Known Issues

Alpha means rough edges. These are the ones I know about, each with a way around it when there is one. If you find one that isn't here, I want it. Things Orbweaver deliberately doesn't do, like text-completion presets or STscript, aren't here. Those are on Coming from SillyTavern. If you hit something that isn't listed, Troubleshooting is the place to start.

Connections

Some roles are unavailable on your own server. Rerank works with vLLM or a llama.cpp reranking server, but not through the built-in Ollama, LM Studio or KoboldCpp routes. Image generation is not available on an own-server connection. See Connect a model.

A model on your own server that reports no context size is assumed to have 8,192 tokens. Orbweaver trims history to that, and the context window row under Advanced says (assumed). Ollama does report the window it uses, but only when the model's Modelfile sets num_ctx or the model is loaded. Otherwise press Override on context window under What this server accepts and type the real one.

Game mode

Gemini can exhaust a short reply budget when the game state travels in the reply. A game turn can stop before producing a usable reply or miss a requested state update. Try Separate pass under Delivery model in the host's Game tab. If you keep In the reply, allow more output space in the preset rather than assuming the ordinary short-reply budget is enough. Check Game actions on this turn and correct or resync missing state. See RPG mode.

Insights

Spending figures are partial, not your provider bill. Some background work is missing from the totals. Backup and restore preserve image-generation history, but not embedding or compaction spend history. Read Accounting coverage beside the figures before relying on them. See Stats and Insights.