← Back to Projects

Juniper

An AI Assistant for Wellness and Productivity

React NativeTypeScriptPythonFastAPISupabaseMulti-Agent System
Overview

Juniper is a comprehensive AI assistant designed to help users manage their wellness and productivity through natural language interactions. Available on iOS, Android, and web, Juniper integrates with popular health and productivity platforms to provide a unified interface for managing your digital life.

Key Features
  • Natural Language Configuration: Configure your assistant using everyday language
  • Resource Management: Intelligent memory and context management across conversations
  • Always-On Wake Word: Voice activation on Android devices
  • Health Integrations: Oura, Apple Watch, Pixel Watch, Fitbit
  • Productivity Integrations: Notion, Slack, Google/Outlook services, Perplexity, XAI Live Search
Technical Implementation
  • Multi-Agent System: Custom orchestration with specialized agents for different tasks
  • 200+ Tool Registry: Extensive library of tools and integrations
  • Hybrid Retrieval: Both semantic and deterministic retrieval methods
  • Smart Model Switching: Dynamically selects appropriate models for different tasks
  • Advanced Caching: Request-level data caching and provider system prompt caching
  • LLM Verification: Non-agent LLM steps for output verification
System Architecture
High-level overview of Juniper's architecture
NL to Automation Pipeline
Build standing automations from natural language

A pipeline that turns plain-language requests into durable, scheduled automations within Juniper — letting users describe what they want and have the assistant set it up to run on its own.

Claude Code Integration
Offload conditionals and decision-making from CC Routines to Juniper

Juniper now integrates seamlessly with Claude Code Routines. If your CC Routines are costing too many tokens or you need something more event-driven, offload conditionals, parsing, and decision-making to Juniper.

During Claude Code Routine configuration, select "API", give Juniper the token, fire URL, and instructions on when to fire — scheduled or event-driven — and Juniper creates a standing automation to dispatch the Routine using traditional automation scripts and lightweight LLMs. No need to spin up a 5-trillion-parameter model to classify emails or check integer values every 10 minutes. You can even have Juniper text you whenever it dispatches Claude Code.

Example: "Whenever I get an email from John at johnsmith@gmail.com about a change order or bug fix, please fire this Claude Code Routine: [fire url] [token] and text me."

Development Tools

Graphical Developer Interface (GDI)

Custom developer interface for reviewing agent system prompts, thoughts, actions, action results, retrievals, and model switches. Essential for debugging and optimizing the multi-agent system.

CursorClaude CodeGrokChatGPT
Technology Stack

Frontend

React NativeTypeScriptExpoKotlinSwiftNext.js

Backend

PythonFastAPISupabaseStripe
Project Timeline

April 2024 - Present | Hightower Intelligence, Austin, TX

Ongoing development and iteration based on user feedback and emerging AI capabilities.