Case study / Personal OSEvidence before claims / 2026
Thomas Howcroft Letchworth / UK
Personal AI operating system

Personal OS: from scattered work to a clear next decision.

How does Personal OS connect daily decisions, specialist work and real completion?

Personal OS connects daily priorities, specialist support and human decisions in one private workspace. Hermes and ongoing agent chats bring together Work and Review, inbox and calendar context, meals, brand production, Finance, Life and bounded automation. Follow each workflow below, from the source material to the proposed action and the evidence needed to call it complete.

StatusWorking internal system
RoleProduct, system and interface design
ScopeFounder operations across desktop and responsive mobile
EvidenceCurrent workspace, shared chats, activity reports and approval behaviour
Known limitDemo records only; a chat role does not imply the tools and permission to perform every task
Evidence in brief

Continuity, current evidence and a visible review boundary.

Read the result, source and limits together.

Current capability

Daily work connects to Hermes, specialist chats, role details, recorded activity and approval cards.

Evidence you can inspect

The current interface with fictional records, including Bots, shared chats, Details and suggested-response behaviour.

Thomas's contribution

Product, system and interface design: the decision model, information hierarchy, shared conversations and review experience.

Limits kept visible

A chat is not proof that every task can run. Recorded activity is not proof of a successful external result. Private records remain outside the tour.

Inspect the operating modules and shared conversations

Keep the current decision in chat and the supporting detail one step away.

The interface separates a useful update from a repeated check, a suggestion from a send and a current result from an unsupported claim. This preserves continuity without making every background event another message.

01 / The operating problem

A daily workspace needs to carry the work through to a proved result.

A direct account of the operating constraint, the product decision and the limits of the evidence.

A personal system becomes useful when it carries context between planning, preparation, decisions and completion. An inbox count alone does not explain what needs a reply. A bot badge does not show what it produced. A completed-looking task does not establish a confirmed real-world result.

Thomas designed Personal OS around those gaps. Work presents the next priorities, Review exposes prepared decisions, and Needs you gathers the actions requiring the owner. Hermes connects the workspace through persistent specialist conversations, while Details exposes current work, dated results, blockers and recorded runs.

The underlying workflows reach beyond chat. They prepare email replies, use read-only calendar context, maintain local meal plans, produce branded content and record bounded automation outcomes. Each module keeps its own evidence and action boundary, so a useful preparation can be inspected without being mistaken for delivery.

System at a glance
Daily decisionsWork prioritises current work, Review holds prepared decisions, and Needs you gathers owner actions.
Conversation modelHermes connects nine featured specialists through ongoing chats and supporting Details.
Practical modulesInbox and calendar context, meal planning, brand production, personal administration and financial planning.
Automation proofChanged work, verified unchanged results and proved blockers have different receipts.
Human reviewExact proposals and email packets expose the decision before an external action.
Public evidenceThe genuine shared interface with fictional records labelled Demo data; device evidence is dated separately.
The full product

Inside Personal OS.

Follow the work through its tools, people, evidence and decisions.

01Working decision surfaces

Work, Review and Needs you

Work answers a practical question: what should I move forward next? It groups current priorities, commitments and project work, then provides a route to the record behind each item. Review holds work that needs checking. Needs you collects decisions that require the owner, so background preparation does not become another personal to-do. The three views serve different moments in the same workflow.

Completion depends on the owning record. A closed decision should leave the active work and review surfaces, while its evidence remains in history. References to the same decision resolve to one item rather than separate copies. Missing evidence stays visible as a gap; a refresh time does not turn an old result into current progress.

02Current shared conversation interface

Hermes and the specialist conversations

Hermes is the main assistant and synthesis layer. It connects permitted workspace context into one grounded answer, brief or next decision. Bots opens ongoing specialist conversations rather than creating a new chat for every dashboard visit. The role card, conversation and Details panel refer to the same specialist, making it possible to ask a follow-up without losing the work being discussed.

