Show HN: Anjadhe – privacy first AI assistant, no account, no server DB
Anjadhe is a macOS personal AI assistant that goes beyond chat by integrating with email, files, and scheduling to automate real tasks It uses local LLMs via llama.cpp by default, with options for self-hosted servers, OpenAI/Anthropic keys, or Anjadhe Cloud Key features include automatic email triage (bills, receipts, bookings), conversational goal-setting with plan management, scheduled routines with logging, and writing style learning All user data is stored locally in SQLite with no account r
Analysis
TL;DR
- Anjadhe is a macOS personal AI assistant that goes beyond chat by integrating with email, files, and scheduling to automate real tasks
- It uses local LLMs via llama.cpp by default, with options for self-hosted servers, OpenAI/Anthropic keys, or Anjadhe Cloud
- Key features include automatic email triage (bills, receipts, bookings), conversational goal-setting with plan management, scheduled routines with logging, and writing style learning
- All user data is stored locally in SQLite with no account required and no telemetry by default
- The project is open-source (app + cloud service), currently in early alpha and macOS-only, requiring 32GB RAM for local model usage
Why It Matters
Anjadhe represents a growing trend toward locally-run, privacy-first AI assistants that act as genuine productivity agents rather than conversational interfaces. Its approach of turning conversational interactions into structured, actionable plans with user confirmation aligns with the industry shift from chatbots to autonomous agents. The open-source, no-account-required model also addresses growing user concerns about data privacy in AI tools.
Technical Details
- Built with Electron and vanilla JavaScript (no framework), running on macOS
- Supports multiple inference backends: local llama.cpp, self-hosted servers, OpenAI/Anthropic APIs, and Anjadhe Cloud (open-weight models)
- Data persistence via local SQLite database; no cloud telemetry unless explicitly enabled
- Email parsing pipeline extracts structured data (due dates, bookings, receipts) and converts them into tasks or plans
- Writing style adaptation uses a readable, editable representation rather than opaque fine-tuning or embeddings
- Minimum hardware requirement of 32GB RAM for local model operation; email features work on smaller models
Industry Insight
- The "local-first AI" approach is becoming a competitive differentiator as privacy-conscious users and enterprises seek alternatives to cloud-dependent AI services; open-sourcing both the app and cloud backend builds trust in a landscape rife with data misuse concerns
- Anjadhe's design of showing planned changes and requesting confirmation before execution reflects a best-practice pattern for agent-based tools—balancing automation with human oversight to prevent costly autonomous errors
- The distinction between features that work on small models (email parsing) versus those needing larger models (full agent) highlights a practical architecture: offloading lightweight tasks locally while reserving heavy reasoning for bigger models or servers, a pattern others should adopt
Disclaimer: The above content is generated by AI and is for reference only.