I use multiple specialized agents (true).
The Main Planner is the entry point for user requests (true).
The executor is deterministic and does not require an LLM call for every individual step (true).
Validator checks the result of workflow execution (checks workflow execution results).
LLMs decide what should happen, while deterministic components execute what was decided (LLMs decide, deterministic components execute).
Workflow steps can reference results generated by previous steps (steps can reference prior results).
NuroFlow uses LangGraph to orchestrate the different agents and execution stages (true).
I built NuroFlow, an AI‑powered workflow assistant that lets me interact with external services like Notion and Google Calendar using natural language. The system follows a Planner → Executor → Validator → Response pipeline orchestrated with LangGraph, and it supports JWT‑based authentication for internal users and OAuth for external integrations.
- Commits: 66 total as of Sep 27 2026.
- Architecture: Planner → Executor → Validator pattern orchestrated with LangGraph.
- Agents: Main Planner, Domain Planners (Calendar, Notion), Domain Executors, Validator, Response Agent.
- Workflow generation: deterministic JSON workflows with step‑to‑step dependencies.
- Clarification loop: planner asks for missing data before generation.
- Validation & retry: failures trigger replanning.
- Authentication: JWT for internal users, OAuth for external services.
- Supported domains: Notion, Google Calendar (easily extensible).
- Latest commit hash: 60419f93d08826775da4b652910f47f329720638.
- Commit date: Sep 27, 2026.