Details separates the role's job from its recorded execution. It exposes current work, last result, blockers, recent runs and the next recorded event where available. A model selected for a conversation is distinct from the model recorded for an automated run. LinkedIn Growth appears under Personal Brand and opens that existing chat; it is not an extra independent agent.

03Current review behaviour

Approval, rejection and feedback in context

An approval card carries the proposal the owner is being asked to judge. LinkedIn cards show the action kind, account, lane, target, context and complete text in the conversation. The response or refusal appears beside that card. The eligible desktop review records the exact local decision; the subsequent worker still needs its own authority and checks before any external action.

Suggested responses solve a different problem. Selecting a multiple-choice option places its code in the message box and focuses it, leaving the user able to edit before sending. Email has its own Review > Email surface with the complete sender, recipients, subject, body and attachments. Rejected email revisions remain rejected; an automatic retry cannot quietly turn the same rejected text back into a fresh approval.

04Read-only context with reviewed drafts

Inbox preparation and calendar context

The inbox workflow reads connected sources, identifies reply obligations and prepares complete drafts for Review > Email. Its operations view distinguishes drafts ready, replies due, held items and source coverage. These states help separate a missing reply from an unavailable mailbox or an incomplete scan. The drafting work prepares the decision rather than sending on the owner's behalf.

Calendar context informs the day shape: agenda, commitments and available time can influence the daily view and priority discussion. That calendar feed is read-only. A proposed booking remains a separate action. Mailbox coverage and historical backfill are not uniformly complete, and the proposed Jev private-email classifier remains gated rather than established as active.

05Meal planning has separate local evidence

Health, meal planning and food feedback

Health covers meals, training, sleep, recovery and energy using available evidence. Its meal-planning home includes individual preferences, candidate recipes, served history and an explicit stock check. Feedback can apply to one week or remain ongoing, and a never rating from either household member takes priority. Repeating the same feedback event does not teach the system twice.

Shopping preparation uses whole packs and checked stock, with expiry or expected delivery dates. Stale pantry entries cannot silently remove a needed purchase. Recipe research retains its source and published nutrition; missing values remain unknown, and researched recipes stay candidates until ingredient and shopping checks pass. Meal planning can remain usable when wider phone-derived health inputs are stale. It provides non-diagnostic planning, with no automatic purchase or individual medical claim.

06Working preparation and review pipeline

Personal Brand from evidence to finished media

The brand pipeline connects topic research, source-backed writing, visual production and review. It supports detailed text posts, infographics and multi-page carousels rather than treating every idea as a short text update. Personal-account binding, media previews and companion copy keep the intended channel visible. Thomas's monochrome visual system carries through the finished material.

Personal Brand also prepares qualified LinkedIn growth work and exact action cards in its shared chat. The user can inspect the proposed target, account and complete wording before deciding. The specialist's dated task and event history gives context, while its Details panel states current access limits. A recorded local review is separate from scheduling or publishing, and the case does not claim an always-visible browser session.

07Bounded specialist workflows

Two distinct routes for paid opportunities

Daily Job and Additional Money share the goal of progressing paid opportunities but operate different workflows. Daily Job checks role suitability and provider evidence, prepares relevant application material and records the confirmed submission result. Its standing instruction permits suitable ordinary initial applications within defined daily limits. Recruiter replies, interviews, assessments and offers remain separate decisions.

Additional Money keeps service opportunities and selected platform applications apart from the job lane. It retains its own reports, eligibility checks and confirmed outcomes. Specifically authorised free initial applications may proceed; new terms, payments, contracts, identity checks and personal screeners have separate boundaries. A local preparation receipt, application submission and settled income are different outcomes and remain labelled accordingly.

08Planning and local preparation

Finance and practical Life administration

Finance supports budgeting, savings, runway and resilience planning from the current permitted evidence. Its recurring work can produce a money-move or planning pack. Source dates and missing inputs remain part of the result, so an old account view cannot become a fresh recommendation simply because the interface refreshed. Finance does not transact.

