Training outline — learn OmniLore as product workflow

Law: The best way to learn OmniLore is not reading operator architecture first.
Teach workflows people can run in Chat, then open the architecture only when someone needs to operate or debug the fabric.

In-product checklist: FIRST_30_MINUTES_OMNILORE.md

0. Learning goals (what "done" looks like)

After onboarding, a person should be able to:

1. State the 30-second story and the three pillars in their own words.
2. Pick the right Chat mode for product truth vs speed vs images vs code.
3. Complete 5–7 guided tasks without inventing features.
4. Create one private agent (preview → yes → new chat).
5. Name what OmniLore cannot claim yet (honesty).
6. Know which role track is next (member / developer / operator / public-law research).

Success metric (trainer): learner completes First 30 Minutes without switching to HOOTL/MCP/OpenBao slides.

1. Plain product story (teach first)

30-second story

> OmniLore is a private AI workspace for teams that need trustworthy answers from approved knowledge — and controlled help with documents, code, tasks, and internal tools.

Differentiator (one line)

> Control: answers are grounded in Source of Truth, agents get only approved tools, and the system fails closed when it cannot verify a source or permission.

Three pillars (only these in the first five minutes)

| Pillar | Plain language |
|--------|----------------|
| Trusted knowledge | Ground answers in approved sources — not empty search invention. |
| Permissioned agents | Small, task-specific agents with only the tools they need. |
| Private control | Org-controlled workspace; gates and membership rights. |

Do not open with

HOOTL · OmniGuard · MCP · OpenBao · dual-seat · Qdrant RF · Sisyphus · "700 tools" · CodeQL · clean_release_go.

Those are operator tracks, not first-day product literacy.

Source docs (story)

• PRODUCT_FRONT_DOOR.md
• WHAT_OMNILORE_IS_IS_NOT.md
• WHAT_OMNILORE_CAN_CANNOT_DO.md

2. What OmniLore can do (by audience)

Members (default track)

| Capability | What it means | Teach with |
|------------|---------------|------------|
| Trusted Q&A | Product/org questions grounded in approved docs | OmniLore mode + "show sources" |
| Chat modes | Right tool for the job | Mode picker table below |
| Private agents | Second Brain, help desk, research helper, … | AGENT_BUILDER.md |
| Personal corpus | Search your Gmail/Drive if linked for that user | Only if provisioned — do not promise every seat |
| Images | Create/edit/animate via local image plane (Comfy/FLUX) | Image mode + paperclip |
| Public-law workflows | Practice-shaped research; not legal advice | PUBLIC_LAW_WORKFLOWS.md |
| Scheduled jobs (v1) | Gated notes/briefs/capped prompts → workspace files | SCHEDULED_JOBS_CHAT_V1.md |
| Password / account | Member-safe identity tools when enabled | Help desk agent or password prompts |
| Sidebar libraries | Bookmarks, prompts, personal memory UI | Not fabric SoT |
| CoTrinity | VS Code (preferred) / Cursor programming assistant | COTRINITY_IDE.md |
| Desktop seat capture | Screenshot own enrolled PC from CoTrinity and/or remote Chat | DESKTOP_SEAT_CAPTURE.md |

Developers / power users

| Capability | What it means |
|------------|---------------|
| CodeHub | Private git / PR workflows under policy (Chat mode) |
| CoTrinity IDE | Local-first code help, review, tests, fabric tools when connected |
| Seat implementer | Optional desktop agent for capture under grants (not free shell) |
| RepoHub | Org forge — rights-based, not free write for everyone |

Operators (separate track — later)

| Capability | What it means |
|------------|---------------|
| Fabric health / routing | Live probes, not chat guesses |
| SoT discipline | Disk docs authority; vectors = discovery |
| Tool governance | MCP gated by role, mode, environment |
| Release honesty | Beta ≠ clean public release; gates stay on |

Full operator path uses ops SoT (PRODUCT_NARRATIVE_DUAL_SEAT, runbooks) — not day-1 member training.

3. What OmniLore cannot honestly claim yet

Teach this after the first win, not as the opener:

• Not unrestricted public SaaS GA / clean governance public release
• Not multiparty HSM complete for clean public claims
• Not autonomous merge/deploy without gates
• Not legal, medical, or financial advice
• Not infinite scale — hardware and model limits matter
• Not "all tools for every user" — permissions are intentional
• Not a free shell / unrestricted scheduled autonomy
• Not a default local MiniMax-H3 / frontier video HQ product path (stills use FLUX image plane)
• Not a guaranteed Chat popup when a schedule fires (v1 → workspace files)

4. Pedagogical sequence (recommended workshop)

Phase A — First 5 minutes: story only

1. Headline + 30-second story.
2. Three pillars.
3. One line on fail-closed.
4. Stop. No architecture slide.

Phase B — First hands-on: OmniLore mode (10 minutes)

Learner opens Chat → mode OmniLore → new chat:

| # | Prompt | Lesson |
|---|--------|--------|
| 1 | What can I do in OmniLore? | Feature discovery via tools |
| 2 | What is OmniLore, and show sources? | Grounded product truth |
| 3 | (Optional) Switch to Quick, ask the same product question | Feel the difference / redirect |

Core rule: product/org/fabric truth → OmniLore mode. Quick is fast general chat without fabric tools.

