Apple pour l’expérience. x86 pour l’échelle.
Vous ne vendez pas des boîtes. Vous vendez une gamme cohérente d’appareils. Apple offre l’expérience locale la plus simple et la plus silencieuse. Le x86 apporte plus de flexibilité budgétaire d’un côté et plus d’extension GPU de l’autre. Il vous faut les deux.
Matériel neuf
Obsidian Personal
Apple Mac mini M2 · 16 GB unified · 1 TB
Best Apple entry
Affordable entry device that runs small local LLMs smoothly with almost no setup friction.
Unified memory behaves like VRAM, so Ollama and LM Studio feel unusually smooth for the size and price.
- Plateforme
- Apple Silicon
- Accél. IA
- Unified memory
- Adéquation modèle
- Private assistant workflows
- Évolutivité
- None
Silent desktop AI for personal assistants, Q&A and offline drafting.
Obsidian Flex
GMKtec Ryzen 7 8845HS · 32 GB
Flexible non-Apple budget device
A lower-cost Windows/Linux device with more RAM flexibility and a strong integrated GPU for local AI.
This is the strongest value-oriented alternative if you want higher RAM headroom without moving into Apple pricing.
- Plateforme
- x86 Ryzen
- Accél. IA
- Radeon 780M iGPU
- Adéquation modèle
- Office knowledge workflows
- Évolutivité
- Moderate
Budget-conscious local AI, Linux-first setups and users who want more tweakability.
Obsidian Core
Apple Mac mini M4 · 32 GB unified
Minimum serious sellable configuration
Balanced Apple-first device for continuous local AI workloads, retrieval and multi-agent prototypes.
The base M4 exists around 929 EUR, but the 32 GB build is the first configuration worth selling for sustained local AI.
- Plateforme
- Apple Silicon
- Accél. IA
- Neural Engine
- Adéquation modèle
- Office knowledge workflows
- Évolutivité
- None
The best balance of simplicity, silence and useful local model headroom.
Obsidian Pro
Apple Mac mini M4 Pro · 64 GB unified
Best Apple high-end
High-memory compact device for larger local models, multi-agent orchestration and heavier professional workflows.
Apple Silicon is fixed at purchase time. RAM and GPU are not field-upgradeable, so this is the tier to buy once and size correctly.
- Plateforme
- Apple Silicon
- Accél. IA
- Neural Engine + stronger GPU
- Adéquation modèle
- Pro knowledge and coding workflows
- Évolutivité
- None
Team-level local AI, orchestration-heavy work and advanced secure inference on the desk.
Obsidian Edge
Minisforum AI X1 Pro · Ryzen AI 9
Scale-first non-Apple path
Expandable device for heavier inference, shared usage and future GPU-backed scaling beyond the Mac mini ceiling.
This is the device to choose when you care more about expansion, eGPU support and scale than absolute simplicity.
- Plateforme
- x86 Ryzen AI
- Accél. IA
- NPU + eGPU path
- Adéquation modèle
- Expandable multi-user inference
- Évolutivité
- High
Server-grade local AI, multi-user deployments and the path toward GPU-backed scale.
Matériel renouvelé
Appareils de seconde vie sourcés professionnellement, vendus par classe de performance et testés avant provisioning.
Renewed Starter
Second-life desktop · 16 GB class · 1 TB
Lower entry price
Professionally sourced second-life hardware configured as an affordable local AI device for everyday private assistance.
Exact hardware may vary within this performance class. selbsai inspects, resets, updates, provisions and tests the unit before shipment.
- Plateforme
- Renewed Apple or x86
- Accél. IA
- Class-matched local runtime
- Adéquation modèle
- Private assistant workflows
- Évolutivité
- Varies by sourced unit
Pilots, personal assistants, email drafting, light research and small document workflows.
Renewed Office
Second-life desktop · 32 GB class · 2 TB
Best renewed value
A second-life office device for serious document, coding, research and back-office AI work at a lower acquisition cost.