Life coordinates household plans, travel and personal obligations. The latest preparation increment separates work owed by the owner, other people's commitments, verification holds and terminal history. It ranks a bounded set of preparation packets and requires independent closure evidence. This first increment produces local plans and holds; complete source-bound drafts, executed administration and real-world closures are later work.

09Recorded local routines

Automation with reports and meaningful receipts

Recurring controllers read their permitted sources, prepare local plans or assets and submit typed results for independent Review. A useful result may be a changed artifact, a verified unchanged result or a proved blocker. These are distinct states. Review checks whether the reported outcome matches its evidence, while OS tracks reliability, source freshness and improvement proposals.

Bot conversations receive relevant changed results and sustained source incidents. Routine unchanged checks stay in reports, Details or expandable Earlier automated checks rather than filling the chat. Direct command receipts keep their own identity, so a repeated check is not presented as a second execution. Recorded local activity supports inspection; external outcomes require the appropriate provider confirmation.

10Device evidence is dated separately

One decision model across application surfaces

The operating model is shared across the desktop application and responsive interface: current work, specialist conversations, evidence and decisions retain the same meaning. A wide screen can place Details beside a conversation; a narrow screen gives the conversation and its controls more focused space. Existing Android, foldable and Wear OS work extends the estate around device-specific tasks.

The current public tour renders the genuine shared interface with fictional records and visible Demo data. It demonstrates layout and workflow without disclosing personal finances, health facts, contacts or provider receipts. Current desktop evidence and earlier physical-device evidence are described separately. The phone lane is held during device repair, so this refresh does not claim fresh wearable measurements, phone push delivery or full device feature parity.

02 / What changed

Personal OS: decisions, controls and operating change.

These are observed properties of the working system. They are not claimed percentages, revenue attribution or a guarantee of future performance.

Operating result

One daily decision surface

Priorities, open work and decisions share a clear route into the supporting detail.

Operating result

Persistent specialist conversations

Users can continue a role's conversation and inspect the work behind its latest update.

Operating result

Current evidence beside the chat

Details shows current work, last result, blockers and recorded activity without inventing a successful run.

Operating result

Review in context

Approval cards bring the proposed action, supporting facts and review controls together.

Operating result

Quieter continuity

Repeated historical checks remain inspectable without filling the main conversation.

Operating result

Separate opportunity workflows

The job lane and additional-money lane retain their own reports, permissions and confirmed outcomes.

Agent team

Hermes and nine featured specialists, with clear jobs.

The public tour follows ten named roles, including Hermes. Each specialist has a defined responsibility, an ongoing conversation and inspectable work. Capabilities vary by command, source and device; the cards explain the useful job and the boundary around it.

Main assistant

Hermes

Connects permitted context across the workspace, answers grounded questions and prepares a clear decision brief. It routes the user into the appropriate ongoing specialist conversation and helps keep ownership visible. Its oversight can surface relevant recorded outcomes and sustained source problems without making every background check another message.

Action boundary
An answer, model selection or oversight update does not prove that a separate task executed.

Priorities and obligations

Work

Connects current priorities, inbox obligations, calendar context and founder administration. It maintains the goal-to-action preparation behind the work view and gives the user a route to the evidence and next decision. Its Details panel distinguishes current tasks, latest results, blockers and recorded runs.

Action boundary
Preparation is separate from confirmed execution; sends and other consequential actions use their own review.

Content and qualified growth

Personal Brand

Prepares source-backed posts, infographics, carousels and companion copy within the personal brand system. It also prepares qualified LinkedIn growth actions with the account, target and exact wording visible in chat. LinkedIn Growth opens this existing specialist conversation, while Details exposes dated work and current access limits.

Action boundary
Local review is separate from publication or outreach; a live browser picture is not available.

Jobs and contracts

Daily Job Agent

Checks suitable ordinary jobs and contracts, prepares evidence-led application material and advances specifically authorised initial applications. Its separate report records provider-confirmed receipts and current failures. Hermes oversight can surface suitable confirmed applications and the exact run result without presenting a local draft as a submitted application.

