Kozmos
Atlas Chat
kozmos official app page

Atlas Chat

Local-first desktop chat for working with local AI models. Atlas now centers multi-thread local conversations, hardware-aware model discovery, profile memory, search, attachments, diagnostics, and isolated code execution around the current AtlasChat GitHub release.

download latestgithub source

Current Release

Version 1.1.5 is pulled from the AtlasChat GitHub release stream. Kozmos links to GitHub as the canonical download source so stale bundled installers do not linger on this page.
version
1.1.5
repository
olliedoganay/AtlasChat
desktop shell
Tauri + React
backend
local Python runtime
Windows x64MSI installer | 55.5 MiBsha256:a7069913e2c2575f5ca85db409de5804ddcb0c9a032239cc9ff12948bcd0e702macOS Apple Siliconunsigned DMG | 51.3 MiBsha256:3a2841d76185f51c4a27a4434f2e9c73b6510b35cccae8685742d3011bf23315macOS Intelunsigned DMG | 54.2 MiBsha256:9d72e100b3318ceaa2dbf6a9808d4b1d21a04363b2f5ebeca37c1f3c35eac00bLinux DebianDEB package | 89.7 MiBsha256:d8ae041f0196e8a711be1f3fbade53cfb5aeb2c73ee028e82a6075c3b448b6f0Linux FedoraRPM package | 89.7 MiBsha256:1525ad9be2b85a02f9ec49f3913f005f36b03893b33a5052d1f4d1f562b81d3cLinux portableAppImage | 163.5 MiBsha256:71c5484099cd48eb8b8c1f4cc8a2c0a0a1f0387bb56e80e634b8717ca9d4f1a8
release notesreadme

What Atlas Does

Atlas is no longer just an Ollama launcher page. It is a fuller local AI workspace with profiles, memory controls, local runtime routing, and code-run tooling.
workspace
  • multi-thread local chat workspace
  • thread rename, duplicate, branch, model lock, and temperature lock
  • local search across the active profile's chats
  • image and file attachments for richer sessions
models
  • Ollama is the default local runtime
  • Discovery page recommends models from detected hardware
  • pull commands are shown inside the app
  • LM Studio, llama.cpp server, vLLM, LocalAI, and generic OpenAI-compatible endpoints are supported
memory
  • per-user profiles with optional password protection
  • optional cross-chat memory with manual remember and forget controls
  • automatic and manual context compaction
  • local embedding model support, including nomic-embed-text:latest
execution
  • reasoning traces, token streaming, stop controls, and saved run diagnostics
  • generated code blocks can run in isolated Atlas Run windows
  • Docker is optional for chat and required for server-side code execution
  • HTML snippets can open in the local client preview

Local Runtime And Trust

Atlas expects a local model runtime on the same machine. Ollama is the default, while OpenAI-compatible local providers can be used when they expose a compatible endpoint.
architecture
  • Tauri shell starts a loopback-only backend.
  • Backend binds to 127.0.0.1 on a random local port.
  • Frontend requests use a per-launch instance token.
  • Unexpected origins are rejected.
  • Local state stays in Atlas-managed user directories.
source build requirements
  • Python 3.14+
  • Node 20+
  • Rust and Cargo
  • Tauri prerequisites
  • Local chat model
  • Local embedding model
runtime providers
  • Ollama
  • LM Studio
  • llama.cpp server
  • vLLM
  • LocalAI
  • generic OpenAI-compatible local runtimes