# AI Training — Session 1 Handout & Pre-Work **Session:** One Brain, Two Hands: Meet the AI Teammates We Actually Use **When:** Wednesday 18 November 2026, 8:00–8:45 PM UTC **Preview (invite-only):** Wednesday 28 October 2026, 8:00–9:00 PM UTC — https://meet.google.com/rup-fsrs-vub **Local:** Boise 1:00 PM · New York 3:00 PM · London 8:00 PM · Berlin 9:00 PM · Johannesburg 10:00 PM · Sydney 7:00 AM (Thu) **Join:** {{MEETING_URL}} **Register:** {{REGISTRATION_URL}} **Notes:** `docs/ai-training-session-notes.md` --- ## 0. What you'll leave with - A clear answer to "which AI should I ask for what" — so you stop paying for the wrong one. - One task you keep doing by hand, handed off before the hour ends. - A one-page map of your own AI stack, free tiers included. - The recording, the notes, and the prompt templates we use. - A door into the next session — not a sales call. **Why we run this:** the same AI stack that runs the session runs the Resonance Tunes product — diagnostic tools for individuals (The Seeker) and teams (The Collective). Seeing the setup is the honest way to see what we build: https://www.resonancetunes.com --- ## 1. What to bring - A laptop and charger. - One real task you keep doing by hand (a report, a post, a meeting summary, a follow-up email). Write it as: "I want AI to ______." - Nothing else. You can attend and watch without installing anything. --- ## 2. Pre-work (10 minutes) **The live trainer is the training page. Do not schedule a setup call.** 1. **Open [https://www.resonancetunes.com/training/](https://www.resonancetunes.com/training/)** — the four setup steps are on that page, with every command and click, including the DeepSeek key and `/connect`. (5 min) 2. **Write your one task** in a sentence: "I want AI to ______." (2 min) 3. **Optional — Notion** — `https://notion.so`. Free plan. If you want your own notes connected during the session, have one page you're willing to open. (3 min) 4. **Optional — Fathom** — `https://fathom.video`. Free tier. If you want meeting capture set up, install the note-taker before the session. (3 min) Nothing else is required. No transcript or recording preparation. You can attend and watch without installing anything. --- ## 3. Tools: what is free, what is paid | Tool | What it does | Cost | Needed? | Where | |---|---|---|---|---| | **Google Meet** | The session room | Free | Required | Browser | | **opencode** | Agent for operations: files, servers, scripts, research, automation | Open-source; you pay only for model tokens | Optional to follow along | `https://opencode.ai` | | **Claude** | Shown briefly as the handoff partner for Notion/Fathom | Free tier | Optional | `https://claude.ai` | | **DeepSeek API** | Cheap model behind opencode for extraction, drafts, logs, batch work | Pay-per-token; a **$5 top-up** is plenty for this session | Optional | `https://platform.deepseek.com` | | **Tavily** | Web research: search, extract, crawl | Free tier ~1,000 credits/month; paid above | Optional | `https://tavily.com` | | **PostHog** | Product analytics: events, funnels, retention | Free tier | Optional | `https://posthog.com` | | **rclone** | Google Drive/Photos sync to local folders | Free | Optional | `https://rclone.org` | | **Ollama + Mistral 7B** | Local/private model | Free | Optional | `https://ollama.com` | | **Notion** | Your notes and documents, readable by the AI | Free plan | Optional | `https://notion.so` | | **Fathom** | Meeting capture and transcript → tasks | Free tier | Optional | `https://fathom.video` | **No purchase is required to attend.** Every tool used in the session has a free tier or is open source. opencode is open source; you pay only for the model tokens you choose to use. --- ## 4. If you choose to pay for anything (optional) Only do this if you want to follow along with a paid tool. Keep receipts. **Claude Pro — $20/month** 1. Sign in at `https://claude.ai`. 2. Settings → Billing → Upgrade to Pro. 3. Pay by card. Confirm the current price and taxes at checkout. **DeepSeek API — prepaid, pay-per-token** *Same account as the free chat, separate prepaid wallet.