Run an open model on your PC.
Some AI models are available to download and run on your own computer. People often call these “open-source” or “OSS” models. Many are more precisely “open-weight”: their learned model files are available, while the permissions to use, modify, or redistribute them depend on their license.
Try it now
Copy the prompt into your AI tool of choice, then answer its questions as they come.
Paste it into any AI chat to start. For most people, this is all you need.
Everything on this page in one paste: the prompt plus the full plan, every step and what to check. Use it for longer moves, to plan ahead, to share with a colleague, or if your AI can't open links.
How it goes
What the AI will do
- Research current downloadable models and what your hardware can run.
- Compare suitable computer options around $1,500, $3,000 and $6,000, adjusting to your budget.
- Help install and test your chosen local model, with an optional small fine-tuning experiment.
What to have ready
- Your interest or use case; a general description is enough.
- Your existing computer or help finding its specifications, and any budget or equipment preferences.
Where you come in
- Choose an approach and confirm setup.
- Try the model and judge its usefulness and speed.
- If you want to fine-tune, choose a small local training experiment and compare the results.
What you'll get
A working local model, an observed trial result, and instructions for using it again; an evaluated adaptation too if you chose fine-tuning.
You're done when
A model runs on your own computer, you have tried it on a task that interests you, and you know how to use it again. If private local processing was your purpose, you have tested the selected workflow offline and established how to keep its content local. If you chose a fine-tuning experiment, you also have an adapted model or adapter and a comparison with the original on separate examples.
These are general instructions, not a description of this user's particular AI system, tools, access, or circumstances. Adapt the method to the user's goal and the capabilities you can actually use, while preserving the activity's essential sequence, review points, and intended result. ### Start with the activity and the user's context Read the complete activity and guide the user through it. Use the activity prompt yourself; the user does not need to paste it again. Keep the responsibilities of the user, the AI, and any colleague clear. Begin with what the user has already supplied. Establish the intended result and resolve missing context through focused questions when needed. Work within the activity's stated scope, permissions, required confirmations, and stopping points, including any deliberate order for introducing information. Make progress with the available information and reasonable, clearly identified assumptions where the activity allows them. ### Check the capabilities needed for this work Inspect relevant available tools, connectors, and data. Use small, relevant, non-destructive checks to establish whether necessary access or a capability actually works. Naming a product does not establish access to it. Let this activity determine what needs checking; do not impose a universal preliminary questionnaire or a full system audit. When useful access is missing: - Explain what you cannot reach and how that material would help. - Investigate practical setup, connection, export, or other access alternatives before asking the user to gather material or troubleshoot. - If progress depends on access, request only the specific user action or permission needed to enable it or provide the material another way. Continue useful independent work while awaiting the response. - Offer to walk the user through an export or upload. Welcome files, links, copied text, rough notes, or an interview as supported by your system and the activity, and use the best available route. If the missing access is optional, continue with what is available. ### Ask focused questions and respect the answers Ask questions only when necessary to complete the work well or satisfy the activity's required review points. Build on answers already supplied. Ask one question at a time, or a small group of closely related questions. Wait on work that depends on the answer, and continue useful work that does not. Explain unfamiliar language with examples where helpful. Confirm answers or decisions you supply on the user's behalf before treating them as the user's final answers or decisions. This does not require separate approval of every sourced finding; follow the activity's specific review requirements. ### Keep evidence and independent contributions distinct Distinguish evidence, inference, and missing information. Check your work against the source material. Use independent agents when your system actually supports delegation and independent investigation or review would improve the work: - Give each agent a bounded brief and the context appropriate to its role. - Keep discovery independent until synthesis where the activity requires it. Preserve any requirement for the user or colleagues to contribute before seeing other explanations. - Follow the activity's stated fallback and any required confirmation when independent agents or fresh contexts are unavailable. Preserve reviewer isolation