Watch understanding become action.
One conversation. A source you can question. A plan you control. A useful preference for next time.
One core route, optional depth
Start with the task-first route. Optional document and support-preference routes rejoin at context review. Adaptive scenes show proposed extensions; privacy and recovery are available throughout. Scene IDs are stable review references, not steps everyone must complete.
The version filter shows cumulative design coverage. Earlier versions are development proofs: authentication arrives at V0.03, real support at V0.04 and a real Work workflow at V0.06.
Who this is for
Proposed first cohort: UK adult higher-education users preparing for a university support meeting. The no-document route remains fully valid. This scenario replaces the investor-meeting example for review; adoption remains open.
Interface buttons illustrate proposed actions; no sign-in, file upload or AI processing occurs. Use route/version filters to explore. Review notes stay in this page until downloaded and are lost on reload. No analytics or external requests are included.
No scenes on this route at the selected increment. Choose a later version or another route.
The last three gates validate the same journey
V0.08 · Trust & recovery
Verify mobile and keyboard completion, context lifecycle, export/deletion, safe failures and cost limits. Every relevant scene needs its recovery states.
V0.09 · Rehearse
Three observed users complete the loop. Close release blockers and prove clean deployment, rollback and restore. No additional product journey.
V0.1 · Admit the cohort
Jointly approve the exact build, complete production smoke and admit invited users with honest beta limits and a support route.
One journey, ten cumulative proof gates
Target: 05 October 2026. This preview is a proposed product/design specification. A scene’s target increment is not evidence that it is implemented. The original DOCX baseline remains distinct from the additions below.
| Increment / date | Baseline proof | Preview delta / requirement anchor |
|---|---|---|
| V0.01 · 7–9 Sep | Reproducible Home/Chat/Work shell at desktop/mobile; synthetic fixtures only. | PRD/TRD-V01-01 · freeze scoped shell; storyboard optional routes before building them. |
| V0.02 · 10–11 Sep | Mock send/stream/stop/retry/error states; ordered events and no duplicate submission. | 02 · simulate task-first interaction; optional setup fixtures do not become mandatory onboarding. |
| V0.03 · 14–15 Sep | Invite-only identity, history, migrations and owner isolation. | 03 · minimal landing/access. Private document storage is additional scope if adopted. |
| V0.04 · 16–18 Sep | Optional discovery, governed context and real support. | 04 · task-first review; source-linked document extraction and sensitive disclosure require approved policy and scope. |
| V0.05 · 21–22 Sep | Explicit Work approval, owned task and editable saved artefact. | 05 · support brief and versioned replan foundations are proposed deltas. |
| V0.06 · 23–24 Sep | Qualified real workflow; FlowState baseline behind adapter. | 06 · demonstrate one approved ten-minute replan. Signed ADR required for runtime substitution. |
| V0.07 · 25 Sep | Outcome → proposal → confirm/edit/reject → later eligible use; feature freeze. | 07 · scoped correction and modest evidence history. No automatic profiling. |
| V0.08 · 28–29 Sep | Lifecycle, security, accessibility, rate/budget controls and recovery. | 08 · document/derived-context deletion and export if ingestion is included. |
| V0.09 · 30 Sep–1 Oct | Observed UAT, clean CI candidate, rollback/restore and no P0/P1. | 09 · rehearse one scenario; 2 Oct is fix buffer, not a new-feature day. |
| V0.1 · 5 Oct | Joint GO, production smoke, controlled first cohort. | 10 · exact commit/artefact/runtime/schema and dated evidence. Begin small; expand only after review. |
Personal foundation, without a personality test
What the user says
Preferences, strengths, difficulties and optional identity disclosures. Preserve uncertainty and the user’s wording. No label is needed to receive support.
What a source says
Document passage, author, date and context. A recommendation, reported difficulty and formal finding are different types of evidence.
What may be used
A purpose-bound, user-approved support approach. Task context and durable learning have separate scope and retention decisions.
Internally, NPR is the governed record behind this foundation. An archetype is a provisional support hypothesis. Neither a report nor conversational patterns automatically establishes a diagnosis or overrides current instructions.
