Notes options google vs own
Status: proposal
The live notes pilot saves drafts and versions, but cannot yet delete drafts or share notes.
Google already supplies the document editing and collaboration this two-person shop needs.
On phones, reliable Google editing means opening its editor outside the EmbeddedEra app.
Keep engineering notes, SOPs and agent-maintained files in the Obsidian Git vault.
Use one app list for Google documents, vault notes and small private captures.
Recommendation: C — hybrid; shrink the pilot. Reuse already-paid Google Docs/Sheets for substantial human documents. Keep the app editor for quick capture and project to-dos. This requires an owner-approved exception to “every click stays inside the app”; a thin Google wrapper cannot honestly promise that experience. No implementation is authorized by this proposal.
Owner decisions — proposed, not yet accepted
- Allow Open in Google Docs/Sheets from the app, returning through the app switcher/back navigation. If never leaving is absolute, reject B/C as described and evaluate an embeddable office engine before expanding custom development.
- Keep Google as the original for proposals, customer documents, collaborative writing and spreadsheets; keep the vault as the original for engineering, SOPs, decisions and agent work.
- Keep capture/project editing; prioritize Discard draft and Trash saved note → Undo. Pause custom collaboration, rich comparison and external-link development while the Google workflow is evaluated.
- Default to private creation and named recipients. Use a named version plus an approved PDF/copy for sign-off; a shared live Google document updates while people type.
- First try one real document with both owners on their phones, then consider a 6–10 focused engineering-day hybrid integration. Confirm the Workspace edition, bill, sharing settings and phone handoff before committing to that estimate.
Verified starting point
The app note records v2026.10.13 live on 2026-10-04, owner-only: Milkdown/Crepe, autosaved drafts, numbered versions/history/restore and project editing. Its release log supersedes the older “not started” wording in the plan and prerelease status in /home/angel/assistant/NOTES_PILOT.md. Project to-do deletion exists; personal-draft deletion is a separate missing action.
The requested search across /home/angel/Notes found Workspace mail on embeddedera.com in site notes, and Google Workspace identity in SOP-009. The app rules require Google through Cloudflare Access. The owner's paid Workspace is established; edition, two licensed seats, exact monthly bill, Drive usage and current admin settings remain unverified. No Google Admin account was accessed.
Site notes record iam.disableServiceAccountKeyCreation blocked and a decision to leave it alone. That is dated evidence, not a fresh policy check. Existing Google login authenticates the app user; it does not grant the app permission to read their Drive.
Comparison: effort and recurring cost
Estimates below are engineering judgment, not supplier promises; include integration/review/device checks, exclude waiting for owner/admin access. Add roughly 25% contingency. Existing server/backup costs continue; additional hosting capacity and labor are not priced.
| Option | Work from today | Extra monthly software cost |
|---|---|---|
| A: extend our editor | 10–16 days for missing core features | $0 licences; our maintenance |
| B: Google-led notes/documents | 4–7 days for a narrow wrapper | Expected $0 within existing seats/quotas |
| C: hybrid | 6–10 days including capture cleanup and vault list | Expected $0 within existing seats/quotas |
| D: another hosted-by-us tool | See candidates below | Often $0 core; features/support may cost |
A — keep building our own
Best match for a consistent phone-first app and literal Markdown ownership. The live pilot has Word-style basic formatting, recoverable local/server drafts and an explicit Save that creates an attributed version. It is not a full Word layout engine; cold offline launch is not supported by the pilot design.
Private author access exists. Inside sharing/contribution, any-two rich comparison, outside links and PDF/print/downloads remain work. The plan budgets 5–8 days for sharing plus comparison and 4–6 for exports/outside sharing: 9–14 days remaining, plus a new 1–2-day allowance for draft discard/note trash and recovery checks = 10–16. Controlled SOP proposals add 2–3 days separately. Original whole-plan total: 19–28 days, not a fresh remaining estimate. Local plan.
Agents and Obsidian already understand the files; Git plus app metadata preserves numbered history. Keep both backed up: Markdown alone does not preserve access grants or version receipts. Privacy is from ordinary app readers, not from privileged filesystem/repository administrators. Custom ACLs, revocation, merge handling, publishing and phone editor maintenance are the main continuing risks. Lowest format lock-in, highest responsibility for us.
