Know what Coach and AI each do
Measured facts, calculated guidance and AI-generated explanations have different roles.
Start with the floating Coach
The green Coach button stays beside the current page. Its compact goal card keeps your current focus and observed progress visible. Ask a question directly below it. Open Goals, history & tools for goal suggestions, saved analysis, the recorded timing guide and completed goals. Goal changes and marking a goal achieved remain explicit actions. Where private Coach history is enabled, confirmed conversations are saved for your account in the current environment.
This calculated guidance is not a model-generated answer. It does not infer a driving line from lap times or establish why a kart or visit was slower.
Choose a measurable focus
When your history contains different comparison contexts, choose one in Coach. The timing guide still summarizes the recorded sessions while you choose. A pace proposal uses a stated retained-average baseline and asks for at least nine retained laps in one session on a later matching visit. Provider all-lap averages do not count toward this target.
Proposals are calculated from your journal and do not call a model. Review suggestion lets you review and edit the title and target before Accept as Home focus saves it; your previous goal remains in history. Completed goals stay completed. Review observed progress and explicitly mark the new goal achieved when appropriate.
Automatic Coach for eligible accounts can read your active goal and its observed progress, completed goals, upcoming plans, relevant private notes and recorded trial outcomes. Saved notes are observations, not verified causes. A plan is not a booking. Saved legacy Session planner forecasts remain a separate feature.
Open saved analysis when you need more
Open saved analysis & AI tools opens the full page for saved Session and Progress analysis. Choose the evidence you want to review. Saved session analysis does not yet support some Event-linked or classified session contexts. Coach chat uses its own eligibility check. This context limit is separate from account access; choose another supported session or use the timing guide. A context that cannot be verified stays blocked until it can be checked again. A saved answer can be read without generating another one; older evidence is labeled so you can see when the answer needs updating.
AI access is limited to eligible accounts. Use the visible Generate, Update or Ask Coach control to request an answer. Opening a page, changing venue or reading a saved answer does not start a paid model request. No general availability or pricing is promised by this guide.
Read the origin label
An AI-generated label identifies content actually produced by a model. Read its source references, date and limitations. Deterministic summaries and fictional Demo examples should be labeled for what they are, rather than presented as generated AI.
A model may explain prepared evidence. It does not replace measured lap times, change eligibility, fill missing data, certify a result, or silently save your goal. A session called Race or a recorded P1 still needs a known ranking basis before it can support a win claim.
Keep your own judgment
Check an explanation against its cited sessions. Treat causes without supporting evidence as unknown. Saving a suggested action is a separate choice, and you can keep using the timing views when AI is unavailable.
Complete raw email and sensor archives are not the prepared coaching evidence. Provider response storage is requested off, but provider security and abuse controls can still apply. See account and privacy for the separate storage and consent boundaries.
Choose a useful next action · Understand the numbers
Prepared Demo examples
The focused Demo labels its prepared coaching examples. They are illustrations grounded in fictional timing, not a claim that AI just generated an answer. Demo never calls a paid model. You can keep supported private goals and notes without enabling real AI access.
Ask a follow-up
For eligible accounts, ask in plain language and use Send or Enter. Coach automatically selects bounded relevant account history for day, progress, kart, field, goals, notes and next-race questions. Factual questions receive recorded facts without requiring a coaching suggestion. Advice questions separate suggestions from measurements. Two named dates or karts can be compared, with source-backed kart labels and admitted-evidence counts. Follow-ups use up to six confirmed messages plus separately checked recorded-source references; a new topic resets the scope. Anonymous field summaries compare like metrics and show coverage counts, with up to six sessions listed together; missing or omitted summaries are labeled. They do not supply competitor lap sequences or identify opponents. New replies scroll into view when you are at the bottom. If you are reading earlier messages, use New reply to jump to the answer. The four suggested questions send immediately when clicked. A question with an ambiguous kart venue asks for clarification. Shift+Enter adds a new line. Live chat can use your AI allowance only when you submit; opening the guide or typing does not request an answer. Coach separates general preparation advice from recorded evidence and limits. Up to six recent confirmed messages provide follow-up context. History selection examines at most 200 results and includes at most six visits and 100 drives. Explicit dates are filtered before this limit; otherwise the newest results are examined. Kart-only questions search the newest 200 account results; older uses can be omitted. Previous-visit kart questions show dated observations from up to six matching visits rather than guaranteeing a pair. Trial notes are labeled when drawn from the whole notebook. Omitted history is not proof of absence. If evidence cannot be verified, your question stays available to retry and no answer is invented. Access readiness is checked separately from question-specific evidence. Chat does not save a goal or action; acceptance remains a separate choice.
