algorithm, node placement, spacing |
Start quickly with pre-built templates for common voice agent patterns:
- Medical Office Triage — Multi-agent routing with support/billing branches
- Personal Shopping Assistant — Customer service with purchase/returns flows
- Video Content Moderation — AI-powered content analysis pipeline
Portable workflow specifications for version control and team collaboration.
YAML Export modal with Mermaid flowchart visualization and LLM-powered regeneration.
# 1. Clone the repository
git clone https://github.com/vericle/graphforge.git
cd graphforge
# 2. Start the frontendcd frontend
pnpm install
cp .env.example .env
pnpm dev
# 3. Open http://localhost:3000Detailed Installation Guide | Frontend Setup
- Node.js 18+ and pnpm 8+ (Frontend)
- Python 3.12+ and uv (Backend & Runtime)
- LiveKit server (Runtime)
- API keys: OpenAI, Deepgram, and optionally ElevenLabs
# Clone the repository
git clone https://github.com/vericle/graphforge.git
cd graphforge
# Frontend setupcd frontend
pnpm install
cp .env.example .env
# Edit .env with your API keys# Backend setup (code generation)cd ../backend
uv venv &&source .venv/bin/activate
uv pip install -e .# Runtime setup (voice execution)cd ../runtime
uv venv &&source .venv/bin/activate
uv pip install -e .
cp .env.example .env
# Edit .env with LiveKit and AI service credentials| Service | URL | Description |
|---|---|---|
| Visual Editor | http://localhost:3000 | Flow designer UI |
| Code Gen API | http://localhost:8000 | Generation endpoints |
| LangGraph Server | http://localhost:2024 | Agent execution |
| Guide | Description |
|---|---|
| Frontend README | Visual editor setup and features |
| Backend README | Code generation CLI and API |
| Runtime README | Voice agent execution |
| Guide | Description |
|---|---|
| Flow Editor Architecture | System design and components |
| UI Style Guide | Component patterns |
| Technical Reference | Runtime architecture |
| Modular Agent Pattern | Agent structuring |
| Document | Description |
|---|---|
| Frontend Contributing | Development guidelines |
| Governance | Project governance model |
| Roadmap | Project direction |
| Layer | Technologies |
|---|---|
| Frontend | Next.js 15, React 19, TypeScript, @xyflow/react, TailwindCSS, Radix UI |
| Backend | Python 3.12, Jinja2, FastAPI, PyYAML |
| Runtime | LangGraph, LiveKit Agents, Deepgram, OpenAI, ElevenLabs, Pydantic |
| Layout | Dagre, ELK.js |
- GitHub Issues — Bug reports and feature requests
- GitHub Discussions — Questions and community discussion
| Document | Description |
|---|---|
| Contributing Guide | How to contribute |
| Code of Conduct | Community standards |
| Governance | Decision-making process |
| Maintainers | Project maintainers |
| Roadmap | Project direction |
MIT — see LICENSE for details.
GraphForge builds upon the work of several open-source projects:
- LangGraph Builder — Visual graph editor patterns and inspiration
- langgraph-gen-py — Code generation approach
- langgraph-livekit-agents — LangGraph + LiveKit integration patterns
- React Flow — Graph visualization library
- LangGraph — AI orchestration framework
- LiveKit — Real-time voice communication
GraphForge — Design voice agents visually. Generate code automatically.
LangGraph · LiveKit · Next.js · React Flow