B — Google Workspace as document engine
Google becomes the default destination for app-created human notes/documents; the engineering vault still survives separately. Strong Word-style writing, autosave, simultaneous contribution, comments and suggesting mode already exist. Native phone editors are the practical path; desktop opens Google's web editor. Autosave, iOS suggestions.
What the thin wrapper can do: show/search authorized files; create a blank document or copy a template; rename/trash; grant viewer/commenter/editor access; list revision metadata; export PDF. Use Drive files.list with queries, files.create/files.copy, files.update, permissions and revisions.list. These are ordinary API operations, not an editor we must recreate. Drive reference, search, sharing.
Docs API reads/creates/updates document structure and fills templates. Sheets API reads/writes cells, formulas and formatting; it is useful for real spreadsheets, not a replacement for our note store or ERP. Neither supplies the Google editing screen as an embeddable UI component. Docs overview, Sheets API.
Google Picker is a ready file-selection/upload dialog, not a document editor. Prefer its existing integration over inventing a Drive browser. Private-file selection needs an OAuth access token; its API key identifies the project and does not authorize private content. Picker itself uses an iframe, so test its phone/account behavior too. Picker, web integration.
The iframe limit: Google's documented Publish-to-web embed is a viewing/publication workflow, not private collaborative editing. Publishing can widen the audience; never use it to “fix” private editor embedding. A docs.google.com edit iframe is not a dependable supported product contract for this PWA. Safari/WebKit blocks third-party cookies by default; Storage Access needs cooperation/user interaction from the embedded site. Our app cannot simply make Google's authenticated editor work by adding an iframe. Google embedding, WebKit policy.
What “in the app” would feel like: search/select/create in EmbeddedEra → tap Open in Google → Google Docs/Sheets app if the installed app and link handling permit, otherwise browser/editor handoff → return to EmbeddedEra. Universal links depend on the destination app, installation and user/browser choices; automatic return is not guaranteed. A home-screen PWA does not remove those boundaries. This is an integration inference requiring tests on the actual iOS Safari/PWA, Android and desktop devices. Apple universal links.
Authentication: recommend per-user OAuth with server-held refresh tokens and minimal scopes. Start with drive.file for documents created by or explicitly selected for the app. It does not authorize a complete search of everyone's Drive; broader discovery needs broader scopes/admin review. Match the connected Google account to the Access identity, and handle revoked tokens/reconnection. Google scopes.
A service account alone sees only resources granted to it; domain-wide delegation lets an administrator authorize impersonation of domain users for specified scopes. It is much broader than necessary for two people. It does not provide a human editor login, and the recorded key-creation block makes a downloaded-key design unsuitable. Keyless delegation is possible architecture, not an integration verified here. OAuth satisfies the token preference; an API key alone cannot access private Docs. Credentials, delegation.
Cost: continue the existing Workspace subscription; verify Drive/Docs/Sheets are enabled and both owners have appropriate seats. Standard Drive, Docs and Sheets API use currently has no additional charge. Their current pages announce later-2026 over-quota billing; do not promise unlimited free API use. At two-person volume, $0 incremental API cost is a planning assumption. Edition upgrades, added storage/seats, formal approvals/eSignature and exact future overage rates are unknown.
History: Docs/Sheets retain attributed version history, changes and restore; named milestones help prevent versions being merged. Google documents limits of 40 named versions per Doc, 15 per Sheet. This is not “every Save gets a permanent integer”: Google groups autosaves. Full history/name/restore workflow is documented for computer use; do not promise equivalent native-phone controls. History.
Drive's revisions.list can omit older revisions of frequently edited Google files. It is not a complete SharePoint-like archive or a turnkey comparison UI; use Google's editor for history rather than recreating it from that list. For approvals, retain a separately identified PDF/copy plus reviewer/date; named versions alone are not an immutable audit archive. Revision limitations.
Sharing: viewer/commenter/editor roles work inside and outside the organization, subject to Workspace admin and shared-drive restrictions. Non-Google recipients can use visitor sharing if enabled. Default to Restricted and chosen recipients, not domain-wide or anyone-with-link. Folder/shared-drive inheritance can grant access despite that intention: create private drafts in an unshared personal location, then explicitly share/move them. Administrators remain privileged. Roles, admin policy, visitors.
Google autosaves edits directly to the shared document. That differs from our Preview → Save → publish model: collaborators can see work in progress. For an approved customer release, send a PDF or separate approved copy; a normal shared URL follows the evolving document. Downloaded/printed copies cannot be revoked. Keep formal signatures/controlled SOP release outside this estimate.
