Installation

Conversation Simulator runs entirely on your computer — no cloud inference, no account, no telemetry. You need a local model before any conversation can start; the app downloads one on first run after you accept its license.

System requirements

Requirement	Minimum	Notes
OS	macOS 13+, Ubuntu 22.04+, Windows 10	macOS 12 Monterey is no longer supported (app minimum is macOS 13)
CPU	Any 64-bit x86 or Apple Silicon	Apple Silicon recommended for CPU inference
RAM	8 GB	16 GB recommended for standard-tier models
GPU VRAM	0 GB (CPU fallback available)	4 GB+ for the starter model; see Local models
Disk	20 GB free	3–15 GB for model weights plus app data

1. Download and verify

Download the installer for your platform from the releases page:

Platform	File
macOS (Apple Silicon)	`ConversationSimulator_<version>_aarch64.dmg`
macOS (Intel)	`ConversationSimulator_<version>_x64.dmg`
Linux (x86_64)	`conversation-simulator_<version>_amd64.AppImage`
Windows (x86_64)	`ConversationSimulator_<version>_x64-setup.exe`

Verify the download against the checksums-sha256.txt file on the release page:

# macOS / Linux
shasum -a 256 ConversationSimulator_<version>_aarch64.dmg

# Windows PowerShell
Get-FileHash "ConversationSimulator_<version>_x64-setup.exe" -Algorithm SHA256

2. Install and launch

macOS: open the .dmg and drag the app to /Applications. On first launch, Gatekeeper may warn about an unidentified developer (pre-release builds are unsigned). Right-click the app → Open → Open to proceed.
Windows: run the .exe installer. SmartScreen may warn about an unrecognised publisher — click More info → Run anyway.

Linux: make the AppImage executable and run it directly — no installation needed:

chmod +x conversation-simulator_<version>_amd64.AppImage
./conversation-simulator_<version>_amd64.AppImage

The app starts its local conversation engine automatically. If the home screen reports that the engine did not start, see Troubleshooting.

3. Install a local model

On first launch the app shows a “No model loaded” banner. Open Settings → Models, pick a model, and accept its license to start the download. No model weights are bundled with the installer, and every download is verified against its SHA-256 checksum before loading.

The recommended starter is Qwen3 4B Instruct Q4_K_M (~2.5 GB, Apache-2.0). See Local models for recommendations by hardware.

The one-time model download is the only step that needs an internet connection. After it completes, everything works offline.

Beta builds — direct-download channel

Beta builds are published to GitHub Releases as versioned pre-releases (e.g. v0.1.0-beta.1). They are distinct from the Steam beta branch.

In-app update notice

The app checks for a new beta on launch and shows a non-intrusive banner on the home screen when one is found. The banner never appears during an active conversation session. Click View notes to open the release page, or Install to open it and download the new build. The check is skipped silently when you are offline.

Rollback

Every beta release remains permanently downloadable from its versioned release page (e.g., releases/tag/v0.1.0-beta.1). To roll back:

Download the installer from the previous versioned release page.
Install over the current version — the Windows installer and macOS DMG both support in-place downgrades.
Data created in a newer beta is forward-compatible with older betas at the same schema version (see Schema versioning).

Data locations

Path	Purpose
`~/.convsim/db/`	Session database (SQLite)
`~/.convsim/data/`	Exported data and pack cache
`~/.convsim/logs/`	Runtime logs
`~/.convsim/models/llm/`	Downloaded model weights

Override any of these with environment variables: CONVSIM_DB_DIR, CONVSIM_DATA_DIR, CONVSIM_LOG_DIR, CONVSIM_MODELS_DIR.

Next steps

Quickstart — run your first conversation
Local models — choose a model for your hardware
Troubleshooting — common setup problems
Platform notes — Gatekeeper, SmartScreen, WebView2, audio permissions