LMIM
lmim xipe · the app
Encrypted peer-to-peer chat. A real terminal. A phone that mirrors the whole thing over your own LAN. An autonomous agent that builds multi-file projects end to end. No cloud. No accounts. No subscription.
"The first sun was raw power. The second is what you can reach with it."
./run --live
No slides, no mockups — every clip is the real system doing the thing on real hardware.
click a module to run its demo · esc collapses
on your phone too
The Android app finds your LMIM instance on your LAN automatically and mirrors the real dashboard — not a stripped-down clone. Same chat, same contacts, same conversations as desktop. Toggle visibility from your phone, and it changes everywhere.
and everything else
Build Crew and Chat get the spotlight. This is what's running underneath, every day.
Point it at any local folder and it becomes your coding partner — reads, writes, edits, sandboxed to that root.
Drop a PDF or Markdown file, ask focused questions. Hybrid retrieval, nothing leaves your machine.
Pull and analyze up to 10 URLs in parallel, respecting robots.txt. Basic extraction or AI-driven analysis.
A local address book wired into messaging. "Send a WhatsApp to John Doe" resolves the name for you.
Standing instructions injected into every session, so you stop repeating yourself.
Whisper STT and Piper TTS, five languages, auto-speak. Hold to talk, or just type.
AES-256-GCM, text, file, or whole directories. Local data protection with no server in the loop.
Bulk WhatsApp or email outreach from a CSV — for the messages that need to reach a list, not a person.
A real calendar, scheduled by talking to it in plain language.
release log
choosing a build
Not a base-plus-DLC split — every variant is fully functional. Pick whichever fits your bandwidth, and point Settings at any GGUF later regardless of which you start with.
Gemma 4 (2B) + MTP draft model
Speculative decoding gives a real speed boost on GPU. The full experience, best quality.
~6.8 GB
Qwen 3.5 (2B, Q4)
No draft-model overhead, still solid quality. The pragmatic pick on slower connections.
~2.9 GB
Gemma 4 (2B)
The reasoning model without speculative decoding — for hardware where MTP isn't worth the overhead.
~2.9 GB
download
AppImage — no install step. Windows installer. Android companion is served by your own backend, no separate store listing.
Full technical README on GitHub →