Skip to content

Model Selection Guide

The choice of language model is decisive for the writing experience. LoneWriter allows you to switch between different engines according to your needs for power, cost, or privacy.

Cloud Models

Recommended for those who do not have powerful hardware or desire maximum reasoning capability.

PRIVACY

Cloud providers may use your data to train their AI models. For total privacy, use Local Models.

ProviderExample ModelUsage ProfileOfficial Listing
Googlegemini-3.1-flash-liteExtremely fast and free model. Ideal for the MPC.View models
Anthropicclaude-opus-4-7The pinnacle of complex reasoning and advanced literary style.View models
OpenAIgpt-5.5The perfect balance between creative power and technical precision.View models
OpenRouteropenrouter/freeAccess to a rotation of high-quality free models.View models

Local Models

Ideal for authors who prioritize absolute privacy and data sovereignty.

ModelParametersOllama IDLM Studio (Repository)Notes
Gemma 48B (E4B)
(16GB RAM)
gemma4:e4blmstudio-community/gemma-4-E4B-it-GGUFThe new standard for local use. Very efficient for writing.
Llama 3.18B
(16GB RAM)
llama3.1:8blmstudio-community/Meta-Llama-3.1-8B-Instruct-GGUFExcellent overall performance and reasoning capability.
Qwen 3.59B
(16GB RAM)
qwen3.5:9blmstudio-community/Qwen3.5-9B-GGUFVery capable in creative tasks and highly multilingual.
Phi-33.8B
(8GB RAM)
phi3:minimicrosoft/Phi-3-mini-4k-instruct-ggufLightweight model ideal for devices with limited resources.

Recommendations by Task

For the Oracle (Lore Queries)

A wide context window is required to process your lore. In the cloud, gemini-3.1-flash-lite is an unbeatable free option. For local use, Llama 3.1 (llama3.1:8b) stands out for its reasoning capability to connect the threads of your story with total privacy.

For Draft Writing

If you are looking for an elegant style in the cloud, claude-opus-4-7 offers very human-like results. For a private, local experience, Qwen 3.5 (qwen3.5:9b) shines with its immense creativity and rich vocabulary in multiple languages.

For the Proposal Monitor (MPC)

The MPC works in the background and needs near-instant responses. gemini-3.1-flash-lite is ideal in the cloud, but if you have local hardware, Gemma 4 (gemma4:e4b) or Phi-3 (phi3:mini) offer near-zero latency without relying on an internet connection.

Designed and developed with ♥ by Sergio Sánchez with Antigravity.
Open Source & MIT License.