* The chat at `chat.deepseek.com` is free. The API is billed per token from its own balance — the chat plan adds nothing to API credit, and topping up the API does not change the chat. 1. Open `https://platform.deepseek.com` and sign in. Try your existing chat login first; if it is not accepted, sign up with the same email. 2. **Top up first.** Left sidebar → **Top up** (some accounts show it under **Billing**) → add **$10** — a light month of opencode use is $5–$20. Pay by card and confirm the balance appears on the Billing page. - Skip this and the key still works, but every call fails with **Insufficient Balance**. That is a wallet problem, not a key problem. 3. **Create the key.** Left sidebar → **API Keys** → **Create new API key**. Name it for the device (e.g. `rebekka-mac-opencode`). Copy it immediately — it starts with `sk-` and is shown only once. 4. **Save a copy in a password manager**, then add it to opencode with `/connect`. 5. **Connect it to opencode** (do not paste it into a config file): open Terminal → run `opencode` → type `/connect` → choose **DeepSeek** → paste the `sk-...` key → press Enter. 6. **Verify:** type `/models` and choose the listed DeepSeek Flash entry (the older `deepseek-v4-flash` ID may be retired). Ask one short question. If it answers, the key and the balance are working. 7. **If it fails:** - `Insufficient Balance` → add credit; the key is fine. - Authentication error → the key was copied wrong or deleted; create a new one. - `Model not found` → run `/models` and pick the entry that is listed. 8. **One key per person and device.** Never share another operator's key. A key is money; revoke it if a laptop is lost or the key is exposed. A key is money: anyone who has it can spend the credit. Revoke it immediately if a laptop is lost or the key is exposed. **Tavily — free tier first** 1. Sign up at `https://tavily.com`. 2. Create an API key. The free tier is enough for this session. 3. Add a card only if you choose to upgrade. No card is needed for the free tier. **Ollama, rclone, Notion, Google Meet:** no payment. **Public attendees:** pay for your own optional tools. You do not need any of them to attend or to get the recording. --- ## 5. How to use this during the session - **Following along?** Keep this handout open and do the same steps in your own window. If an install slows you down, stop and watch; the recording and handout will be here after. - **Just watching?** Write your one task and one question in the chat. That is enough. - **Demo data only.** We use fake contacts and a sandbox repo. Do not paste client, investor, or personnel data into any tool during the session. - **Ask in the chat.** the moderator collects questions; we answer the ones that apply to the most people first. - **One action before you leave.** Write down the single task you will hand off this week. --- ## 6. Token and cost discipline The goal is not to use less AI. It is to stop paying for the same context twice. 1. **Write context once.** Put your brief, voice, and examples in one doc or queue. Link it; don't re-paste it into every chat. 2. **Route by task.** Premium model for judgment, brand voice, and final copy. Cheap model for extraction, summaries, logs, and first drafts. Local model for anything private. 3. **Summarize once.** Turn a transcript into a durable summary and work from the summary, not the raw transcript. 4. **Batch.** Five related tasks in one well-scoped prompt beat five separate chats. 5. **Cap the output.** Ask for bullets, a table, or "under 200 words." Long outputs cost more and get ignored. 6. **Search before you read.** Use search/extract on the specific source instead of dumping whole files or sites into the model. 7. **Use the free tiers first.** Tavily free credits, Groq free tier, Gemini free tier, Pollinations for images, Ollama for local. 8. **Set a spend alert.