For preparation and best-lap questions, the main recorded evidence is the calculated minimum of the admitted selection, not an arbitrarily selected session. Its dates, source session and coverage stay visible. A six-visit minimum is not labeled as your lifetime best; explicit dates and selected sessions keep their own scope. Where private history is enabled, reopening Coach loads your confirmed conversation across devices in the same environment. DEVELOPMENT and PRODUCTION histories are separate. History offers older conversations, New chat, Export and confirmed Delete. Questions show accepted, answering, complete, failed or outcome-unknown status. Refresh status only checks saved status; Retry is an explicit action that can use allowance. Unknown does not mean unbilled. Revoked access clears the visible conversation. Old temporary chats are not automatically imported. If generation is disabled, saved history remains readable and exportable; sending is unavailable.
Broad best-lap questions use a complete recorded-history snapshot when supported, with venue and coverage shown. Unsupported average, date-window, kart, layout, class or other qualifying conditions ask for clarification instead of presenting an all-time minimum.
Previous interface referencePreserved guidance · 2026-09-28
Coach in Driver OS
The floating Coach is a deterministic guide to selected evidence. It explains supported pace, variation, kart-use and field observations and lets you edit a private goal. Its conversation stays in the current tab. It does not infer a racing line from lap times or turn a note into a verified cause.
Open saved analysis & AI tools opens a native full-width page. Choose a recorded session to read its saved analysis, or use Progress analysis. Current saved answers restore without a model request; older evidence and versions are labelled. Source links, insights, limits and AI provenance remain visible. Eligible Generate, Update and Ask Coach buttons use the existing protected service and allowances only when clicked.
Plan explanations and Coach recaps remain separate requests and require supported evidence. They are not presented as durable saved history. Other historical detailed-tool and Race Day workflows described below remain preserved at their existing routes, but are not links from the new interface. No backend record or saved analysis is removed.
Floating Coach
The floating Coach stays beside the page while you navigate. Its calculated quick questions and matching saved AI do not trigger a new model request. On a phone, the panel can minimize so navigation remains reachable.
For the authorized owner, choose a Chat session, Check AI access, type a timing, consistency or result question, then Send to Coach. Chat stays pinned to that recorded session even if the page changes. Every question is independent; no conversation is saved. Only Send can use the existing shared monthly allowance. Broader progress, kart-cause, technique and unrelated questions receive bounded guidance. Sources, limits and AI provenance remain visible.
In the wide review, Follow the evidence → Saved analysis & session notes reads an existing Session answer in place. It keeps the whole source-session scope, source references, generated time, model, versions and current/stale status visible. The availability check only reads authorized saved information; expanding the evidence does not call the model. Notes and tags remain driver observations, not new AI conclusions.
Controls for requesting or refreshing saved Session/Progress analysis remain in the named detailed tools. Opening these tools does not grant AI access or expand the account allowance.
Selected evidence in main Coach
Opening main Coach from a day, session or Progress comparison keeps that selected evidence visible. It does not save a plan, complete a priority or submit a paid question. The floating Coach has its own explicit Chat session selection, which stays pinned while you navigate.
Start with Driver OS evidence
Lap times, representative pace, lap time variation, comparisons, plans, eligibility, confidence and timestamps come from deterministic Driver OS code. Review those values and their sources before asking for an explanation.
When an AI action appears
Model backed access is limited to explicitly entitled accounts and is not a general early access capability. Session and Progress tools run only after their visible request action. Opening Home, Sessions or Race Day tools does not start a new analysis. Race Day coaching requires its explicit Generate or Update action.
