Check memory and storage
See what your computer can handle.
A DEVICETERRA PRODUCT
Technology for emerging-market businesses and institutions.
Explore Deviceterra ↗LOCAL AI · YOUR HARDWARE · YOUR TASK
Find local AI that fits your computer.
Choose your computer and task. Get model options, software guidance and a storage plan.
Save or restore setup
Optional · saved only on this browser
STEP 1 · YOUR COMPUTER
Tell us about your computer
Not sure about a detail? Use the help beside each field.
More computer settingsCPU, storage type and model library
Use a device starterOptional example settings you can edit
Example settings only. Check and correct them for your computer.
Details are entered by you. Model matches are estimates to test.
System memory32 GBUsable verified VRAMnot assumedGPU catalogue151 profilesSpeedbenchmark required
How MamiLens helps
Choose suitable software
Get options for your model and setup.
Install and test
Follow a practical setup and test plan.
WANT MORE CHOICES?Explore all 98 modelsOpen a searchable list of every source-checked model⌄
OPTIONAL MODEL EXPLORER
98 source-checked profiles
Use this only when you want to search beyond the three recommendations.
⌕
Q
Qwen 3.5
Alibaba Cloud · 9B
Officially supported35/100 evidence coverage
Qwen 3.5 9B is a curated dense profile for general, coding, reasoning workloads.
MODEL FILE~6.6 GB
SAFE RAM FLOOR13 GB
GPU TARGET10 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
L
Llama 3.1
Meta · 8B
Officially supported35/100 evidence coverage
Llama 3.1 8B is a curated dense profile for general, writing, coding workloads.
MODEL FILE~4.9 GB
SAFE RAM FLOOR11 GB
GPU TARGET8 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
Q
Qwen 3
Alibaba Cloud · 8B
Officially supported35/100 evidence coverage
Qwen 3 8B is a curated dense profile for general, coding, reasoning workloads.
MODEL FILE~5.2 GB
SAFE RAM FLOOR12 GB
GPU TARGET8 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
G
Granite 3.2
IBM · 8B
Officially supported35/100 evidence coverage
Granite 3.2 8B is a curated dense profile for general, reasoning, tools workloads.
MODEL FILE~5 GB
SAFE RAM FLOOR11 GB
GPU TARGET8 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
Q
Qwen 2.5
Alibaba Cloud · 7B
Officially supported35/100 evidence coverage
Qwen 2.5 7B is a curated dense profile for general, writing, coding workloads.
MODEL FILE~4.7 GB
SAFE RAM FLOOR11 GB
GPU TARGET7 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
Q
Qwen 2.5 VL
Alibaba Cloud · 7B
Officially supported35/100 evidence coverage
Qwen 2.5 VL 7B is a curated dense profile for vision, general, research workloads.
MODEL FILE~6 GB
SAFE RAM FLOOR12 GB
GPU TARGET9 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
M
Mistral
Mistral AI · 7B
Officially supported35/100 evidence coverage
Mistral 7B is a curated dense profile for general, writing, tools workloads.
MODEL FILE~4.1 GB
SAFE RAM FLOOR11 GB
GPU TARGET7 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
Q
Qwen 3.5
Alibaba Cloud · 4B
Officially supported35/100 evidence coverage
Qwen 3.5 4B is a curated dense profile for general, coding, reasoning workloads.
MODEL FILE~3.4 GB
SAFE RAM FLOOR10 GB
GPU TARGET6 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
Q
Qwen 3
Alibaba Cloud · 4B
Officially supported35/100 evidence coverage
Qwen 3 4B is a curated dense profile for general, coding, reasoning workloads.
MODEL FILE~2.5 GB
SAFE RAM FLOOR9 GB
GPU TARGET5 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
G
Gemma 4
Google · E4B QAT
Officially supported35/100 evidence coverage
Gemma 4 E4B QAT is a curated dense profile for general, writing, reasoning workloads.
MODEL FILE~6.1 GB
SAFE RAM FLOOR13 GB
GPU TARGET8 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
N
Nemotron 3 Nano
NVIDIA · 4B
Officially supported35/100 evidence coverage
Nemotron 3 Nano 4B is a curated dense profile for general, reasoning, tools workloads.
MODEL FILE~2.8 GB
SAFE RAM FLOOR9 GB
GPU TARGET6 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
G
Gemma 3
Google · 4B
Officially supported35/100 evidence coverage
Gemma 3 4B is a curated dense profile for general, writing, reasoning workloads.
MODEL FILE~3.3 GB
SAFE RAM FLOOR10 GB
GPU TARGET6 GB
ARCHITECTUREDense
The task and system-memory plan fit. Speed still needs a test on this computer.
Showing 12 of 98 matches
Need help or more planning tools?Free reviews, setup support, saved shortcuts and DeviceTerra
READY FOR HUMAN REVIEW?
Get a checked setup before you spend time downloading.
During the public launch, Deviceterra can check the selected model, artifact, engine and hardware at no cost.
OPTIONAL PLANNING TOOLS
Open all free tools →Go further after you have a model and engine.
Estimate memory, compare candidates or plan a first deployment. No sign-up and no hardware information leaves your browser.