where required; describe shared-context or sequential passes accurately. ### Produce, check, and revise the result Do the substantive investigation and production. Carry the activity through to the most complete, usable result within its scope, rather than ending with a plan or handing the main work back to the user. When missing information does not prevent useful work, proceed with clearly identified assumptions where the activity permits them. Check the result against the activity's completion criteria. At each substantive delivery, state what is complete and any material unfinished work, assumptions, helpful missing information, and remaining uncertainty. Explain how these affect what the user can use, and help the user revise the result. Preserve the activity's stopping point. Do not turn a rehearsal, investigation, or test-design activity into execution that the user has not authorized. ### Help me explore and run a model on my own computer Help me choose and run a downloadable AI language model locally. I am comfortable using AI but may know little about computer hardware or model training. Explain technical terms when they help me make a decision, and do the research and supported setup yourself. ### Start with what interests me Ask me what I want to try. Offer these starting points, and welcome my own description: - Just curious to play around. - Want to process sensitive data locally. - Want to train my own model. - Other. Wait for my answer. Then ask focused questions one at a time to establish the task, what a useful result would look like, and whether I want to use my existing computer or consider buying one. Identify my operating system, CPU, RAM, GPU and its dedicated memory, and available storage. Inspect these details if you have appropriate access; otherwise help me find them. A computer's name or price alone is not enough to determine what it can run. Establish my country/currency and whether I want a laptop, desktop, upgrade, or have no preference. Use US dollars and the budget bands below as starting points when I have not specified a budget. If they are clearly inappropriate, ask what range makes sense and adapt before researching an extensive shopping list. Do not ask me to paste sensitive documents, private training data, or confidential examples into this research chat; a general description or synthetic example is enough. ### Research current models and what they need Search the web now. Start with model creators' official releases, model cards, licenses, and repositories, then the current documentation for compatible local runners such as LM Studio, Ollama, or an appropriate alternative. Useful starting points include https://lmstudio.ai/docs/app/system-requirements and https://docs.ollama.com/faq. Verify the exact current versions; do not make recommendations from memory alone. Find a small shortlist matched to my purpose. Explain open-source versus open-weight in plain language and check the actual license for my intended use. For each candidate, report its strengths, relevant limitations, available download/version, supported local software, and source links. Compare its capabilities with current hosted frontier models where useful, using task-relevant evidence. Do not assume downloadable models are always far behind, or that a small local model matches every capability of a larger hosted system. Give both minimum requirements and a configuration likely to feel comfortable for my task: CPU/GPU compatibility, system RAM, dedicated GPU memory or Apple unified memory, disk space, and operating system. Name the exact model size and precision or quantized version behind each estimate. Explain quantization as making model files smaller using lower-precision numbers, which can change quality. Account for the length of the material I want to process and memory used beyond the model files. Distinguish running a model from training or fine-tuning it; training requirements must be researched separately. Investigate NVIDIA GPU options and explain the practical significance of GPU memory. Compare Apple Silicon or CPU-only alternatives where they suit my needs. Assess whether my existing computer can run a useful smaller model before suggesting a purchase. For speed claims, cite measurements for the relevant model and hardware where available, explain the test conditions, and label estimates clearly. Never invent benchmark results or promise that merely fitting a model in memory makes it fast. If my goal is fine-tuning, establish what I want the model to learn and what examples I could provide locally. Explain how adapting behavior with training differs from supplying instructions or searching documents at answer time. Research a feasible small learning experiment, including a compatible model, training method such as LoRA, software, data format, training-memory requirements, and how we would compare the result with the original model on separate examples. Let me choose whether to pursue that experiment; do not silently replace my learning goal with document search. For a work use case, search relevant accessible company software catalogs and AI/IT guidance for permitted local runners, equipment, data handling, and installation routes. Use those findings to select a workable setup, without searching unrelated correspondence or processing the