ZEBRA PIPE
Local LLM Orchestrator
SETTINGS :8081
...
Model
Main GGUF model
Vision projector / mmproj
RESCAN MODELS
LOAD / RELOAD
STOP
Runtime
Context
GPU layers
Threads (0=auto)
Parallel slots
Batch
UBatch
Flash Attention
Jinja templates
Auto-load selected model on container start
Generation Defaults
Temperature
Top P
Top K
Additional llama-server arguments
SAVE WITHOUT RELOAD
Discovered models
MCP / Agent Engine
Configure MCP servers here. Streamable HTTP is preferred; stdio and legacy SSE are supported. Agent jobs on :8082 can execute only tools in the allowlist.
Enable MCP agents
Max agent iterations
Tool timeout (seconds)
Policy
Allowlist only
Allow all discovered tools
MCP servers JSON
SAVE MCP CONFIG
DISCOVER TOOLS
llama-server log