Action boundary
Defined suitability and daily limits apply. Recruiter replies, interviews, offers and assessments remain separate decisions.

Service and platform opportunities

Additional Money Agent

Maintains a separate workflow for service opportunities and selected income platforms. It checks eligibility, prepares next steps and records the difference between local work, a confirmed initial application and a real financial outcome. Routine zero-result checks stay in reports and Details so they do not crowd the conversation.

Action boundary
Only specifically authorised free initial applications may proceed; new terms, payments, contracts and personal checks remain separate.

Budgeting and resilience

Finance

Uses permitted current finance evidence to support budgets, savings, runway and financial control points. Recurring work can maintain a planning or money-move pack, while source dates and missing inputs stay visible. Its chat supports questions about that evidence rather than implying that a polished card makes every input current.

Action boundary
Does not transact. Stale evidence can block a specific review even when the conversation is available.

Meals, training and recovery

Health

Supports non-diagnostic planning for meals, training, sleep, recovery and energy. Meal planning combines individual feedback, checked whole-pack stock, candidate recipes and served history. A weekly local meal mission has its own proof, allowing food planning to remain separate from the freshness of phone and wearable inputs.

Action boundary
No diagnosis or automatic purchase. Current device data and complete provider coverage are not assumed.

Personal administration and follow-through

Life

Organises household plans, travel and personal obligations. The latest local preparation increment separates work owed by the owner, other people's commitments, verification holds and history, then ranks preparation packets with explicit closure evidence. It keeps incomplete assets and missing source coverage visible instead of promoting a template to a finished task.

Action boundary
The first preparation increment is live; full drafting, executed administration and proved real closures remain later work.

Independent evidence and completion

Review

Checks whether controller outcomes match their supporting evidence. It distinguishes a genuine changed asset, a verified unchanged result and a proved blocker, and challenges unsupported success or stale state. This gives the system a completion check beyond the agent that prepared the work.

Action boundary
Judges the result; it does not execute the work it is evaluating.

Reliability and recovery

OS

Tracks source truth, reliability, recovery and safe improvement proposals. It maintains recorded control information and cost evidence, helping the user distinguish a source problem from completed work. Its role supports the existing specialist estate rather than creating another chief-of-staff identity.

Action boundary
Reports and limited recovery do not imply unrestricted system changes or guaranteed availability.

03 / Interface tour

Role views and decision surfaces inside Personal OS.

Every image preserves the interface purpose and control logic while replacing live people, projects, records and values with staged data.

Personal OS personal agent roster and Work conversation using complete fictional records.
Current app interface / demo data / 30 September 2026Specialists in the same workspace. The Personal bots list opens directly into a role's conversation. Work's demonstration brief records its status, output, evidence and preparation boundary. View full-size image
Personal OS Work agent details showing responsibility, current task, result and review boundary.
Current app interface / demo data / 30 September 2026What the agent does and what it has done. The native details panel exposes responsibility, current work, result, next action and dated run evidence. A chat reply alone does not mark work complete. View full-size image
Personal OS Personal Brand exact-message review card with fictional recipient and draft text.
Current app interface / demo data / 30 September 2026Review the exact message. The current review card shows the recipient, account, target and exact text together. This read-only preview cannot approve, reject or send the fictional action. View full-size image
Responsive Personal OS Hermes conversation at 390 pixels wide with three suggested review choices.
Current app interface / demo data / 30 September 2026Suggested replies on a narrow screen. The same current interface adapts its conversation and choices for a phone-sized browser. Selecting a choice fills the composer; it does not send the reply. View full-size image
Responsive Personal OS exact-message review card on a phone-sized browser with fictional data.
Current app interface / demo data / 30 September 2026Exact review on a narrow screen. Recipient, target and message remain readable in the responsive review card. This is a current browser capture, not evidence of a newly released Android or Wear OS build. View full-size image
04 / Feature evidence

Explore the evidence for Personal OS.

Each entry links to the exact view it describes. Product captures and explanatory diagrams retain their own evidence labels.

