MOTHER LLM
The sovereign general-purpose assistant — everyday chat and document making.
MOTHER LLM is the general-purpose assistant of the MOTHER family — for everyday conversation, drafting and document creation. It runs on the sovereign MOTHER stack with retrieval grounding on your own knowledge base, and today reasons on the MOTHER CORE backbone while its own dedicated pipeline is trained.
Capabilities
General conversation
Open-ended chat, Q&A and everyday assistance across topics.
Document making
Draft, edit and format documents — reports, letters, briefs and more.
Retrieval-grounded
Ground answers in your own knowledge base for accurate, cited responses.
Tool & agent aware
Hands off to the MOTHER agent fleet for tasks that need tools.
Sovereign & private
Runs on UK sovereign infrastructure with no data egress.
Architecture
- 1
MOTHER stack
Served on the sovereign MOTHER platform; general chat + document workflows route through the shared reasoning core.
- 2
CORE backbone (today)
MOTHER LLM currently reasons on the MOTHER CORE model while its own dedicated general pipeline is trained.
- 3
Grounding
Retrieval over your documents keeps answers accurate and attributable.
Specifications
| Pipeline | MSAIGlobal/train-llm-from-scratch · decoder-only · bf16 |
| Tokenizer | tiktoken r50k_base · EOT 50256 |
| Corpus prep | prepare_corpus_hdf5.py — JSONL → HDF5 |
| Scale | 13M → 1B params on a single GPU |
| Today's chat | resolves to mother-core-7b / HF fallback |
| Registry | unit mother-llm · MOTHER_LLM_URL |
| Hosting | UK sovereign · on-node |
Model evaluation — industry benchmarks
| Benchmark | Metric | Result | Status |
|---|---|---|---|
| Today (via CORE-7B) | agentic rate | 77% | ◉ measured |
| MMLU | 5-shot acc | — | ○ to run |
| HellaSwag | acc | — | ○ to run |
| LAMBADA | perplexity | — | ○ to run |
| WikiText | perplexity | — | ○ to run |
Benchmarks scheduled on this build — measured results are recorded here as each run completes.
What it's for
Everyday assistant
A private, sovereign assistant for general questions and drafting.
Document workflows
Create and refine documents grounded in your own material.
Front door to the fleet
A natural-language entry point that delegates to MOTHER agents and tools.
Safety & governance
- Human-in-the-loop on sensitive or outward-facing actions.
- Content governance and refusal policy applied to outputs.
- No data egress — inference stays on sovereign infrastructure.
Frequently asked
What is MOTHER LLM?
MOTHER LLM is the sovereign general-purpose assistant of the MOTHER family for everyday chat and document making, grounded in your own knowledge base and running on UK infrastructure.
Is MOTHER LLM its own model?
It has its own pipeline and registry entry; today it reasons on the MOTHER CORE backbone while its dedicated general model is trained.
Is it sovereign?
Yes — it runs on UK sovereign infrastructure with no data egress, on-prem or air-gapped options available.
The MOTHER model family
A frontier world model — not text tokens, but Eyes, Ears, a Mouth, a Brain, memory and understanding, all at once.
A sovereign British reasoning model — the mind behind MOTHER.
The eyes of MOTHER — real-time detection, tracking and segmentation across every feed.
Describe it — MOTHER Code builds the website, app or game.
Build on MOTHER LLM
Sovereign, on-node and observe-and-advise by design. Talk to us about access and deployment.