Use cases
From documents
to everyday work.
Local AI for professionals, businesses and personal projects: six practical workflows, with starting data, example requests and checks on the result.
Illustrative scenarios, not customer testimonials or guaranteed outcomes. Choose features and models according to your data, permissions and hardware.
Professional services
Compare confidential documents with local AI
A consultancy needs to prepare a proposal from a specification, supplier offers and internal notes. The challenge is not producing more text: it is finding differences, missing requirements and questions without scattering documents across services.
What to prepare
Current document versions, comparison criteria, intended audience and proposal objectives. Import only files you are authorised to process.
Working with T-Chat
- Create a dedicated project and attach the relevant documents, keeping different clients’ files separate.
- Request a requirement-by-requirement comparison with source references and a list of missing information.
- Review the text in the Workspace and prepare a draft. Sending and sharing remain separate decisions.
A request to start with
Compare these two offers against the specification. For each requirement, identify the source, differences and questions for the supplier. Do not send anything.
What to verify
A comparison matrix and a draft grounded in the selected documents. Check references, amounts, dates and unsupported claims before using it.
Keep in mind. AI does not replace professional review. An offline workflow requires models and tools already available on the Mac; email and web research need an authorised connection.
Small businesses and teams
A business Second Brain for procedures and handovers
A team keeps procedures, manuals and decisions in different folders. When a key person changes roles, knowledge becomes hard to find. A business Brain gathers selected sources for questions without mixing them with personal archives.
What to prepare
Approved procedures, internal manuals, documented decisions and named content owners. Define who can access the archive and who maintains it.
Working with T-Chat
- Create a business Brain and explicitly import files or folders, checking the source preview.
- Connect it to authorised projects and ask for referenced answers that distinguish current procedures from information requiring review.
- Update sources, curate confirmed memories and configure backup retention. Test recovery into a separate archive.
A request to start with
What is our procedure for taking on a new client project? Cite the Brain documents and flag any steps without a source.
What to verify
A searchable knowledge base for onboarding and operational continuity. Verify that every answer uses the correct version of each procedure.
Keep in mind. Not every stored format can be indexed automatically. Backups need configuration and safe custody; they do not replace an organisational continuity plan.
Personal knowledge
Build your Second Brain with memories, writing and photographs
Notes, experiences and choices capture different parts of a person. A personal archive can bring them together and help you find connections without turning AI interpretations into facts about your life.
What to prepare
Writing, documents and memories you want to preserve. In the 4.5 preview, add values, preferences and goals to your profile and explicitly choose whether answers should use them.
Working with T-Chat
- Choose your Brain or create several for separate areas, without automatically merging their contents.
- In the 4.5 preview, request local analysis of a photo or PDF pages, review the description and add the people, context and memories yourself.
- Confirm what should become searchable and keep encrypted archive copies with a recovery procedure.
A request to start with
Help me reconstruct the criteria I used in work decisions. Use sources from my Brain and distinguish what I wrote from your interpretations.
What to verify
A documented personal memory you can correct and consult. The profile retains revisions, while originals and personal context remain separate.
Keep in mind. Profiles and visual analysis are 4.5 preview features, not part of the 4.4 installers. AI is not a complete replica of a person and cannot infer certain identities or memories from a photograph.
Sales and planning
Estimate sales trends from historical data
A business wants to plan purchasing and capacity for the coming months. TimesFM 3 can produce local time-series forecasts; the usefulness of the work also depends on data quality and comparison against known outcomes.
What to prepare
A regularly spaced series with consistent units and identified periods. Note closures, promotions and changes that make historical periods less comparable.
Working with T-Chat
- Activate TimesFM 3 and enable forecasting for the personality you will use in chat.
- Hold back the latest known periods and request a forecast using earlier data only. Compare errors against a simple baseline, such as the same month a year earlier.
- If the comparison is useful, generate a future forecast and request quantiles where available. Review it against operating constraints.
A request to start with
Use TimesFM on the first 36 months of this series and forecast the next 12. Do not include the last 12 as input. Show the tool output; then let’s compare it with actual data.
What to verify
A forecast accompanied by its input data and a historical check. Not a sales target or certainty about the future.
Keep in mind. A good result on one dataset does not guarantee accuracy elsewhere. Estimates are not investment recommendations and do not automatically anticipate shocks or market changes.
Operations and development
Create and update an internal app without losing its project
An operations manager needs a small register for requests or supplies. A first version is only the beginning: organised files, tests and a quick way back to editing matter when the process changes.
What to prepare
Required fields, roles, validation rules and a few fictional examples. Avoid real personal data or credentials in the prototype.
Working with T-Chat
- Open a project and describe the workflow. Choose the autonomy level and ask the agent to clarify decisions that would change the result.
- Review code and preview in the Workspace. Test valid data, missing fields and failure cases before distributing the app.
- Use “Edit app” in Applications to return to the linked project and conversation. Review changes before publishing again.
A request to start with
Create a supplies-register prototype with quantities and minimum levels. First ask me how to handle users; use fictional data and do not publish without confirmation.
What to verify
An inspectable prototype and a project you can reopen, with files and context preserved. Authentication, data and deployment need specific checks.
Keep in mind. Generated code needs testing and review. A preview does not certify security or production readiness; the built-in XLSX editor does not replace every Excel feature.
Research and prototyping
Prepare an Arduino prototype with HUMANOID
A lab wants a bench-top environmental sensor with a status light. HUMANOID organises identification, manuals, code and presets; Companion keeps the device-upload stage separate.
What to prepare
The exact board model, components, power supply, wiring diagram and manufacturer documentation. A USB connection cannot reveal the entire circuit.
Working with T-Chat
- Use Companion to detect selected USB devices and associate the device with a HUMANOID project.
- Attach manuals or authorise official sources. Develop the code in the project and prepare a test procedure.
- Use a compatible toolchain and verify board, port and firmware before confirming the upload. Test the prototype and retain presets and results.
A request to start with
Let’s prepare a bench-top environmental sensor using this board and manual. Ask about missing wiring details and propose safe tests before any upload.
What to verify
A versioned, reusable hardware project with source code, documentation and test results. Successful upload remains distinct from physical operation.
Keep in mind. HUMANOID is an experimental lab: device support and toolchains must be verified. Universal compatibility, electrical certification and robotic autonomy are not implied.
Start with a real process.
Choose an activity, prepare a non-sensitive data sample and let’s define the expected outcome, hardware and checks together.