01
Current app interface / demo data / 30 September 2026

Hermes connects the work

What this evidence shows

The current shared interface brings a persistent conversation and a reviewable response together. This fictional example connects Work, Personal Brand and Review without sending anything. Simple role icons replace the portrait placeholders in this demonstration capture.

02
Current app interface / demo data / 30 September 2026

Specialists in the same workspace

What this evidence shows

The Personal bots list opens directly into a role's conversation. Work's demonstration brief records its status, output, evidence and preparation boundary.

03
Current app interface / demo data / 30 September 2026

What the agent does and what it has done

What this evidence shows

The native details panel exposes responsibility, current work, result, next action and dated run evidence. A chat reply alone does not mark work complete.

04
Current app interface / demo data / 30 September 2026

Review the exact message

What this evidence shows

The current review card shows the recipient, account, target and exact text together. This read-only preview cannot approve, reject or send the fictional action.

05
Current app interface / demo data / 30 September 2026

Suggested replies on a narrow screen

What this evidence shows

The same current interface adapts its conversation and choices for a phone-sized browser. Selecting a choice fills the composer; it does not send the reply.

06
Current app interface / demo data / 30 September 2026

Exact review on a narrow screen

What this evidence shows

Recipient, target and message remain readable in the responsive review card. This is a current browser capture, not evidence of a newly released Android or Wear OS build.

05 / System design

Context, evidence and control layers behind Personal OS.

The visible dashboard is only the decision surface. Context, data, permissions, review and learning determine whether it can be trusted.

01

Current source evidence

Permitted sources supply dated facts for each module. Coverage, checked time and missing inputs remain visible, so a newly loaded interface cannot silently refresh an old result.

02

Work and obligation model

Priorities, projects and commitments point back to their owning records. One decision keeps one identity across Work, Review and Needs you; terminal history remains available without appearing as new work.

03

Hermes and specialist continuity

The main assistant and specialist chats connect the operating context. Details exposes the responsibility, current task, last result and recorded activity behind a conversation.

04

Preparation and review

Local plans, complete drafts and exact action cards are prepared for inspection. Email sending, local LinkedIn review and suggested-response selection retain distinct meanings and controls.

05

Bounded recurring work

Controllers maintain permitted local assets, respect task-specific limits and report changed, unchanged or blocked outcomes. Independent Review checks the evidence before a result is accepted.

06

Feedback and history

Meal preferences, exact rejected revisions and retained activity influence the next preparation. Repeated checks stay inspectable without becoming repeated chat interruptions.

07

Application surfaces

Desktop and responsive layouts share the decision model. Earlier Android and Wear OS evidence supports a broader estate, with device-specific availability and freshness described separately.

06 / Working flow

How a signal becomes a governed decision.

A useful workflow names the input, owner, check, approval boundary and proof that closes the loop.

  1. Orient

    Open the daily view or ask Hermes a question about current work.

  2. Inspect

    Choose the specialist role and read its current work, latest result and source limitations.

  3. Prepare

    Review the recommendation, proposed action or exact draft.

  4. Decide

    Use the appropriate review control, or select a suggested response to place it in the message box.

  5. Confirm

    Check the actual result and retained evidence before treating the work as completed.

07 / Working features

Working controls inside Personal OS.

Each feature exists because it changes a decision, reduces unsafe ambiguity or makes a failure easier to see.

Working feature

Work, Review and Needs you

Different entry points for priorities, prepared decisions and owner actions, connected to the same underlying work.

Working feature

Persistent specialist conversations

Hermes and Bots retain the conversation, while Details supplies current tasks, results, blockers and recent runs.

Working feature

Inbox preparation

Connected email sources produce obligations and complete review drafts, with held items and coverage gaps visible.

Working feature

Read-only calendar context

Agenda and available time inform the day's shape without automatic booking or event changes.

Working feature

Meal planning and explicit feedback

Individual ratings, served history, checked stock and recipe evidence support local plans and shopping drafts.

Working feature

Brand production and exact review

Detailed posts, infographics and carousels combine source evidence, finished media and account-specific review.