sensitive data itself. If a necessary requirement cannot be established, identify the specific missing information and help me obtain it. A personal learning experiment does not need an organizational investigation. ### Show what different budgets buy Unless I have given a different range, research a few complete computer configurations in each of these bands: - Around $1,500, give or take $500. - Around $3,000. - Around $6,000. Aim for two genuinely suitable options per band where the market supports them. Do not manufacture a second option or include unsuitable machines just to fill a table. If one band cannot meet my requirements, explain what it could do and the specific constraint. Include an existing-computer or upgrade route when useful, separate from complete-system prices. Use current manufacturer specifications and direct seller listings for the exact configuration and local availability. Distinguish new, refurbished, and used equipment. Give a checked date, currency, purchase links, and whether tax, shipping, or required extras are excluded. A GPU-only price must not be presented as a complete computer price. Use a readable comparison table with exact CPU/GPU, GPU memory or unified memory, RAM, storage, total price, compatible shortlisted models, and what I could reasonably do with each. Keep running and fine-tuning suitability separate. Explain what spending more would buy for my purpose and where it might make little difference. Treat performance estimates as estimates, and reconcile conflicting specifications before recommending a machine. If the research reveals a substantial budget mismatch, ask me how to adjust the scope or budget and wait. Recommend a starting model and setup, with the reason it fits my interests. Ask whether I want to proceed, revise the options, or keep exploring. Do not buy anything. A researched shortlist is an intermediate result; actually running a model is the activity's completion point. ### Get it running locally and try it Once I choose an available setup and confirm installation, follow current official instructions for that exact operating system and hardware. Help download the local runner and model from their original sources, configure the model, and run a small non-sensitive trial matched to my interest. Perform the actions you can; guide me through actions requiring my access one at a time. If this chat cannot operate my computer, provide a complete handoff to a suitable coding assistant or an accessible guided setup, carrying forward the selected model, hardware, sources, purpose, and choices. For the local workflow, verify that inference, document processing or embeddings if used, and any other required components run on this computer. Check the actual application's cloud models, fallback, web tools, plugins, and telemetry settings against current documentation. Explain what sends data out and configure the selected workflow to keep its content local. Download what is needed first, then test it with networking disabled using non-sensitive material. An offline trial demonstrates that the tested workflow works without a server; do not treat that as proof about every mode of the application when networking is restored. For sensitive work, keep the tested workflow offline or establish the network controls needed to preserve that boundary before using the real data. Use https://lmstudio.ai/docs/app/offline and the selected runner's current documentation as starting points. Evaluate the trial together: usefulness, mistakes, responsiveness, and whether the computer remains comfortable to use. Investigate and fix setup problems, or try a better-fitting model within the agreed scope. If I chose the fine-tuning experiment, obtain my agreement on the small dataset and training run, perform it locally or guide the necessary local actions, and compare the adapted model with the original using held-out examples. Keep raw data, training output, and evaluation local when that is my requirement. Do not start a substantial training run without agreeing its expected time and resources with me. Finish with a model that runs on my computer, the results of our trial, and short instructions for reopening it and repeating the task. Include the exact model/version, configuration, and any remaining limitations. If I also chose fine-tuning, include the saved adapted model or adapter and the comparison results. If hardware, access, or installation is still missing, leave a specific resumption handoff and say that running the model remains unfinished. These instructions were written in September 2026 using the frontier models available at the time. You may now have better tools or models, or restrictions that make some tasks harder. Your AI should adapt to the tools and access you actually have and find the best available way to help. Everything here is a starting suggestion for you and your AI, not a proven best practice for every setup. Feel free to rewrite any of it to better meet your needs.
View full AI instructions
Run an open model on your PC.
Some AI models are available to download and run on your own computer. People often call these “open-source” or “OSS” models. Many are more precisely “open-weight”: their learned model files are available, while the permissions to use, modify, or redistribute them depend on their license.
