← all repositories
matthiasn/lotti

Your logbook, now with opinionated roommates

A local-first journal and task manager that runs thousands of persistent AI agents on your own hardware, each with a memory, a personality, and a grievance system.

1.1k stars Dart AgentsLLMOps · Eval
lotti
Velocity · 7d
+0.3
★ / day
Trend
steady
star history

What it does

Lotti is a cross-platform logbook (tasks, time tracking, voice notes, journal entries, habits, health data) with a twist: persistent AI agents that read your entries, remember past reports, and propose next steps. Everything stores locally by default, syncs between your devices over Matrix with end-to-end encryption (Vodozemac), and can run entirely offline if your hardware cooperates.

The interesting bit

The agents aren’t stateless chatbots. They’re long-living entities with “missions,” reusable “souls” (personalities), and a formal grievance system — you can literally complain about an agent’s behavior, and it gets addressed in your next scheduled 1-on-1. The author runs over a thousand task agents on a phone by having them subscribe to fine-grained entity changes rather than polling; they sleep cheap and wake only when relevant data moves.

Key highlights

  • Local model support validated with Qwen 3.6 35B A3B (~3B active parameters), though the author runs hybrid: local for sensitive categories, Gemini Flash for routine stuff
  • Per-category provider selection — different trust boundaries for journaling vs. open-source work
  • Voice transcription via local Whisper or Voxtral
  • Strict two-database architecture: agents read from your facts, but can’t write to them without explicit human approval
  • Cross-platform Flutter app; Windows currently build-from-source only

Caveats

  • Local agent inference is power-hungry: tested on an M4 Max with 128GB RAM, produces audible fan noise and noticeable battery drain under sustained load
  • Image generation for cover art is cloud-only for now; local Python-based service is an open contribution wanted
  • iOS and Android Play Store builds are invitation-only TestFlight/internal testing; general mobile users need APK or sideload

Verdict

Worth a look if you’re privacy-obsessed, curious about agentic workflows, and have the hardware to run local models — or the patience to configure hybrid cloud fallbacks. Skip if you want a polished, zero-config consumer app; this is explicitly a system you tune and manage like a relationship.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.