OpenAI API
First met in aicp-c3-basecamp-02 · Models & model access
OpenAI APIFirst met in aicp-c3-basecamp-02 · Models & model access
Direct API access to OpenAI's models — the canonical example of buying credits and pasting a key into something you built.
A generated app needs its own model access and you want the vendor relationship direct rather than routed.
The teaching example for the whole API-key pattern: create key, top up credits, paste into the app
Cost reality check — the BC2 AI project generator's whole session cost 3-4 cents
Serves OpenAI models to your own applications with a per-token bill
Powers features in vibe-coded apps that need an LLM
Doesn't come free — but the amounts are small enough that fear of cost is usually misplaced
Doesn't belong in front-end code; keys go server-side (Supabase edge functions or equivalent)
Anywhere your own software needs a model and you've settled on the vendor.
Models & model access
The pipeline calls no model APIs directly.
The 3-4 cent figure is a useful anchor when estimating any future API-backed feature — the cost objection is usually theoretical.
current
openrouter: one key for many models vs a direct vendor relationship
supabase: edge functions keep the key off the front end
| Session | Coverage | Moment | What happened |
|---|---|---|---|
| Basecamp 2: Vibe Coding | demonstrated | 1:22:56 | Key creation in console, credits top-up, powers the AI project generator; total spend for the session's feature ~3-4 cents (2:29:42) |
| Basecamp 3: Introduction to n8n — I | demonstrated | 0:37:53 | Credential setup walkthrough as the canonical API-key example |
| Basecamp 4: Introduction to n8n — II | demonstrated | 1:36:01 | Agent LLM; agent-prompt-vs-chat-input error debugged live; picked for high thinking |
| Session 19: Content Strategy Pipeline — the YouTube Growth Stack, Built by an Agent Team of Chats | demonstrated | Keys created live; realtime-voice rejections all session; chained Transcribe + 4o-mini fallback | |
| Session 20: Complete LinkedIn Agency Automation — Carousels + Newsletter from One Content System | demonstrated | Responses API structured output + GPT Image; Azure-variable hijack fought off with explicit env isolation; 429 throttling diagnosed live | |
| Session 22: Automate Admin Tasks — Advanced RAG (Enterprise 'Internal Perplexity' Built Live) | demonstrated | Demo LLM + embeddings; stale-key failure diagnosed live | |
| Basecamp 6: Voice Agents | explained | 1:07:52 | Retell's suggested models; counterintuitive tested finding: 4.1 beats 5.1 for voice work |
| Session 18: Nano Banana + Veo 3.1 Multi-Shot Video — UGC Genie, an AI-UGC Product Built Live | explained | Magic-wand prompt-enhancement endpoint built in, key deliberately not added on stage | |
| Workbooks 1–5: Guided Builds with Completion Proof | explained | Workbook 5's model access: key handling doctrine, small funded balance, usage limits, JSON output mode |