Jarvis Helper

Jarvis Helper is a versioned Ollama package of OpenBMB MiniCPM5-1B, configured for lightweight, local helper tasks in Jarvis Voice.

It is intended for short, bounded workloads such as progress phrases, metadata summaries, and other internal text transformations without consuming the primary cloud or local chat model.

Quick start

ollama pull bigsk1/jarvis-helper:minicpm5-1b-q4_k_m-v1
ollama run bigsk1/jarvis-helper:minicpm5-1b-q4_k_m-v1

Model details

• Base model: OpenBMB MiniCPM5-1B
• GGUF source: OpenBMB MiniCPM5-1B-GGUF
• Quantization: Q4_K_M
• Model size: approximately 688 MB
• Context window: 8,192 tokens
• Interface: ChatML
• Modality: text only
• License: Apache 2.0

This is a packaging and configuration of the upstream OpenBMB model. It is not a separately trained foundation model.

Intended use

Jarvis currently uses this model as an optional provider for:

• Short spoken status updates
• Stash metadata summaries
• Long-text summaries
• Other bounded, non-critical helper operations

Each role is explicitly enabled. The helper remains independent of Jarvis’s primary chat, vision, and embedding models.

CPU and GPU operation

The Q4_K_M model is small enough to run on CPU-only Ollama installations. Ollama can also use available GPU acceleration automatically.

Jarvis exposes a separate device setting:

JARVIS_HELPER_LLM_DEVICE="auto"

Set it to cpu to force CPU inference.

Jarvis configuration

JARVIS_HELPER_LLM_BASE_URL="http://127.0.0.1:11434"
JARVIS_HELPER_LLM_MODEL="bigsk1/jarvis-helper:minicpm5-1b-q4_k_m-v1"

STATUS_LLM_PROVIDER=helper
STASH_SUMMARIZE_LLM_PROVIDER=helper
TEXT_SUMMARIZER_LLM_PROVIDER=helper

Jarvis users can install the configured model with:

./bin/setup-helper-llm --mode cloud

Replace cloud with local when configuring local mode.

Limitations

This is a compact 1B-parameter helper model. It is not intended to replace a larger primary chat model.

It is not recommended for:

• Embeddings or semantic retrieval
• Image or vision analysis
• Complex reasoning
• High-stakes decisions
• Reliable tool selection
• Unsupervised classification

Instruction following can vary, particularly for complex schemas and classification tasks. Applications should validate structured output and retain deterministic or primary-model fallbacks.

Versioning

The minicpm5-1b-q4_k_m-v1 tag identifies this Jarvis packaging release. Future model or configuration changes will use a new versioned tag rather than requiring users to manage model digests manually.

Attribution

MiniCPM5 was developed by OpenBMB. Jarvis Helper packages the official Q4_K_M GGUF for convenient Ollama installation and Jarvis integration.

Github repo

https://github.com/bigsk1/jarvis-voice