PDF/print/offline: Drive files.export supports application/pdf, with a 10 MB exported-content limit on that method. The wrapper can download/share that PDF; Google phone apps offer print/export. Native Docs/Sheets apps support selected offline files and later synchronization; the wrapper/PWA does not inherit their offline editor. Desktop offline requires Google's supported setup, not Safari iframe caching. Export, iOS print, offline.
Ownership and risk: we control document access and can export portable DOCX/PDF/Markdown/XLSX copies, but editing, identity and full collaboration history depend on Google. Exports are not complete backups of comments, suggestions, grants and history. Main risks are account handoff friction, OAuth upkeep, inherited sharing, grouped history and service dependency. Company shared drives improve organizational custody where available; they are not a private-draft location by default. Shared-drive access.
C — hybrid: recommended destination
Use B's editor, history, sharing, PDF and offline capabilities for human documents. Retain A's small editor and Git integration for capture/project fields. This preserves existing investment while avoiding a second collaboration/export engine. Main extra risk: people editing the wrong copy; every item must show its source and one authoritative home.
| Content | Authoritative home |
|---|---|
| Proposals, customer deliverables, collaborative drafts | Google Docs |
| Calculations and collaborative spreadsheets | Google Sheets |
| Engineering notes, SOPs, decisions, agent-maintained records | Existing Obsidian Git vault |
| Quick private capture; project summary/Open to-dos | Existing lightweight app editor/vault |
| Assigned shop tasks and ERP transactions | Existing Kanboard/ERP; links from notes |
One Notes / Documents list merges Drive API results with the existing vault/Notes API; Drive API cannot list local vault files by itself. Show source, owner/audience and modified date. Recheck each user's access for list, search, preview and export; never put another user's document titles into a shared index. Google items open in Google; vault items continue to render/edit in the app.
Agents and Markdown: Google now officially supports Docs export as text/markdown (.md) through Drive export. Sheets exports include XLSX/CSV rather than Markdown; use Sheets API for structured values. This makes selected content readable to agents, but does not preserve Obsidian frontmatter, wikilinks, every layout feature, comments, suggestions or Google history as a lossless round trip. Supported export formats.
Prefer direct authorized API reads for private Google content. Where a file copy is useful, make one-way, read-only exports into a separately permissioned cache excluded from notebook publishing, recording source file ID/link, export time and content hash. A vault copy is appropriate only when all vault readers are allowed that content. Recheck permissions and remove stale cache access after revocation; a Git commit of private content is not easily recalled.
Do not run automatic two-way Docs↔Markdown sync: format loss, concurrent edits, deletions and permissions would recreate a substantial sync product. Agents continue writing canonical engineering Markdown as today. For a Google-owned document, an agent may propose edits via authorized Docs API use; do not push an edited cache file over human work. Promotion of a capture to Google should be explicit, retain the original, and record the new canonical link after successful creation.
Effort breakdown: B's 4–7 days covers OAuth, file list/Picker, template creation, sharing/export controls and handoff/device checks. C adds roughly 2–3 days for mixed-source navigation, pilot cleanup and a bounded agent-read/export proof: 6–10 days total, overlapping work counted once. Bulk migration, unattended sync, formal signatures and advanced approval workflows are excluded. Let Google handle review/history in its UI initially.
What to do with the pilot: keep existing drafts, versions, restore and project adapters. Add separate discard-draft and trash-note actions with clear confirmation/undo; discard must remove local and server checkpoints and prevent stale-device drafts reappearing. Preserve saved history according to a stated retention policy. Keep quick capture private; when contribution is needed, explicitly promote to Google. Retire only the planned full-office ambitions after the trial succeeds, not the owner's existing content.
D — existing alternatives worth checking
These are product-fit judgments from current upstream documentation, not device trials. Estimates cover a small evaluation/deployment, Google/Access identity, backup and basic integration; a seamless app wrapper is extra. Self-hosting retains custody but adds patching, restore tests and availability work. $0 means software fees only. Native-app/offline and private-sharing behavior must be demonstrated before adoption.
Nextcloud + Collabora / ONLYOFFICE — 5–10 days; $0 community licences, supported/mobile commercial costs unknown. Strongest alternative if embedded office editing is mandatory: office layouts, collaborative writing, PDF/print, internal grants and external links. Nextcloud has named/restorable versions with retention rules; WebDAV/API and office exports enable agent access, but are not a live Obsidian round trip. Files stay with us; authentication, office-server integration and backups become our responsibility. Sharing, versions.