Every complete model answer carries an AI-generated label, evidence references and limitations. If those elements are missing, do not treat the text as a Driver OS AI explanation.
Temporary questions
A temporary protected question may include a bounded amount of selected session context, including a note when that context is part of the chosen request. The answer is not saved as Driver OS product data, although operational usage metadata is retained.
Saved Session and Progress analysis
Saved analysis uses a narrower prepared fact packet and saves the validated answer, evidence references, audit fields, version fields and model usage metadata to the account for that exact evidence.
If the evidence changes, the earlier result remains visible as stale until you explicitly request an update. An unchanged current result should restore without another provider request.
All provider calls request application response storage to be disabled. Provider security and abuse controls may still apply under the provider terms. Driver OS does not promise that the provider retains nothing.
Owner Race Day Review coaching
Where enabled for the exact owner account, Race Day Review shows AI coaching analysis inline. In the wide review, opening these tools reads an existing saved answer without a new model request. Choose Generate coaching analysis for a missing answer or Update coaching analysis for changed evidence. The official results must have settled for 30 minutes after their latest import or update, or the venue-local day must have ended. Opening early still shows the complete timing review immediately.
All six coaching sections expand into the page. Read the analysis, evidence links and source details using normal page scrolling.
The analysis is saved for the evidence it reviewed and restored on return without another model call. New results preserve the earlier analysis as outdated. Driver OS offers an explicit update after the next quiet period. A failed attempt does not retry automatically; Retry coaching analysis lets the owner try again within the same limits.
Each analysis must fit a maximum $0.02 reservation and accepted cost, a $1.00 monthly Race Day Review allowance and the existing account limits. If access, budget or provider availability prevents analysis, the deterministic review stays complete. Model, saved time, cited evidence and limitations remain visible. Saving the suggested focus is a separate action controlled by the driver.
What AI never replaces
AI does not receive complete raw email, complete sensor archives, GPS routes, heart rate or raw sensor channels. It cannot import or repair results, certify records, invent measurements, validate a weak sensor signal or silently change a plan.
Ranking facts stay fixed
Coaching uses the same deterministic ranking facts as Sessions and Progress. Session purpose, optional Event membership and ranking basis are separate. A numeric P1 or a session named “Race” cannot establish a win. Unknown basis remains explicit, and the model selects prepared claims rather than rewriting measurements.
Qualifying position does not establish the starting grid. Net positions gained needs the recorded start and classified finish in that race; it does not count overtakes or explain what caused the outcome. Changed ranking context makes older evidence snapshots stale. Owner-confirmed context currently uses deterministic review; it does not enable another paid coaching path.
Use the explanation well
Check every statement against the cited Session or Progress evidence. Keep one useful interpretation, then turn it into your own practical action in Coach.
Choose the next action Understand the numbers Review account and privacy boundaries
Saved Demo coaching examples
Demo Testers see precomputed, deterministic examples derived from fictional scenario records. The examples show their source references and limitations and are labeled synthetic. They do not call a model, incur AI usage or make claims about the tester's real driving. Event context preserves qualifying, race and stint distinctions; unsupported live Event AI remains unavailable.
Setup choices and recovery
The AI question during onboarding records interest only. You can change it in Profile → Your driving choices. Selecting it grants no AI access, accepts no consent, starts no analysis and creates no charge. Existing account approval, availability and usage limits still apply; pricing has not been decided.
Demo Testers with the race-day refresh can choose a suggested priority, put it first on Home, dismiss suggestions and mark priorities complete. These choices use fictional session evidence and do not call AI. Existing plans and saved work stay in place. Older Demo versions show the saved examples and Session planner while priority controls remain unavailable.
Venue context and selected evidence
The venue you are browsing limits relevant timing summaries. General saved goals remain available and are not rewritten when you switch venues. Direct Session and Event entries keep their selected evidence; a missing or unavailable session does not substitute unrelated Home history. Saved explanations must match their evidence and scope before reuse. Opening a venue, page or saved explanation never starts a paid request.