Exact hardware may vary within this performance class. selbsai records source, condition and QA before assigning the device to an order.
- Plateforme
- Renewed Apple or business x86
- Accél. IA
- Class-matched local runtime
- Adéquation modèle
- Office knowledge workflows
- Évolutivité
- Varies by sourced unit
Small firms, agencies, advisors and internal teams that want private AI without buying new hardware.
Renewed Pro
Second-life high-memory desktop · 64 GB class
High capability, lower capex
A sourced high-memory renewed device for larger local models and heavier professional workflows when new hardware is not required.
Exact hardware may vary within this performance class. Availability depends on vetted second-life supply and final QA.
- Plateforme
- Renewed Apple or workstation x86
- Accél. IA
- High-memory local runtime
- Adéquation modèle
- Pro knowledge and coding workflows
- Évolutivité
- Varies by sourced unit
Professional document estates, coding assistants, multi-workflow setups and larger private knowledge bases.
Vérification terrain
Les Mac mini sont d’excellentes appliances IA parce qu’ils sont silencieux, stables et sans friction. Le revers, c’est l’absence d’évolution : la RAM et le GPU sont figés à l’achat. La gamme Apple doit donc être vendue avec la bonne configuration mémoire dès le départ.
C’est pourquoi la configuration M4 32 Go est le premier véritable Core sérieux, et pourquoi l’appareil Edge x86 existe au sommet de la gamme.
Règle de packaging
Apple est le meilleur choix quand la priorité est le silence, la simplicité et une expérience desktop premium. Le x86 est le bon choix quand la priorité est l’extension, l’usage partagé et l’évolutivité future via GPU.
- Apple = expérience + simplicité
- x86 = puissance + évolutivité
- La ligne Obsidian doit présenter les deux comme des appareils de premier rang
Vue d’ensemble des capacités
| Capacité | Personal Apple Mac mini M2 · 16 GB unified · 1 TB 579,00 € | Flex GMKtec Ryzen 7 8845HS · 32 GB 359,99 € | Renewed Starter Second-life desktop · 16 GB class · 1 TB 399,00 € | Renewed Office Second-life desktop · 32 GB class · 2 TB 749,00 € | Renewed Pro Second-life high-memory desktop · 64 GB class 1 299,00 € | Core Apple Mac mini M4 · 32 GB unified 1 143,97 € | Pro Apple Mac mini M4 Pro · 64 GB unified 1 903,99 € | Edge Minisforum AI X1 Pro · Ryzen AI 9 799,00 € |
|---|---|---|---|---|---|---|---|---|
| Architecture | Apple Silicon | x86 Ryzen | Renewed Apple or x86 | Renewed Apple or business x86 | Renewed Apple or workstation x86 | Apple Silicon | Apple Silicon | x86 Ryzen AI |
| Memory | 16 GB | 32 GB | 16 GB | 32 GB | 64 GB | 32 GB | 64 GB | 96 GB |
| GPU path | Integrated only | Integrated only | Integrated only | Integrated only | GPU-ready | Integrated only | GPU-ready | GPU-ready |
| AI acceleration | Unified memory | Radeon 780M iGPU | Class-matched local runtime | Class-matched local runtime | High-memory local runtime | Neural Engine | Neural Engine + stronger GPU | NPU + eGPU path |
| Model capability | Private assistant workflows | Office knowledge workflows | Private assistant workflows | Office knowledge workflows | Pro knowledge and coding workflows | Office knowledge workflows | Pro knowledge and coding workflows | Expandable multi-user inference |
| Upgradeability | None | Moderate | Varies by sourced unit | Varies by sourced unit | Varies by sourced unit | None | None | High |
| Noise / efficiency | 5/5 quiet | 4/5 quiet | Inspected quiet office profile | Inspected quiet office profile | Inspected workstation profile | 5/5 quiet | 5/5 quiet | 3/5 workstation |
| Best use | Silent desktop AI for personal assistants, Q&A and offline drafting. | Budget-conscious local AI, Linux-first setups and users who want more tweakability. | Pilots, personal assistants, email drafting, light research and small document workflows. | Small firms, agencies, advisors and internal teams that want private AI without buying new hardware. | Professional document estates, coding assistants, multi-workflow setups and larger private knowledge bases. | The best balance of simplicity, silence and useful local model headroom. | Team-level local AI, orchestration-heavy work and advanced secure inference on the desk. | Server-grade local AI, multi-user deployments and the path toward GPU-backed scale. |
Personal
Apple Mac mini M2 · 16 GB unified · 1 TB
- Architecture
- Apple Silicon
- Memory
- 16 GB
- GPU path
- Integrated only
- AI acceleration
- Unified memory
- Model capability
- Private assistant workflows
- Upgradeability
- None
- Noise / efficiency
- 5/5 quiet
- Best use
- Silent desktop AI for personal assistants, Q&A and offline drafting.