Document boundaries
| Source | Potential use | Boundary |
|---|---|---|
| EP report / DSA statement | Strengths, recommendations and study support to review. | Initial wedge candidates; distinguish assessment findings from recommendations and current user preference. |
| PIP documentation | Reported assistance needs and decision history. | Future exploration; not benefits eligibility, legal advice or automated casework. |
| EHCP | Recorded needs, outcomes and provision. | Future exploration; child/guardian and institutional workflows are outside this adult beta. |
| CV | Skills, experience and goals. | Future career route; not evidence of psychological traits or neurodivergence. |
Architecture and decision register
UI → authenticated product API → eligible context projection + policy checks → SupportRuntime adapter → approved product write. akilii owns identity, data, context and permissions. The runtime executes bounded work; it does not become the authoritative personal record.
Mock is for deterministic tests; direct provider enables real support; FlowState remains the baseline real Work runtime. Capabilities, errors, ordered events, cancellation and idempotency use product-owned contracts. No hidden replay onto another provider.
| ADR-V01 | Decision / gate |
|---|---|
| 001 / 010 | Canon, capacity, budget and thresholds · kickoff/G01 |
| 002 / 006 | Runtime boundary, events, idempotency and cancel · G02 |
| 004 / 005 | Existing stack/auth/data · G03; NPR lifecycle and Context Compiler · G04 |
| 007 / 003 | One workflow and internal actions · G05; qualified real runtime choice · G06 |
| 008 / 009 | Privacy, routing, retention and deletion · G08; clean CI and recovery · G09 |
Preserve the plan’s ADR IDs. Document ingestion, sensitive disclosures and scenario changes require amendments or new accepted records; this preview does not sign them.
Working agreement and release floor
André Skepple: product, UX, design, canon and product acceptance. George Nangle: implementation, engineering feasibility, tests, CI/CD and technical challenge. Joint: runtime, data/privacy, capacity, cost, scope and release.
Daily short async update; Monday capacity review, Wednesday demo, Friday gate/risk review plus target-date gates. One active build increment; next handoff prepared one working day ahead. No weekend capacity assumed.
Versioned PRD/contract governs behaviour; exact Figma nodes govern visuals; accepted ADRs/contracts/migrations govern engineering; a clean CI artefact governs release. This HTML is a review asset, not a competing source of production truth.
- Fresh clone, pinned toolchain/dependencies, non-secret environment example, migrations and synthetic seed. CI checks types, build, relevant contracts/state/isolation and critical journey.
- Provider keys remain server-side. Enforce ownership, input/output validation, minimal context, safe rendering and prompt-injection boundaries for documents.
- Proposed £150 sprint and £5 daily live-runtime limits require adoption. Alerts at 50/80%; stop new paid runs at the adopted ceiling. Inventory unused Azure and verify residual charges.
- Every gate needs linked implementation/tests, prior gates passing, no blocking defect and both owners’ dated acceptance.
- Cut visual breadth, extra document formats, extra workflows and comparison depth first. Never waive identity, user control, real execution, truthful lifecycle or recovery.
Final checklist: scope/source freeze; local/CI reproduction; one complete loop; real runtime trace; context controls; isolation/secrets; privacy/safety; accessible UAT; cost/reliability; rollback/restore/support cover; no P0/P1; production smoke and controlled admission. All are pending evidence.
A focused beta hypothesis
Proposed wedge: UK adults in higher education, with an upcoming university support meeting. Existing EP/DSA information can help, but documents and diagnostic labels are optional.
Promise to test: “Bring your existing support information, decide what still fits, and prepare clear questions and next steps.” Candidate channels: willing disability advisers, specialist study-support practitioners and student communities. No partnerships or demand are claimed.
