Nodes/Simple LlamaCPP Client/🦙 Llama.cpp Router Client (Chat + Vision + Audio + Video + Stream)
ComfyUI Node

🦙 Llama.cpp Router Client (Chat + Vision + Audio + Video + Stream)

A ComfyUI node in LLM / Simple llama.cpp with 17 inputs and 5 outputs.

By ai-joe-git·Created 6 months ago·Updated 2 months ago· 3
🦙 Llama.cpp Router Client (Chat + Vision + Audio + Video + Stream)
  • image
  • response
  • thinking
  • json
  • raw
  • model_used
â—„server_urlhttp://127.0.0.1:8888â–º
â—„modelâ–º
â—„system_promptYou are a helpful assistant.â–º
â—„promptHello!â–º
â—„streamtrueâ–º
â—„max_tokens0â–º
â—„temperature0.80â–º
â—„seed-1â–º
â—„json_modefalseâ–º
â—„unload_after_generatefalseâ–º
â—„text_postprocessfix_mojibakeâ–º
â—„audio_pathâ–º
â—„video_pathâ–º
â—„api_keyâ–º
â—„json_schema_hintâ–º
â—„stop_customâ–º
CategoryLLM / Simple llama.cpp

Inputs (17)

NameTypeDefaultDescription
server_urlSTRINGhttp://127.0.0.1:8888llama.cpp server base URL (router mode or single-model)
modelSTRINGModel name — click Fetch Models button to pick from the server
system_promptSTRINGYou are a helpful assistant.System prompt sent before the user message
promptSTRINGHello!User message / question
streamBOOLEANtrueEnable SSE streaming for real-time token output
max_tokensINT00–131072Max tokens to generate (0 = server default)
temperatureFLOAT0.800–2Sampling temperature
seedINT-1-1–2147483647RNG seed (-1 = random)
json_modeBOOLEANfalseForce JSON-only output via response_format
unload_after_generateBOOLEANfalseCall /models/unload on the server after generation
text_postprocessCOMBOfix_mojibakeText post-processing mode
imageoptIMAGEOptional image for vision models (first frame used)
audio_pathoptSTRINGOptional path to audio file (WAV/MP3) for audio models
video_pathoptSTRINGOptional path to video file for video models
api_keyoptSTRINGAPI key for Bearer token auth (leave empty if none)
json_schema_hintoptSTRINGJSON schema hint appended to system prompt when json_mode is on
stop_customoptSTRINGCustom stop string (leave empty for none)

Outputs (5)

NameTypeDescription
responseSTRINGClean response text with thinking blocks removed
thinkingSTRINGExtracted thinking/reasoning content (empty if model produced none)
jsonSTRINGParsed JSON object (empty if json_mode is off or no JSON found)
rawSTRINGFull raw server response as JSON string
model_usedSTRINGThe model name that was actually used for this request