Flex
GMKtec Ryzen 7 8845HS · 32 GB
- Architecture
- x86 Ryzen
- Memory
- 32 GB
- GPU path
- Integrated only
- AI acceleration
- Radeon 780M iGPU
- Model capability
- Office knowledge workflows
- Upgradeability
- Moderate
- Noise / efficiency
- 4/5 quiet
- Best use
- Budget-conscious local AI, Linux-first setups and users who want more tweakability.
Renewed Starter
Second-life desktop · 16 GB class · 1 TB
- Architecture
- Renewed Apple or x86
- Memory
- 16 GB
- GPU path
- Integrated only
- AI acceleration
- Class-matched local runtime
- Model capability
- Private assistant workflows
- Upgradeability
- Varies by sourced unit
- Noise / efficiency
- Inspected quiet office profile
- Best use
- Pilots, personal assistants, email drafting, light research and small document workflows.
Renewed Office
Second-life desktop · 32 GB class · 2 TB
- Architecture
- Renewed Apple or business x86
- Memory
- 32 GB
- GPU path
- Integrated only
- AI acceleration
- Class-matched local runtime
- Model capability
- Office knowledge workflows
- Upgradeability
- Varies by sourced unit
- Noise / efficiency
- Inspected quiet office profile
- Best use
- Small firms, agencies, advisors and internal teams that want private AI without buying new hardware.
Renewed Pro
Second-life high-memory desktop · 64 GB class
- Architecture
- Renewed Apple or workstation x86
- Memory
- 64 GB
- GPU path
- GPU-ready
- AI acceleration
- High-memory local runtime
- Model capability
- Pro knowledge and coding workflows
- Upgradeability
- Varies by sourced unit
- Noise / efficiency
- Inspected workstation profile
- Best use
- Professional document estates, coding assistants, multi-workflow setups and larger private knowledge bases.
Core
Apple Mac mini M4 · 32 GB unified
- Architecture
- Apple Silicon
- Memory
- 32 GB
- GPU path
- Integrated only
- AI acceleration
- Neural Engine
- Model capability
- Office knowledge workflows
- Upgradeability
- None
- Noise / efficiency
- 5/5 quiet
- Best use
- The best balance of simplicity, silence and useful local model headroom.
Pro
Apple Mac mini M4 Pro · 64 GB unified
- Architecture
- Apple Silicon
- Memory
- 64 GB
- GPU path
- GPU-ready
- AI acceleration
- Neural Engine + stronger GPU
- Model capability
- Pro knowledge and coding workflows
- Upgradeability
- None
- Noise / efficiency
- 5/5 quiet
- Best use
- Team-level local AI, orchestration-heavy work and advanced secure inference on the desk.
Edge
Minisforum AI X1 Pro · Ryzen AI 9
- Architecture
- x86 Ryzen AI
- Memory
- 96 GB
- GPU path
- GPU-ready
- AI acceleration
- NPU + eGPU path
- Model capability
- Expandable multi-user inference
- Upgradeability
- High
- Noise / efficiency
- 3/5 workstation
- Best use
- Server-grade local AI, multi-user deployments and the path toward GPU-backed scale.