Working feature

Personal admin preparation

Life separates obligations, verification holds and other people's commitments before ranking its next preparation packets.

Working feature

Finance planning

Current permitted evidence supports budget and resilience packs while stale sources remain a limitation.

Working feature

Independent completion checks

Review distinguishes a changed artifact, verified unchanged work and a proved blocker.

Working feature

Quiet automation history

Relevant changed results remain visible; routine checks are retained in reports, Details and expandable history.

Working feature

Separate opportunity workflows

Job applications and service or platform opportunities use distinct reports, limits and confirmed-outcome definitions.

Working feature

Source and device honesty

Unavailable inputs and device limits remain explicit rather than being inferred from a polished screenshot.

08 / Issues and changes

Incidents that hardened Personal OS.

The strongest evidence in an operating system is often the failure behaviour. These issues became design constraints and regression checks.

Issue 01

Fragmented conversations

What failed

Role cards and chats could show different versions of current work.

What changed

Connect named roles to the same persistent conversation and supporting details.

Issue 02

Repeated automatic checks

What failed

Routine updates filled chats without adding a new decision.

What changed

Keep repeated checks in reports or expandable history; show meaningful changed work and unresolved issues.

Issue 03

A suggested answer looked like an action

What failed

A choice could be mistaken for a sent instruction.

What changed

Selecting a suggestion only fills and focuses the message box. Sending remains a separate step.

Issue 04

Status without evidence

What failed

A polished summary could imply a completed task without a current result.

What changed

Keep source freshness, recorded activity and explicit limitations available for inspection.

Issue 05

Different work mixed together

What failed

Jobs, service opportunities and platform activity need different rules and proof.

What changed

Maintain distinct workflows and reports, and count outcomes only when confirmed.

09 / Controls and limits

What Personal OS proves, and what it does not.

A public case study should make it easy to see what is working, what remains internal and what has not been proven.

Known limit

Demo records only; a chat role does not imply the tools and permission to perform every task.

The visuals are captures from the working private product with demonstration data. They do not prove a public user or productivity result.

10 / Capabilities demonstrated

Capabilities demonstrated by Personal OS.

Strategy, interface, governance and implementation are treated as one operating problem rather than separate deliverables.

Product strategyPersonal operating systemsInformation architectureShared agent conversationsDecision designApproval designData freshnessResponsive UXBehavioural workflow design
11 / Follow the evidence

Each link adds a different layer of first-hand evidence rather than repeating the same page under another keyword.

12 / Clear answers

Personal OS: practical questions and answers.

Practical answers about the build, its use and its limits.

What does Personal OS actually do?

It connects daily Work, Review and Needs you to specialist conversations and practical preparation workflows: inbox replies, calendar context, meal plans, brand content, personal administration, financial planning and bounded opportunity work.

How does it know an agent has done useful work?

Details and reports retain current tasks, dated results, blockers and run evidence. Independent Review checks controller outcomes. A local draft, a confirmed submission and a real-world result remain different states.

How does feedback change the next result?

Meal feedback has explicit weekly or ongoing scope. Rejected email revisions stay rejected, while a changed proposal is reviewed as its own exact version. Routine repeated checks remain in history instead of returning as new decisions.

Does it send emails or publish posts by itself?

Ordinary email drafts use exact Review > Email approval. LinkedIn local review is separate from external execution. Only specifically authorised initial job or free platform application workflows can proceed within their defined limits.

Is every module fully autonomous?

No. Some modules prepare local work; some commands are unavailable or require fresher sources. Life's first admin increment is preparation, wider Health has stale device inputs, and private-email Jev activation remains unproved.

What do the screenshots establish?

Current shared interface, layout and review presentation with fictional Demo data. They do not prove native execution, a new phone or watch release, private-data availability or measured productivity savings.

Product strategy / Personal operating systems / Information architecture

Make the next decision clear, with the work behind it visible.

Thomas can map a founder's daily decisions and build the workspace, specialist roles, evidence views and review controls around them.