Open-source tooling and open-weight models, matched to the device.
selbsai is configured around memory, thermal and latency envelopes. That means the configurator maps to model classes, not to one frozen marketing name. The stack combines classic open-source software with open-weight model releases, and the exact deployment is matched to your hardware tier, languages, workload and licensing profile at provisioning time.
7B-8B class
Open-weight 7B-8B instruct stack
Fast local assistance for drafting, document Q&A and day-to-day offline copilots.
- RAM floor
- 16 GB
- Target speed
- 15-40 t/s
7B-13B class
Open-weight 7B-13B reasoning and retrieval stack
Balanced local reasoning for research, document-heavy workflows and multi-agent prototypes.
- RAM floor
- 32 GB
- Target speed
- 12-30 t/s
13B-30B class
Open-weight 13B-30B advanced local model stack
High-memory local intelligence for larger models, orchestration and heavier secure workloads.
- RAM floor
- 64 GB
- Target speed
- 8-18 t/s
Representative model families we actually watch.
These are representative families used to calibrate the current selbsai device tiers. Exact shipped models can change as newer open releases prove better on independent benchmarks.
Llama 3.1 8B Instruct
Meta Llama · Meta
Our compact baseline for fast local assistants, offline drafting and lighter retrieval on personal desktop nodes.
Ministral 3 14B Instruct
Mistral / Ministral · Mistral
A strong local step-up for multilingual work, longer-context tasks and richer instruction-following on balanced desktop systems.
Qwen3 30B A3B / 32B class
Qwen · Qwen
One of the main higher-capability open families we watch for larger reasoning, coding and multilingual deployments on high-memory local nodes.
Artificial Analysis
Independent model pages with direct comparisons across intelligence, speed, price, context window and methodology notes.
Hugging Face Open LLM Leaderboard
A widely used open-model benchmark hub for comparing community and lab releases across standard eval suites.
Arena Leaderboard
Useful for broad human-preference comparisons and keeping an eye on how major open releases stack up in live arena-style evaluation.
What actually gets installed
We do not promise that every Standard, Professional or Elite unit always ships with the same named model. A serious local AI hardware shop should choose the right open-weight stack for the job and update that recommendation as the ecosystem improves.
- Tier determines feasible model scale, memory floor and latency envelope.
- Persona presets determine tooling around the base model: OCR, retrieval, translation, image workflows or audit features.
- Final selection depends on language mix, data sensitivity, licensing constraints and whether the build optimizes for speed, depth or multimodality.
- Benchmark positions move over time, so this page links out to live third-party references instead of freezing stale claims into marketing copy.
OCR and document extraction
For invoice, receipt and document-heavy presets we pair the language model with open OCR and document-understanding tooling rather than relying on the base LLM alone.
Retrieval and reranking
Search-heavy presets use additional embedding and reranking components so large local indexes stay usable at real-world scale.
Preset workloads
The Researcher
Reads what you don't have time to read.
Drop in thousands of PDFs and ask questions across all of them. Citations included.
- PDF & DOCX ingestion
- Citation-aware Q&A
- Long-context search
The Vault
Sensitive data stays on the device.
Drafting, summarising, and review for legal, medical, and financial workflows. Nothing leaves the box.
- Air-gap mode
- Audit log
- Redaction filters
The Polyglot
Speaks 30 languages, no internet required.
Translate documents, hold conversations, and switch languages mid-sentence — entirely offline.
- DE / EN / FR / ES / IT / TR …
- Offline translation
- Voice in & out
The Accountant
Reads invoices and receipts, finds the numbers.
Pre-loaded with OCR, German tax-form templates, and CSV export. Built to talk to DATEV exports.
- OCR (Tesseract + LayoutLM)
- DATEV export
- Tax-form templates
The Archivist
A search engine for the last decade of your files.
Optimised for very large local indexes. Photos, scans, emails, contracts — instant recall.
- Vector index up to 4 TB
- Image-aware search
- Email / EML import
The Creative
Local image generation and a copywriting partner.
Pre-loaded with local image diffusion models and brand-voice prompt templates.
- Local image diffusion
- Brand-voice prompt library
- Inpainting & variations