Offline by design.
Your model runs on your own computer and the app talks to it at 127.0.0.1:1234 by default. Your saves, Shards and memory stay with you. No accounts, no telemetry, no cloud AI.
WordBlock Labs · A local roleplay frontend · SID 6.7.5-beta · Windows test build out
Private, turn-based roleplay with a partner who stays in its lane.
SoloRoleplayer// is a frontend for your own model, built in the style of the MUSH games. You write your side. The companion writes its side. The app keeps the facts of the scene. It runs on your computer, with nothing watching.
Recommended local stack · for adults 18+
Mistral Nemo (abliterated) + Anti-H 2.0
Status: SID 6.7.5-beta is out as a free Windows test build, with the full source, on GitHub. I built it on a Mac and have no Windows computer, so I checked the Windows portable file under Wine on a Mac: it starts, runs the full setup, creates a companion and loads a world. It has not been run on a real Windows PC yet. Please report problems on the support page.
One companion. One lane.
The companion writes only its own words, actions and thoughts. The app resists head-hopping in two ways: a trained LoRA, Anti-H 2.0, and guards in the app itself.
Bring your own server.
Start any OpenAI-compatible local server (LM Studio, llama.cpp, KoboldCPP, Ollama) and the app connects. Type /connect if yours uses another port.
Why this exists
Most roleplay tools break the same two ways.
Head-hoppingThe AI starts writing both sides, steals your reactions and turns into a floating narrator.
Lost factsClothing changes, objects move, promises vanish, and the persona softens when you are not looking.
Our fixThe app keeps the facts (STASIS scene memory and a scribe), sends your companion's persona every turn, and Anti-H 2.0 keeps the model in its lane. Measured in a published benchmark, with the weak spots published.
SoloRoleplayer//
The frontend you play in. Free for noncommercial use; commercial use by license. A Windows test build and the full source are on GitHub.
Download on GitHubMistral Nemo (abliterated)
The recommended model: a public abliterated build of Mistral Nemo 2407 on Hugging Face.
Open on Hugging FaceAnti-H 2.0
Our LoRA for Mistral Nemo: keeps the companion in its own lane and the writing clean. Published on Hugging Face as solo-r-lane-lora.
Get Anti-H 2.0 on Hugging FaceMade by WordBlock Labs.
A solo project. SoloRoleplayer// is for adults, 18 and over. It runs offline on your own computer and is not monitored: you are responsible for how you use it.