Intellect (AI)

Intellect runs AI on infrastructure you pay for. Live mode is always-on; Batch mode is on-demand Spot. Both are fixed costs — you know what you'll pay before you deploy.

Add Intellect if:

• Visitors are sending messages your team can't keep up with
• Visitors ask specific, answerable questions about your products, menu, or services
• You want AI to process content in the background (summarize, translate, extract)

Don't add Intellect if:

• The chatbox is sitting idle most of the time
• Your questions are simple enough for an FAQ page or accordion
• You only want batch processing — Batch mode is significantly cheaper than Live

Start with Core's static chatbox. Add Intellect when you have real volume that justifies the infrastructure.

Ma's Bakery started with a simple contact form — visitors filled it out, Ma responded via email within a day. It worked fine until the bakery went viral on a local social media post.

Overnight, Ma was getting 80–120 chatbox messages a day. Half of them were the same questions:

• "What's fresh today?"
• "Do you have anything gluten-free?"
• "Can I order a custom cake?"
• "What are your hours on Sunday?"

Ma added Intellect. She wrote a system prompt that introduced the AI as her bakery assistant and gave it access to the week's menu and her FAQ. The AI now handles the first 90% of conversations — the routine ones. When a conversation needs Ma, the AI escalates to her email.

Pa's Donuts hasn't added Intellect — their chatbox gets maybe 3–5 messages a week.

Cousin's Coffeehouse uses Intellect in Batch mode only — they run a nightly job that reads that week's Coffee Notes blog post and generates a newsletter blurb. No always-on instance.

Core ships a ChatboxWidget that works as a contact form out of the box. When Intellect is provisioned and OLLAMA_HOST is configured, the chatbox automatically upgrades to a live AI assistant.

Without Intellect: visitor messages are submitted as a contact form.
With Intellect Live: visitor messages get real-time AI responses, streamed from your Ollama instance.

No code changes required — the upgrade is automatic when the env var is set.

The server-side chatbox route can inject context from your own data before sending a message to the model.

Ma's Bakery example: The chatbox route reads Ma's current menu and injects it into the system prompt before every conversation. When someone asks "what's fresh today?", the AI reads the actual menu — not its training data.

If Commerce is provisioned, the chatbox can check your Products table and respond with live inventory:

"We have 6 of those almond croissants left today — would you like to start an order?"

The AI only accesses data you explicitly provide to it. You control what it can see and what it's allowed to say.

The BatchRunner handles background AI workloads that don't need real-time responses:

• Nightly content summarization (Cousin's Coffeehouse blog → newsletter blurb)
• Bulk entity extraction from uploaded documents
• Page translation into other languages
• Structured data generation from unstructured input
• Weekly reporting from Ocean pipeline data

Batch uses Spot EC2 instances — spun up on demand, terminated when done. Significantly cheaper than an always-on Live instance.

Intellect runs in one of two modes. The mode determines the cost model and what the instance can do.

Live provisions an always-on EC2 instance at your chosen tier. It handles real-time chatbox traffic continuously, and any batch jobs you queue run on that same instance. No additional infrastructure is needed for background processing if you already have Live running.

Batch provisions Spot EC2 on demand. An SQS trigger spins up an instance, runs the job, and terminates it. This is background-only — no real-time chatbox. The right choice when you want AI capability without the cost of an always-on instance.

If you have Live, you do not need Batch. The instance is already running.

Tiers:

Tier	Specs	Best for
Lean	CPU-only or small GPU (Gemma 2B)	Batch-only deployments, dev/eval
Small	Entry GPU, 7B-class models	Single-user chatbox or light batch
Medium	Mid GPU, 13B-class models	Moderate chatbox traffic
Large	Performance GPU, 30B-class models	High-traffic chatbox, multi-user
Max	High-end GPU, 70B+ models	Enterprise scale

Your tier choice drives your solvency cap directly. A larger instance means more infrastructure cost, which means more ad impressions running your publisher ID before SirSluginston Co earns anything above that line.

Intellect ships only OllamaProvider. There are no wrappers for OpenAI, Anthropic, Bedrock, or any other cloud AI provider.

The InferenceProvider interface is open (SLIME source-available) — if you want to wire in a cloud AI provider yourself, the interface is there. That integration, and any token costs, are yours to manage.

Why? Cloud AI costs are unpredictable, don't belong in the solvency cap formula, and vary per call. Intellect only works cleanly with infrastructure you own and pay for at a fixed rate.

Intellect is selected during adapter init:

npx @sirsluginston/aws-adapter init
# ...
? AI / Intellect?
  ❯ Live (always-on EC2 — handles chatbox and batch)
    Batch (Spot EC2 — background jobs only, no chatbox)
    No

Then select a tier (Lean → Small → Medium → Large → Max).

Once provisioned, set your model:

OLLAMA_HOST=http://your-ec2-private-ip:11434
OLLAMA_MODEL=mistral

The chatbox upgrades automatically when OLLAMA_HOST is set. No code changes required.

Intellect costs depend on mode and tier. Everything runs through your solvency cap — your ad publisher ID covers the full infrastructure bill before SirSluginston Co earns anything above it. A larger instance means a higher cap, which means more impressions running your publisher ID. The AI earns its keep.

Mode	Tier	Approximate monthly
Batch (Spot)	Lean	~$5–20
Batch (Spot)	Small	~$20–60
Live	Lean	~$30–80
Live	Small	~$100–200
Live	Medium	~$200–400
Live	Large	~$400–800
Live	Max	$800+

Batch rows apply to deployments without a Live instance. If you have Live, batch jobs run on that instance — no separate cost.

Actual numbers vary by AWS region and current EC2 pricing. The Cost Explorer Lambda updates your solvency cap within roughly 24 hours of any instance change.

Intellect + Ocean — The chatbox can query Lighthouse snapshots from your Ocean pipeline to give live, accurate answers. Ma's AI knows today's prices because it reads from Ocean's Lighthouse cache — not from training data.

Intellect + Commerce — The chatbox can check your Products table to answer inventory questions and guide visitors toward checkout. Requires both Intellect and Commerce.

Intellect + Core (Batch only) — Batch mode can process your Core page content — summarizing new blog posts, generating social copy, translating Rich Text blocks. Requires only Core + Batch Intellect.