Collabora supports a touch editor, but free CODE is a rolling development edition explicitly not recommended upstream for production. ONLYOFFICE Community lacks mobile web editing; free native mobile apps are a different workflow and leave our app. Check commercial terms before assuming a free iframe editor meets the phone rule. Too much operation for two people unless that rule outweighs using paid Workspace. Collabora FAQ, ONLYOFFICE mobile limits.
AFFiNE — 2–4 days evaluation; cloud Free $0, self-host commercial terms need confirmation. Rich documents/whiteboard, collaboration, Markdown/PDF export; mobile writing and external permission details need a real trial. Current pricing gives cloud Free three collaborators and seven-day history; lists Self-Hosted Team at $10/seat/month with ten-seat minimum ($100/month). The editor is largely MIT, but current backend terms are source-available/commercial: do not assume unrestricted free self-hosting. Exports help agents; native workspace storage adds lock-in. No clear advantage over existing Workspace for this job. Pricing/licensing, documents, PDF export.
Outline — 3–5 days; $0 community self-host core, licensed PDF edition cost unknown. Polished collaborative knowledge base with automatic history, user/group sharing and public links; responsive browser UX needs phone testing. API tokens and Markdown exports suit agents, but it owns a separate database. Native PDF export requires a licensed edition plus Gotenberg; browser printing is a fallback to test. BSL community restrictions and another service reduce the appeal. Licence, history, sharing, API, PDF.
Docmost — 3–5 days; $0 Community, Business $60/month minimum at annual billing. Collaborative rich-text wiki with comments, history/restore, spaces and public sharing; phone/browser ergonomics need testing. Markdown/HTML exports aid agents; dedicated PDF export was not established here, so test browser print. Current Business is $6/seat/month, minimum ten, and lists SSO, granular page permissions and API-key management. That cost/identity boundary is poor for two people; data stays on our server but in another application's database. Features/pricing, pages/history/export.
BookStack — 2–4 days; $0 MIT software. Strong free structured handbook: WYSIWYG, revisions, permissions, PDF/Markdown exports and REST API. Internal roles and guest/public permissions support publishing, but do not assume the same recipient-link collaboration as Docs. Phone writing, printing and concurrent-edit workflow need a trial. We own the database/files; exports reduce lock-in. Best wiki alternative, but moving SOPs from the established vault adds another source of truth. Features, permissions, exports/API, revision controls.
SilverBullet — 1–3 days; $0 open-source software. Closest to agent/vault ownership: plain Markdown, offline-first browser sync, token API and now documented Git-backed revision history. It is Markdown-oriented rather than Word-like; collaboration is experimental, and fine-grained private/customer sharing and polished PDF/print require evaluation or extra work. Lowest content lock-in, but a separate UI and sync/revision writer could conflict with existing vault automation. Good engineering-notebook candidate, weak replacement for customer documents. Overview, sync, history, collaboration, API auth.
Decision gate and evidence limits
Proposed trial: private creation → partner contribution → comment/suggestion → named milestone/history recovery → outsider access → PDF/print → offline edit/reconnect → return to EmbeddedEra. Include wrong-account/revoked-access cases and one Markdown export containing a table/image. Use both owners' actual phones plus desktop. If the editor handoff or desktop-only history tasks are unacceptable, revisit Nextcloud/Collabora; do not disguise that failure with a fragile iframe.
Research checked 2026-10-04 against linked upstream documentation and local notes; no service integration/device trial was performed. Only this proposal was written; no code, installs, migration, settings or sync/publication changes. Research/review ran in the primary Codex session; no helper models ran because allowance was unknown and enforceable read-only delegation was unavailable.
10-line summary
- Recommend the hybrid: Google documents plus the existing engineering vault.
- The live pilot is useful, but its missing delete/share controls matter.
- Keep and shrink it to private capture and project editing.
- Add draft discard and saved-note trash/undo before expanding it.
- Use Docs/Sheets for writing together, customer work and spreadsheets.
- Allow an explicit Google editor handoff; private iframe editing is unreliable.
- Keep Google OAuth separate from Cloudflare Access login.
- Keep SOPs and agent-maintained Markdown canonical in Obsidian/Git.
- Expect $0 extra standard API cost; verify seats, edition and future overages.
- Trial the workflow first; budget 6–10 engineering days if accepted.