** Put a small top-up on DeepSeek/Tavily and check the dashboard after the session. If a tool has a usage page, look at it once a week. **Rule of thumb:** if you would not pay a human for the task, don't pay a premium model for it. --- ## 7. Session 1 agenda (45 minutes) | Time | Segment | What happens | |---|---|---| | 0:00–0:03 | Welcome | Framing, recording consent, "one idea, one live build, one action" | | 0:03–0:08 | Install opencode | Everyone leaves with it running, or watches the install | | 0:08–0:16 | opencode as the operations agent | Files, research, writing, automation — the work, not the theory | | 0:16–0:22 | The routing rule | Premium vs cheap vs local: judgment, plumbing, privacy | | 0:22–0:30 | Notion connected | Your own notes become context; stop re-pasting | | 0:30–0:38 | Fathom connected | Meeting → transcript → summary and action list | | 0:38–0:42 | Resonance Lens + youth sessions | RT's team layer; youth sessions coming soon | | 0:42–0:45 | One action + next date | Write your one task; next session; register link | **Additional AI topics in this session (introduced, not fully demoed):** - Agentic AI vs chat AI: plan → act → verify. - Model routing and cost control. - Local models and data sovereignty. - Research with sources and citations. - Meeting capture and transcript-to-action. - Shared prompt queues and second-brain context. - Automation (n8n, Brevo, CRM) — preview for the December/January sessions. - Ethical AI: consent, anonymization, EU hosting, Mistral/Zhipu positioning. - Text-to-game and youth sessions — preview for the July session. --- ## 8. Security and privacy - Give the key to your AI through `/connect` or a local env file. Keep it off public pages, slides, and shared docs. - Use environment variables or the tool's auth screen. - Rotate keys before any public session or screen share. - Use demo contacts and a sandbox repo. Do not show real client, investor, CRM, or personnel records. - Recording consent is stated at the start. If you share something sensitive, say so and we will stop recording for that part. --- ## 9. After the session 1. Do the one task you wrote down. Only that one. 2. Watch for the recording and the prompt list by email. 3. Bring the result (or the failure) to the next session. 4. Next session: **Wednesday 16 December 2026, 8:00 PM UTC — Connect Your World: Files, Drive, and a Single Prompt Queue.** --- ## 10. Links - Full series overview: `docs/ai-training-series-overview.md` - Program calendar and invite copy: `docs/ai-training-sessions.md` - Living session notes: `docs/ai-training-session-notes.md` - Strategic partner ecosystem doc: `docs/strategic-partner-ecosystem-sessions.md` --- ## 11. Optional — instrument one action If you want to leave with one tracking loop that actually works: 1. **Pick one decision** — a pilot request, a completed reading, a social click. Not a pageview. 2. **Name it in the past tense** — `pilot_requested`, `reading_completed`, `social_click`. 3. **Add it once, in one shared tracker** — not copy-pasted into every page. 4. **Add only the properties you will use** — `tool`, `tier`, `source`, `page`. No names, emails, birth data, or free text. 5. **Test in a real browser** — headless browsers are often filtered as bots. Do the action, watch the request. 6. **Check it in PostHog within 60 seconds** — if it is not there, it did not fire. 7. **Bump the cache** — a shared file can live in browsers for a week; change the `?v=`. 8. **Write one line in the event map** — event, where it fires, properties, owner. **Event map — five lines to keep** | Event | Fires when | Properties | Owner | |---|---|---|---| | `pilot_requested` | Pilot form submitted | `source`, `team_size`, `tool` | Growth | | `reading_completed` | A reading reaches the result | `tool`, `tier` | Product | | `social_click` | Someone clicks out to a social profile | `network`, `page` | Marketing | | `signup_completed` | A new account is created | `method` | Product | | `purchase_completed` | Stripe confirms payment | `tier`, `amount` | Finance | **Do not** - Put a personal API key in the page. The write key is public; the read key is not. - Track names, emails, birth data, or anything you would not put on a slide. - Trust a headless test. Use a real browser; PostHog filters bots. - Count pageviews as demand. A view is not a decision. **What needs a human** - The canonical funnel name: `assessment_started/completed` or `interpretation_started/viewed`. - UTMs in every social link. The tracker can read them; the post has to include them. - Native social metrics. Likes, comments, and shares need platform API access, not a JavaScript tag.