KEEP YOUR LOCAL AI SHORTCUT
Come back when your models change.
Save your result on this device, install MamiLens for quick access, or receive occasional updates when new hardware guides and model profiles are published.
D
THE DEVICETERRA MISSION
Use technology today. Compete globally tomorrow.
Deviceterra helps small businesses and institutions in emerging countries increase profit using technology today, and build for the global market tomorrow. MamiLens delivers one part of that mission: practical, private and affordable AI adoption.
06 · LOCAL AI ANSWERS
Clear answers before you download.
Short, practical guidance for the questions people ask when starting with local AI. Each answer links to a complete, evidence-reviewed guide.
Can 8GB of RAM run local AI?
Yes. An 8GB computer can test small, quantized local models, usually in the 1B to 4B range. Close memory-heavy apps, keep context short and expect CPU generation to be slower than GPU inference.
Read the complete guide →Do I need a GPU for local AI?
No. A dedicated GPU improves speed, but compact models can run on a CPU with system RAM. The useful test is whether the model completes your real task at an acceptable quality and waiting time.
Read the complete guide →Ollama or LM Studio: which is better?
LM Studio is a strong visual starting point. Ollama is often better for command-line workflows, automation and applications that need a local API. The right choice depends on how you plan to use the model.
Read the complete guide →Is local AI completely private?
Only when the complete workflow stays local. A local model can still leak data through cloud transcription, web search, plugins, remote storage or exposed APIs. Audit every component, not only the model.
Read the complete guide →Which local AI model should I download?
Choose the smallest proven model that fits your available memory and passes a test for your exact job. Do not choose by parameter count or popularity alone.
Read the complete guide →How much RAM does a local AI model need?
The model file, context cache, runtime, operating system and other applications all consume memory. Leave headroom above the download size and benchmark the exact quantization on your machine.
Read the complete guide →Services and business deploymentLaunch access, configuration review and installation help
PUBLIC LAUNCH ACCESS
Use MamiLens freely while we learn with the community.
The planning tools and configuration review are public during this launch phase. Clear paid services may come later, after MamiLens has earned trust and demonstrated consistent value.
Free
Plan what to test first
- 98 curated profiles
- Conservative hardware matching
- Exact verified Ollama commands
- Official evidence links
Free during public launch
A written check of your selected setup
- Model and artifact verification
- Engine and hardware review
- Risk and limitation notes
- One safer alternative when needed
Request help
Tell Deviceterra where your setup is blocked
- Engine and model troubleshooting
- Safe first-task guidance
- Clear next steps
- No purchase obligation
DEVICETERRA PROJECT SOVEREIGNTurn a compatible model
Turn a compatible model
into a working system.
The model is only one component. Deviceterra installs the runtime, secures the data path, connects private knowledge and validates the workflow.
- Worldwide remote installation
- Private knowledge assistants
- Local coding environments
- School and SMB deployments
REVENUE PATH
One assessment. A clear outcome.
We first confirm that your hardware and use case are suitable. If local AI is the wrong answer, we say so before proposing a deployment.
Current recommended starting pointComplete the setup questions32 GB RAM · General · Windows
MAMILENS KNOWLEDGE BASE
Explore all guides →Learn before you install.
Complete, evidence-first guides live here on MamiLens. Each one turns a technical decision into a practical next step.
START HERE · 8 MIN
What is local AI - and when should you use it?
Understand privacy, offline use, hardware requirements and where cloud AI still makes more sense.
Read complete guide →HARDWARE · 10 MINRAM, VRAM and model size explained simply
Learn why a model file fitting on disk does not automatically mean it will run well in memory.
Read complete guide →SETUP · 12 MINOllama vs LM Studio: which should you install?
Choose the right local AI runtime for beginners, developers and business deployments.
Read complete guide →MODELS · 11 MINQwen, Gemma, Llama or DeepSeek?
A practical family-level comparison for coding, reasoning, writing, vision and general work.
Read complete guide →BUSINESS · 10 MINHow to keep company knowledge local
Plan a private document assistant without handing sensitive files to an unknown cloud service.
Read complete guide →SAFETY · 9 MINWhy “it runs” does not mean “it is reliable”
Test output quality, speed and workflow fit before trusting a local model with important work.
Read complete guide →LOCAL CONTENT · GLOBAL AUTHORITY
MamiLens owns the full guides and internal search journey. Relevant articles link to Deviceterra for the wider technology and deployment perspective.
EVIDENCE POLICY
What “works” means here.
A profile passes only when weights, runtime and selected context fit conservatively. Responsiveness still requires a benchmark on the actual machine.
MamiLens labels full-VRAM fits, partial offload and CPU/RAM execution separately instead of treating every runnable model as equal.
Combined VRAM is useful only when the runtime supports the model, quantization and sharding topology. Interconnect and offload strategy matter.
MoE models load all weights but activate only some experts per token. Active parameters can improve compute efficiency without reducing storage.
A published maximum is not a promise that your hardware can use it. The engine adds more headroom as the selected context grows.
Official pages establish specifications. Community and creator tests inform practical fit, but MamiLens never copies speed results to untested hardware.