No snapshot yet - click ⊕ CONTEXT in the Chat tab to generate one.
Context Management
turn-pairs kept
Canvas context occupies a single CONTEXT slot - replaced on each injection, never stacked.
ADD DRIFT CORRECTION
Describe what the LLM got wrong. Appended to corrections.md permanently.
Startup Command
llama-server -hf unsloth/Qwen3.5-9B-GGUF:UD-Q6_K_XL \
--host 0.0.0.0 --port 8180 \
--ctx-size 262144 --n-gpu-layers 99 \
--chat-template-kwargs '{"enable_thinking":false}'
Prompt files: world_model_data/prompts/ - edit and save, reloads on next request.