Proposed first-cohort decision gates
These are internal hypotheses for 20 users—not industry benchmarks, clinical evidence or established product-market fit.
| Question | Measure |
|---|---|
| Recruitment | Invitation → eligible interest → consent → activation, by channel. |
| First value | Aim for 14/20 to approve relevant context and produce a usable preparation artefact; record time and human assistance. |
| Repeat value | Aim for 10/20 to return for a second meaningful task within 14 days; distinguish prompted from unprompted returns. |
| Personalisation benefit | Counterbalance comparable tasks with/without approved context. Compare usefulness, corrections and completion. |
| Onboarding effort | Compare task-first with optional guided discovery. Measure abandonment, questions skipped, time-to-draft and feeling accurately represented. |
| Commercial evidence | Test a specific paid continuation or funded pilot offer. Distinguish actual commitment from stated willingness. |
| Trust and economics | Extraction errors, unwanted inference, honoured corrections, cost per retained user and founder-support minutes. |
Competitive context
Desk research reviewed 05 September 2026. Published capabilities are not a hands-on comparison. The opportunity below remains a hypothesis.
| Reference | Established overlap | What akilii must prove |
|---|---|---|
| NotebookLM | Document-grounded answers and citations. | Approved context → useful action → later adaptation beyond summarisation. |
| Goblin Tools | Task breakdown and practical assistance. | Context improves outcomes enough to justify setup. |
| Tiimo | Neuroinclusive planning; company reported 50k+ paying subscribers in August 2024. | Repeat use and payment behaviour, not interest alone. |
| Cognassist / Kai | Cognitive-profile-based strategies with alignment explanations. | User-owned existing evidence and corrections provide a better entry and continuing value. |
| Genio Notes / Glean | Focused capture-to-application workflow and DSA assessor materials. | An equally clear job and viable recruitment route. |
Pre-seed potential: promising, unproven
Tiimo reported a $1.6m round in August 2024, bringing total funding then to $4.8m. Gresham House confirms £4m invested in Cognassist in April 2023. These show adjacent investment activity; they are not valuation comparables or funding predictions for akilii.
Build the case with a narrow recurring problem, observed retention, comparison against simpler alternatives, a repeatable channel, real paid commitments and measured costs. Defensibility must come from the governed context-to-action loop, evaluation and distribution. Sensitive documents are not a moat or presumed training data.
Clinical boundary
User-reported neurodivergence, experiences and preferences remain distinct. The system does not diagnose from chat, reports or interface choices. Screening and diagnosis require separate scope and qualified assessment pathways. NICE ADHD guidance · NICE autism guidance.
Review before implementation
New in 3.1: Live walkthrough provides scripted streaming, a + attachment/mode menu, source confirmation, an editable Work plan, explicit replan approval, learning consent and a fresh-conversation demonstration. It is an interactive product simulation, not a connected AI system. No uploaded file content is read.
Preview revision 3.1 · 05 September 2026 · target delivery 05 October 2026. Product journey and additional scope await joint review. No real accounts, files, payments, messages or AI calls are available here.
- Walk through the task-first core route.
- Explore optional evidence and personal-foundation branches.
- Review the replan, support brief and scoped learning.
- Check the build deltas, ownership and release gates.
- Record changes, concerns and decisions using stable SB scene IDs.
Notes are held only in this open page and are lost on reload; download them before closing. The clean preview excludes review notes. Send the downloaded HTML to George or another reviewer; it works offline in a modern browser. No public URL has been created and nothing has been sent automatically.
Decisions requested
- Adopt the adult university-support-meeting scenario?
- Retain task-first entry with optional, bounded discovery?
- Include document ingestion and sensitive disclosures in October, or defer until privacy/capacity gates are met?
- Include replan and support brief within the one workflow?
- Which exact screens become canonical after review?
Visual alignment · preview 2.1
Uses exported canonical brandmark, wordmark and navigation/panel icons; embedded Inter and DM Serif Display fonts; semantic forest/paper/sage colours; the definitive 240px sidebar, 76px header, 414px contextual panel and canonical card/button/composer grammar. Compact review cards use a collapsed rail; Scene by scene shows the expanded desktop shell. Mobile preview uses a compact header and contextual disclosure. The calm background omits the ambient raster artwork rather than substituting an unverified pattern. Proposed scenes remain adaptations, not approved pixel-perfect production screens.
Visual references: definitive sidebar · Home layout. Existing legacy sidebar lineage has not been copied.
Journey changes from preview 1.0
The investor example is replaced with a proposed university support meeting. Added optional foundation/disclosure, document review and disagreement, evidence-to-action, adaptive explanation/replan, support brief, scoped learning and lifecycle scenes. Legacy SB-05 is superseded by SB-35 targeted discovery. Other IDs remain stable. Figma and the DOCX plan have not been rewritten by this preview update.