Requirements
The short answer
Any machine with Docker, 8 GB of RAM and 50 GB of free disk will run Chronicler. Whether it runs pleasantly depends mostly on whether it has a GPU.
Memory
The number that matters is the memory Docker can use, not the memory in the machine. On Windows, Docker runs through WSL2 with its own allocation. See Performance and memory for how to raise it.
| Docker memory | What you get |
|---|---|
| Under 4 GB | Won't work. The backend gets killed as soon as a model loads. |
| 4–8 GB | Runs, small model only, slow. Fine for trying it out. |
| 8–16 GB | Comfortable for a handful of people with the default model. |
| 16 GB+ | Room for a larger model, or several people asking at once. |
Graphics card
Optional, and the single biggest difference to how the product feels.
- No GPU - everything works, on the CPU. A cold answer takes one to three minutes; once the model is warm, answers come in tens of seconds.
- NVIDIA - answers in seconds. Needs the NVIDIA Container Toolkit so that containers can see the card (on Windows, a current driver plus Docker Desktop's WSL2 backend).
- AMD - supported on Linux only, via ROCm, and best-effort: it works on common cards but is less tested than NVIDIA. Roughly which model suits which machine - the full list, with what each is good at, is on Models:
| Hardware | Model |
|---|---|
| No GPU, 8 GB RAM | gemma4:e2b - the default |
| No GPU, 16 GB RAM | gemma4:e4b - better answers, slower on CPU |
| 10 GB VRAM | gemma4:12b |
| 20–24 GB VRAM | gemma4:26b / gemma4:31b |
Disk
| Item | Size |
|---|---|
| Container images | ~6 GB |
| Default model | ~3 GB (larger models: 6–20 GB) |
| Database and documents | Grows with your library - budget several times the size of your files, since text and search index are stored alongside them |
50 GB free is a sane starting point; 100 GB if you're indexing a large archive. The Admin → Settings page caps how much of the disk Chronicler will use.
Operating system
Windows 10/11 with Docker Desktop using the WSL2 backend.
The Hyper-V backend cannot pass a GPU through to containers at all. If you have a graphics card and want to use it, switch in Docker Desktop → Settings → General → Use the WSL 2 based engine.
docker --version
docker compose version
Any distribution with Docker Engine 24+ and the Compose plugin. This is the best platform for a shared server: no VM in the way, and both NVIDIA and AMD GPUs can be used.
docker --version
docker compose version
Network
- The app reaches the server on port 8410. On one machine that's
http://localhost:8410; for colleagues it'shttp://<server-ip>:8410. - Outbound HTTPS to
api.chroniclerlm.comfor licence checks, and to Docker Hub and GitHub for installs and updates. - Exposing the server beyond a trusted network means putting HTTPS in front of it - see Security and networking.
A licence key
Self-hosted servers require one. Without it the server starts, but the parts that touch documents refuse to work. See Licensing and activation.