Running locally can let you work with sensitive information entirely on your own device, even without an internet connection once everything is downloaded. The setup matters: the model, document processing, and any connected tools need to stay local. The AI guiding this activity will help you check that.
With a suitable model and hardware, local AI can feel surprisingly fast. Smaller models make useful experiments and practical tasks possible on consumer computers. What will run well depends on memory and processing power. An NVIDIA graphics card with enough dedicated memory is an important option to investigate; Apple Silicon and some CPU-only setups are possibilities too. Your existing computer may already be enough to start.
Choosing a smaller model involves tradeoffs in capability, speed, and how much material it can handle. Try it on the work you care about. You can also learn by adapting an existing model with your own training examples, a process called fine-tuning. That is different from simply letting it search your documents.
Models and hardware change quickly. Use the prompt below to find current options for your interests and budget, then get a model running.
For the user
Tell the AI what interests you: playing with a local model, processing sensitive information on your own computer, learning to train a model, or something else. It will help you identify your computer’s specifications, research current models and equipment, and compare what different budgets buy. You will choose a setup, confirm installation, and try the result. Describe sensitive work in general terms while choosing the setup.
Done when
A model runs on your own computer, you have tried it on a task that interests you, and you know how to use it again. If private local processing was your purpose, you have tested the selected workflow offline and established how to keep its content local. If you chose a fine-tuning experiment, you also have an adapted model or adapter and a comparison with the original on separate examples.
Guidance for the AI
These are general instructions, not a description of this user's particular AI system, tools, access, or circumstances. Adapt the method to the user's goal and the capabilities you can actually use, while preserving the activity's essential sequence, review points, and intended result.
Start with the activity and the user's context
Read the complete activity and guide the user through it. Use the activity prompt yourself; the user does not need to paste it again. Keep the responsibilities of the user, the AI, and any colleague clear.
Begin with what the user has already supplied. Establish the intended result and resolve missing context through focused questions when needed. Work within the activity's stated scope, permissions, required confirmations, and stopping points, including any deliberate order for introducing information. Make progress with the available information and reasonable, clearly identified assumptions where the activity allows them.
Check the capabilities needed for this work
Inspect relevant available tools, connectors, and data. Use small, relevant, non-destructive checks to establish whether necessary access or a capability actually works. Naming a product does not establish access to it. Let this activity determine what needs checking; do not impose a universal preliminary questionnaire or a full system audit.
When useful access is missing:
- Explain what you cannot reach and how that material would help.
- Investigate practical setup, connection, export, or other access alternatives before asking the user to gather material or troubleshoot.
- If progress depends on access, request only the specific user action or permission needed to enable it or provide the material another way. Continue useful independent work while awaiting the response.
- Offer to walk the user through an export or upload. Welcome files, links, copied text, rough notes, or an interview as supported by your system and the activity, and use the best available route. If the missing access is optional, continue with what is available.
Ask focused questions and respect the answers
Ask questions only when necessary to complete the work well or satisfy the activity's required review points. Build on answers already supplied. Ask one question at a time, or a small group of closely related questions. Wait on work that depends on the answer, and continue useful work that does not. Explain unfamiliar language with examples where helpful.
Confirm answers or decisions you supply on the user's behalf before treating them as the user's final answers or decisions. This does not require separate approval of every sourced finding; follow the activity's specific review requirements.
Keep evidence and independent contributions distinct
Distinguish evidence, inference, and missing information. Check your work against the source material.
Use independent agents when your system actually supports delegation and independent investigation or review would improve the work:
- Give each agent a bounded brief and the context appropriate to its role.
- Keep discovery independent until synthesis where the activity requires it. Preserve any requirement for the user or colleagues to contribute before seeing other explanations.
- Follow the activity's stated fallback and any required confirmation when independent agents or fresh contexts are unavailable. Preserve reviewer isolation where required; describe shared-context or sequential passes accurately.
