Start with a source snapshot
Lilith's current repository separates shared Core services, Agent behaviour, and the console host. The website gives each release an exact source commit. Download that snapshot instead of assuming today's development branch has the same behaviour as a recorded demonstration.
V1 is the local conversation foundation. V2 adds model-callable C# tools. V3 is a development snapshot adding fresh scenario conversations. The website's release pages describe only what that particular version introduced.
Prepare a provider and model
The console supports Ollama and LM Studio. Use an installed model suited to your hardware. Tool and scenario requests require a model that supports tool calling; ordinary text generation alone does not establish that capability. A routine application build does not require uninstalling your provider or deleting model files.
Build and launch
dotnet build Interfaces/ConsoleApp/LilithConsole/LilithConsole.csproj -c Release
dotnet run --project Interfaces/ConsoleApp/LilithConsole/LilithConsole.csprojUse the snapshot's README for its exact setup behaviour. Current projects use the .NET 10 SDK, while Core targets .NET 8. The Windows recording host has additional Windows Forms and FFmpeg requirements.
Add a voice when you want one
Kokoro narration is optional for ordinary chat. Its ONNX model is a separate asset; the application does not claim that text success proves speech playback. Narrated recordings require real synthesis, while a layout preview deliberately uses silent fixtures.
Keep the conversation recoverable
Saved chats use unique identities and atomic file replacement. A failed request must not silently become a completed conversation. Inspect history through the application methods while idle, and keep generated chats separate from source you intend to publish.