AI Without Limits on the Appliance

The self-hosted Glassy appliance (INSTANCE_ID=self_hosted) handles AI differently from Glassy Cloud. There are no cost caps, no metering, and no system cloud keys. This lesson explains what that means and how to set up AI on your appliance.

No metering, no caps

On Glassy Cloud, AI usage is metered per tier with a monthly cap. On the self-hosted appliance, the metering system is a no-op — deductAiCredits() does nothing. There is no credit ledger, no cap, and no billing.

This means:

• Browser WebGPU AI — always free, runs in your browser (same as cloud)
• Ollama — always free, runs on your server’s hardware
• BYOK — you pay your provider directly, with no Glassy-imposed limit

System cloud keys are disabled

The appliance does not load Glassy’s managed cloud AI keys (GEMINI_API_KEY, OPENAI_API_KEY, ANTHROPIC_API_KEY) from the .env file. Even if you set them, they’re ignored.

On startup, you’ll see this log message:

[AI Providers] Self-hosted — cloud system keys disabled.
Use BYOK (Settings → API Keys) or local Ollama.

This is intentional — it keeps your AI queries from routing through Glassy’s cloud provider accounts. Your data only goes where you explicitly send it.

Your two paths to cloud-capable AI

Since system keys are disabled, you have two options for AI that exceeds what the browser model can handle:

Option 1: Ollama (recommended — fully local)

Install Ollama on your host or use the bundled sidecar (see the previous lesson). Ollama runs on your hardware, no API costs, no data leaves your machine. This is the most private option.

Option 2: BYOK (bring your own cloud key)

Add your own OpenAI, Anthropic, or Gemini API key via Settings → API Keys. Keys are encrypted at rest with your API_KEY_ENCRYPTION_KEY. See BYOK: Your Own API Keys for setup.

You can use both — Ollama for everyday tasks, BYOK for when you need a specific cloud model.

The NO_AI_PROVIDER error

If you attempt a cloud AI task and no provider is configured, Glassy returns:

402 { code: 'NO_AI_PROVIDER', error: 'No AI provider configured.
Add your own API key in Settings → API Keys, or run a local Ollama model.' }

This is not a billing error — there’s no credit copy, no upgrade suggestion. It simply means you need to configure a provider. Browser WebGPU AI (embeddings, small tasks) still works regardless.

What still works without any provider

Even with no Ollama and no BYOK keys, these features work via browser WebGPU:

• Voice transcription (Whisper in-browser)
• Embeddings for Related Notes and Smart Tags
• Small text generation (Qwen 0.5B-3B via WebLLM)
• Semantic search in the Knowledge Base

You only need Ollama or BYOK for larger, more capable text generation.

What’s next?

You’ve completed the Local AI track. Explore the rest of the self-hosting experience in the Self-Hosting track — deployment, configuration, backups, and upgrades.