Produce, check, and revise the result
Do the substantive investigation and production. Carry the activity through to the most complete, usable result within its scope, rather than ending with a plan or handing the main work back to the user. When missing information does not prevent useful work, proceed with clearly identified assumptions where the activity permits them.
Check the result against the activity's completion criteria. At each substantive delivery, state what is complete and any material unfinished work, assumptions, helpful missing information, and remaining uncertainty. Explain how these affect what the user can use, and help the user revise the result.
Preserve the activity's stopping point. Do not turn a rehearsal, investigation, or test-design activity into execution that the user has not authorized.
Your AI prompt
Help me explore and run a model on my own computer
Help me choose and run a downloadable AI language model locally. I am comfortable using AI but may know little about computer hardware or model training. Explain technical terms when they help me make a decision, and do the research and supported setup yourself.
Start with what interests me
Ask me what I want to try. Offer these starting points, and welcome my own description:
- Just curious to play around.
- Want to process sensitive data locally.
- Want to train my own model.
- Other.
Wait for my answer. Then ask focused questions one at a time to establish the task, what a useful result would look like, and whether I want to use my existing computer or consider buying one. Identify my operating system, CPU, RAM, GPU and its dedicated memory, and available storage. Inspect these details if you have appropriate access; otherwise help me find them. A computer's name or price alone is not enough to determine what it can run.
Establish my country/currency and whether I want a laptop, desktop, upgrade, or have no preference. Use US dollars and the budget bands below as starting points when I have not specified a budget. If they are clearly inappropriate, ask what range makes sense and adapt before researching an extensive shopping list. Do not ask me to paste sensitive documents, private training data, or confidential examples into this research chat; a general description or synthetic example is enough.
Research current models and what they need
Search the web now. Start with model creators' official releases, model cards, licenses, and repositories, then the current documentation for compatible local runners such as LM Studio, Ollama, or an appropriate alternative. Useful starting points include https://lmstudio.ai/docs/app/system-requirements and https://docs.ollama.com/faq. Verify the exact current versions; do not make recommendations from memory alone.
Find a small shortlist matched to my purpose. Explain open-source versus open-weight in plain language and check the actual license for my intended use. For each candidate, report its strengths, relevant limitations, available download/version, supported local software, and source links. Compare its capabilities with current hosted frontier models where useful, using task-relevant evidence. Do not assume downloadable models are always far behind, or that a small local model matches every capability of a larger hosted system.
Give both minimum requirements and a configuration likely to feel comfortable for my task: CPU/GPU compatibility, system RAM, dedicated GPU memory or Apple unified memory, disk space, and operating system. Name the exact model size and precision or quantized version behind each estimate. Explain quantization as making model files smaller using lower-precision numbers, which can change quality. Account for the length of the material I want to process and memory used beyond the model files. Distinguish running a model from training or fine-tuning it; training requirements must be researched separately.
Investigate NVIDIA GPU options and explain the practical significance of GPU memory. Compare Apple Silicon or CPU-only alternatives where they suit my needs. Assess whether my existing computer can run a useful smaller model before suggesting a purchase. For speed claims, cite measurements for the relevant model and hardware where available, explain the test conditions, and label estimates clearly. Never invent benchmark results or promise that merely fitting a model in memory makes it fast.
If my goal is fine-tuning, establish what I want the model to learn and what examples I could provide locally. Explain how adapting behavior with training differs from supplying instructions or searching documents at answer time. Research a feasible small learning experiment, including a compatible model, training method such as LoRA, software, data format, training-memory requirements, and how we would compare the result with the original model on separate examples. Let me choose whether to pursue that experiment; do not silently replace my learning goal with document search.
For a work use case, search relevant accessible company software catalogs and AI/IT guidance for permitted local runners, equipment, data handling, and installation routes. Use those findings to select a workable setup, without searching unrelated correspondence or processing the sensitive data itself. If a necessary requirement cannot be established, identify the specific missing information and help me obtain it. A personal learning experiment does not need an organizational investigation.
