Provide a seven-day workbook with truthful completed and open items.
You will complete a seven-day plan that inventories hardware, chooses processing, runs a first model, evaluates output, and records a safe workflow.
Before you begin
Choose a low-risk goal, identify whether its data may leave your device, and reserve time for official installation and privacy documentation.
Inventory checklist
Record OS and version, CPU, RAM, GPU and dedicated or shared memory, free storage, and connectivity. Classify task data as public, personal, confidential, or regulated. Note ownership, disk encryption, and account protection. For cloud use, identify account-security and spending controls. This prevents choosing a tool before understanding the task, privacy boundary, and operating constraints.
Choose one starting option
For local AI, consult current runtime support and choose a modest instruct model whose format, size, and license you understand. For cloud AI, review current retention, training use, price, quota, and multifactor authentication. Prepare three non-sensitive examples and expected qualities. Avoid installing many runtimes or subscribing to several services at once; one baseline simplifies troubleshooting.
Run and evaluate
Record application, runtime or provider, model, date, and important settings. Ask one clear question, one source-grounded summary, and one request with missing information. Check support, instruction following, uncertainty, speed, and resource or usage impact. The result is a written observation, not proof every task works. Stop if the computer is unstable, cost is unclear, or the data boundary differs from expectations.
Make the workflow repeatable
Save a prompt stating goal, context, constraints, and output format. Add a checklist to remove sensitive data, verify names and numbers, inspect sources, and obtain approval before external action. Keep original input and reviewed output separate. Add one capability at a time and retest after updates. You should finish with one useful task, one working configuration, known limits, and a safe stop or removal path.
Days 1–3: decide and establish a baseline
On day one, write the task and success check. On day two, complete the hardware and data inventory and choose local, cloud, or hybrid processing. On day three, follow official setup guidance and test one model or service. Record versions and run the same three non-sensitive examples. Do not optimize yet. The baseline should answer: does it work, is the data boundary acceptable, is the cost or resource use visible, and can you stop or remove it safely?
Days 4–7: improve one factor at a time
On day four, improve prompt structure. On day five, add source or output validation. On day six, test an empty, ambiguous, and adversarial input. On day seven, document the workflow, known limitations, and review gate. Compare results with the baseline. If you try another model, keep all other factors stable. The goal is not maximum automation in a week; it is one useful, understood process that fails safely and can be explained to another beginner.
Pick one starter stack
Decision: Start with one local runtime and model or one permitted hosted product; avoid gateways and agents in week one.
Relevant options: Ollama offers CLI/API local use, LM Studio a desktop local path, and ChatGPT or Claude hosted use. Subscriptions and APIs are separate unless official terms say otherwise.
Worked example: Run three fixed low-risk summaries and record setup, data boundary, quality, resource or visible usage, and removal path.
Failure mode: Several runtimes or trials make it unclear which process, model, bill, or privacy setting produced the result.
Validation: Confirm the model/provider, repeat examples, review settings and terms, and prove the workflow stops safely.
Useful next step: Keep one baseline for a week and add one capability only for a documented gap.
Official sources checked 2026-10-24: Ollama · LM Studio · ChatGPT · Claude. Product availability, support, terms, pricing, and licenses can change. These are vendor or project sources; no independent benchmark is claimed.
Seven-day journal — completed and open items
Download the workbook. The example records real hardware, Ollama 0.35.1, qwen2.5:1.5b, the task, and observed runs.
Days 1–3 are complete. Days 4–7—repeatability, real-data privacy, update behavior, and removal—are NOT_TESTED.
Authoritative references
Reference scope checked 2026-10-27. These sources support the concepts and risk controls; product-specific behavior still requires the nearby vendor documentation and a dated test.
Your next step
Start with a verifiable low-risk task and leave every unperformed day marked NOT_TESTED.
Editorial provenance
Nyfir Studios is the publisher and editorially responsible organization. AI tools assisted with drafting and structured review for this candidate; factual claims, test labels, links, and retained evidence require human review before deployment. No named employee, credential, or unrecorded hands-on result is implied.