NanoRP

A lightweight, self-hosted AI roleplay chat.

Bring your own characters. Bring your own models. Everything stays on your machine.

• Small by design
• Quick start
  • Your data
• First steps
• Features
• Configuration
• Development
  • Layout
• Contributing
• License

Most self-hosted chat UIs arrive as a Node server, a Python API, a Postgres instance, and a Redis cache. NanoRP is one binary and one file.

• One static binary. No Node.js, no Python, no Electron at runtime. The server is Rust (Axum); the whole app ships as a single executable plus static assets.
• No database server. State lives in one SQLite file, with SQLite compiled into the binary — nothing to install, provision, or back up separately.
• No JS framework runtime. The UI is Rust compiled to WebAssembly (Leptos), server-rendered first and hydrated by a size-optimized WASM bundle (opt-level = "z" + LTO). Assets are pre-compressed at build time, so the server never re-compresses a response.

It's built to sit quietly on a Raspberry Pi, a free-tier VPS, or your laptop alongside everything else. Your GPU should be busy running the model, not your web stack.

docker run -d --name nanorp \
  -p 3000:3000 \
  -v nanorp-data:/data \
  ghcr.io/coder3101/nanorp:latest

Or with Compose:

docker compose up -d

Then open http://localhost:3000.

Images are published for linux/amd64 and linux/arm64 on every release.

Connecting to Ollama on the host? Use http://host.docker.internal:11434 as the provider URL. On Linux, add --add-host=host.docker.internal:host-gateway to docker run.

The SQLite database, avatars, and attachments are persisted in the nanorp-data volume, mounted at /data.

The container runs as a non-root user (uid 10001). Named volumes work out of the box; if you bind-mount a host directory instead (-v /path/on/host:/data), chown -R 10001 it first or pass --user to match its owner.

1. Open Settings → Providers and add one:
   • Ollama — a URL like http://localhost:11434, no API key
   • OpenAI-compatible — a URL like https://api.openai.com plus a key
2. Go to Characters and create one.
3. Hit Chat, pick a model, and start talking.

Characters

• Name, role, personality, system prompt, greeting, and avatar
• Generate a character from a plain-text description using any configured model (structured JSON output) — review, then save or retry
• {{char}} / {{user}} placeholders resolved at prompt-build time
• Duplicate, search, and import/export as JSON

Models

• Ollama and any OpenAI-compatible endpoint
• Configure several providers and switch model mid-conversation
• API keys are encrypted at rest (ChaCha20-Poly1305)
• Sampling controls: temperature, top-p, max tokens

Chat

• Token-by-token streaming, with stop-and-cancel that reaches the provider
• Collapsible reasoning ("thinking") blocks for reasoning models
• Edit your messages — text and images — then regenerate from that point
• Image attachments for vision models (up to 5 per message)
• Markdown rendering, one-click copy

Interface

• Responsive down to phone widths, with a proper mobile layout
• Light and dark themes, no flash on first paint
• Keyboard navigable menus and dialogs, focus trapping, ARIA labels, and prefers-reduced-motion support

Yours

• No telemetry, no accounts, no cloud dependency
• Every message, image, and key stays in a directory you control

Everything is optional; the defaults below are what the container uses.

You'll need the Rust toolchain, the wasm32-unknown-unknown target, and cargo-leptos:

rustup target add wasm32-unknown-unknown
cargo install cargo-leptos --locked

cargo leptos watch                      # dev server on http://127.0.0.1:3000
cargo leptos build --release            # production build into target/

Tests and lints, the same way CI runs them:

cargo fmt --all --check
cargo test --features ssr --no-default-features
cargo clippy --all-targets --features ssr --no-default-features -- -D warnings
cargo clippy --lib --target wasm32-unknown-unknown \
  --features hydrate --no-default-features -- -D warnings

In development, app data lives in ~/.config/nanorp/ (macOS: ~/Library/Application Support/nanorp/).

Issues and pull requests are welcome. Keep changes focused, make sure cargo fmt, cargo test, and cargo clippy pass, and match the surrounding style.

MIT © coder3101