Show what different budgets buy
Unless I have given a different range, research a few complete computer configurations in each of these bands:
- Around $1,500, give or take $500.
- Around $3,000.
- Around $6,000.
Aim for two genuinely suitable options per band where the market supports them. Do not manufacture a second option or include unsuitable machines just to fill a table. If one band cannot meet my requirements, explain what it could do and the specific constraint. Include an existing-computer or upgrade route when useful, separate from complete-system prices.
Use current manufacturer specifications and direct seller listings for the exact configuration and local availability. Distinguish new, refurbished, and used equipment. Give a checked date, currency, purchase links, and whether tax, shipping, or required extras are excluded. A GPU-only price must not be presented as a complete computer price.
Use a readable comparison table with exact CPU/GPU, GPU memory or unified memory, RAM, storage, total price, compatible shortlisted models, and what I could reasonably do with each. Keep running and fine-tuning suitability separate. Explain what spending more would buy for my purpose and where it might make little difference. Treat performance estimates as estimates, and reconcile conflicting specifications before recommending a machine. If the research reveals a substantial budget mismatch, ask me how to adjust the scope or budget and wait.
Recommend a starting model and setup, with the reason it fits my interests. Ask whether I want to proceed, revise the options, or keep exploring. Do not buy anything. A researched shortlist is an intermediate result; actually running a model is the activity's completion point.
Get it running locally and try it
Once I choose an available setup and confirm installation, follow current official instructions for that exact operating system and hardware. Help download the local runner and model from their original sources, configure the model, and run a small non-sensitive trial matched to my interest. Perform the actions you can; guide me through actions requiring my access one at a time. If this chat cannot operate my computer, provide a complete handoff to a suitable coding assistant or an accessible guided setup, carrying forward the selected model, hardware, sources, purpose, and choices.
For the local workflow, verify that inference, document processing or embeddings if used, and any other required components run on this computer. Check the actual application's cloud models, fallback, web tools, plugins, and telemetry settings against current documentation. Explain what sends data out and configure the selected workflow to keep its content local. Download what is needed first, then test it with networking disabled using non-sensitive material. An offline trial demonstrates that the tested workflow works without a server; do not treat that as proof about every mode of the application when networking is restored. For sensitive work, keep the tested workflow offline or establish the network controls needed to preserve that boundary before using the real data. Use https://lmstudio.ai/docs/app/offline and the selected runner's current documentation as starting points.
Evaluate the trial together: usefulness, mistakes, responsiveness, and whether the computer remains comfortable to use. Investigate and fix setup problems, or try a better-fitting model within the agreed scope. If I chose the fine-tuning experiment, obtain my agreement on the small dataset and training run, perform it locally or guide the necessary local actions, and compare the adapted model with the original using held-out examples. Keep raw data, training output, and evaluation local when that is my requirement. Do not start a substantial training run without agreeing its expected time and resources with me.
Finish with a model that runs on my computer, the results of our trial, and short instructions for reopening it and repeating the task. Include the exact model/version, configuration, and any remaining limitations. If I also chose fine-tuning, include the saved adapted model or adapter and the comparison results. If hardware, access, or installation is still missing, leave a specific resumption handoff and say that running the model remains unfinished.
These instructions were written in September 2026 using the frontier models available at the time. You may now have better tools or models, or restrictions that make some tasks harder. Your AI should adapt to the tools and access you actually have and find the best available way to help. Everything here is a starting suggestion for you and your AI, not a proven best practice for every setup. Feel free to rewrite any of it to better meet your needs.
These instructions were written in September 2026 using the frontier models available at the time. You may now have better tools or models, or restrictions that make some tasks harder. Your AI should adapt to the tools and access you actually have and find the best available way to help. Everything here is a starting suggestion for you and your AI, not a proven best practice for every setup. Feel free to rewrite any of it to better meet your needs.