BetaHow MasterNode runs work
Agent Factory explains the parallel pipeline from one request to one result. Features below match what ships in Public beta (August 2026).
01 — agent factory
From one request to parallel execution
How tasks flow through MasterNode.ai: from a single request to parallel execution and back to one coherent result. This is the same pipeline used in Tasks and optional pipeline mode in Chat.
Agent Factory Flow
Hover over components to see details. Rotate, zoom, and pan to explore.
High-level flow: Master Agent → Decomposer → Parallel Agents → Aggregator → Supervisor
02 — pipeline
Pipeline stages
Master Agent
Receives the task, analyzes scope and complexity, and decides how to decompose it for parallel execution.
Decomposer
Splits the work into independent subtasks that can run concurrently. Each subtask is assigned to a parallel agent.
Parallel Agents
Multiple LLM agents run subtasks at the same time. You can use different models (OpenAI, Gemini, Claude, etc.) via Settings.
Aggregator
Collects results from all parallel agents and merges them into a single, coherent response.
Supervisor
Validates the aggregated output for quality and consistency, then delivers the final result to your app.
03 — rationale
Why parallel?
Complex tasks are broken into smaller pieces and run at the same time. That means faster results and better use of your LLM quota. You get one coherent answer without managing threads or queues yourself.
Deliverable routing sends document and presentation requests down fast paths; code and research tasks use the full swarm. Configure task text, parallelism, RAG, and model providers via the API — see docs for request shapes and WebSocket events.
04 — features
What you can use today
Updated August 2026. Every item below is available in the product.
Creator chat workspace
Streaming chat, attachments, web search, pipeline toggle, ghost mode, response versions, and in-chat pricing.
Parallel task pipeline
Full LangGraph run with plan questions, WebSocket DAG, human-in-the-loop on /tasks/execute, and deliverable routing.
Assistants & intake forms
Gallery of saved templates, detail dialogs, variable tags, and pre-send intake cards in chat.
Knowledge base (RAG)
Upload → chunk → embed → retrieve on Neon pgvector. Memory page with working memory and #knowledge section.
Web search & session location
Intent-gated search with citations, locality-aware queries, and approximate location from IP for local recommendations.
Integrations (wave 1)
n8n, Canva OAuth, Google Docs OAuth, Notion token — connect, test, and invoke from the hub.
OAuth, onboarding & tour
Google OAuth, five-step onboarding, environment provisioning, and restartable chat intro tour from Settings.
LLM evaluation suite
Offline and live benchmark scripts, case library, and provider comparison docs — thesis-ready proof artifacts.
Multi-provider models
Route across OpenAI, Google Gemini, Anthropic Claude, Groq, DeepSeek, Mistral, and others. Bring your own API keys.
Billing & plans
Free and paid plans with Stripe checkout and Razorpay support. In-chat upgrade board and usage visibility.
Have a request? Share feedback, read the about page, or check the changelog for what shipped.
Documentation & endpoints
Guides, API reference, and route index in the repo