Phase C — Mode picker (5 minutes)

| Mode | Teach users to use it for | Avoid for |
|------|---------------------------|-----------|
| OmniLore | Real product/org answers, tools, sources, fabric questions | Nothing "wrong," but CodeHub is cleaner for multi-file PR sprints |
| Image | Create / edit / animate images (paperclip) | Product definition |
| Quick | Fast general chat/code, no fabric tools | "What is OmniLore?", dual-seat, completeness |
| CodeHub | RepoHub/git/code/PR workflows | Product encyclopedia |
| Advanced · Models | Fixed model testing, no tools | Product truth |

Deep reference: USER_GUIDE_CHAT_OMNILORE_AI.md.

Phase D — Guided starter tasks (15–20 minutes)

Pick 5–7 from this menu (adapt to role; do not force personal Google if not linked):

| # | Task | Mode | Pass criteria |
|---|------|------|----------------|
| 1 | Grounded product question + sources | OmniLore | Answer cites docs / tools; not pure invention |
| 2 | "What can I do?" or features list | OmniLore | Mentions real capabilities without overclaim |
| 3 | One Image create or simple edit | Image | Job completes or honest queue/fail message |
| 4 | Create one private agent (preview first) | OmniLore | Preview → yes → agent appears under Agents |
| 5 | List my agents / open new chat on that agent | OmniLore / Agents | Learner switches agent intentionally |
| 6 | One scheduled note (optional; if gate allowed) | OmniLore | Returns schedule_id + next_run_at |
| 7 | Password / account question or public-law "not legal advice" framing | OmniLore | Correct mode + honest limits |
| 8 | (Developers) CodeHub: list my repos / open a PR draft path | CodeHub | Policy-respecting forge action or clear deny |
| 9 | (Developers) CoTrinity: one code review or doc question | Desktop | Local help works; Cursor account wall understood |

Phase E — Create one private agent (required for most tracks)

Best starter prompts (from Agent Builder):

Build me a private Second Brain for my Gmail and Drive.
Preview first — I'll say yes when it looks good.

Create a help desk agent for password and account questions.
Preview first.

What personal agents can you create for me?

Pattern to name out loud: small, task-specific, permissioned agents — not one giant assistant with every tool.

Phase F — Role tracks (after First 30 Minutes)

| Track | Focus | Source docs |
|-------|--------|-------------|
| Member | Chat, modes, agents, sidebar, schedules, images | FEATURES_USER · FIRST_30_MINUTES · SCHEDULED_JOBS |
| Developer | CoTrinity + CodeHub + local-first coding | COTRINITY_IDE · chat user guide CodeHub |
| Operator | Fabric health, SoT, release honesty, gates | PRODUCT_NARRATIVE (ops) · runbooks — not first for members |
| Public-law research | Public Law workflows; strict not-legal-advice | PUBLIC_LAW_WORKFLOWS · KNOWN_LIMITATIONS |

5. Trainer anti-patterns

| Anti-pattern | Do instead |
|--------------|------------|
| Start with dual-seat topology | Start with Chat modes and a product question |
| Demo 50 MCP tools | Demo one mode + one agent |
| Promise H3 / unlimited video | Honest image plane limits |
| Skip "cannot claim" | One honest slide after first win |
| Train on Auto for product truth | Force OmniLore mode for SoT questions |
| Create agent without preview | Always preview → yes |
| Treat sidebar Memory as fabric SoT | Call it personal convenience UI |

6. Suggested session formats

| Format | Length | Content |
|--------|--------|---------|
| Lightning | 15 min | Story + OmniLore mode ×2 prompts |
| First 30 Minutes | 30 min | FIRST_30_MINUTES_OMNILORE.md |
| Half day | 2–3 h | First 30 + agent deep dive + role track |
| Operator day | Full day | After product literacy: SoT, health, gates, release board |

7. Source doc map (for curriculum authors)

| Need | Doc |
|------|-----|
| 30s story / pillars | PRODUCT_FRONT_DOOR.md |
| First login | GETTING_STARTED.md |
| Modes & UI | USER_GUIDE_CHAT_OMNILORE_AI.md |
| Agents | AGENT_BUILDER.md · USER_GUIDE_CREATE_CUSTOM_AGENTS_CHAT.md |
| Can / cannot | WHAT_OMNILORE_CAN_CANNOT_DO.md · FEATURES_USER.md |
| Limits | KNOWN_LIMITATIONS.md · STATUS.md |
| Schedules | SCHEDULED_JOBS_CHAT_V1.md |
| Desktop IDE | COTRINITY_IDE.md |
| Ops product truth | docs/PRODUCT_NARRATIVE_DUAL_SEAT.md (after product literacy) |

8. Best next product lifts

1. Keep First 30 Minutes linked from Chat help and starter chips.
2. Optional: Chat starter chips that map 1:1 to checklist steps.
3. Role-specific "Day 2" one-pagers (developer / operator) — only after First 30 is green.

Do not replace this with another long architecture document for first-time learners.

End training outline.