Text Embeddings_

Appwrite generates text embeddings with built-in models, so you can turn text into vectors without running a separate embedding service. Embedding usage is metered in tokens, and each model is priced and counted separately.

How embedding usage is counted

Every call to createTextEmbeddings sends one or more strings to a model. Appwrite counts the tokens in those strings and records them against the model that produced the embeddings.

How it works:

1. Generate embeddings through the Appwrite SDK or API.
2. Appwrite tracks the token count per model, not per request.
3. Token counts are aggregated per project and rolled up to your organization.
4. Allowances are refreshed monthly on your billing cycle.

Each model keeps its own counter. Embedding 2M tokens with nomic-embed-text and 2M tokens with all-minilm uses 2M against each model's allowance, not 4M against a shared pool.

Pricing

Rates are per model, since models differ in cost to run.

Free plan

• Included: 1,000,000 tokens per model per month.
• Overage: Not available. createTextEmbeddings returns a 402 error once the cap is reached. Only embedding creation is blocked, and the rest of your project keeps working.

Pro and Scale plans

• Included: 10,000,000 tokens per model per month on Pro and 100,000,000 tokens per model per month on Scale.
• Overage: Billed at the model's rate, rounded up per started million above the allowance.

Enterprise plan

• Included: Custom allowances.
• Overage: Custom rates.

How overage is calculated

The included allowance is a ceiling, not a floor. Usage inside it is covered by your plan price and adds nothing to your invoice, so 5M tokens against a 10M allowance bills $0 for embeddings.

Above the allowance, Appwrite bills each started million at the model's rate:

overage = ceil((tokens - allowance) / 1,000,000) x rate

For example, a Pro organization that embeds 12M tokens with nomic-embed-text in one cycle is billed ceil((12M - 10M) / 1M) x $0.10 = $0.20. Each model is calculated independently against its own allowance and rate.

Overage on paid plans counts toward your budget cap, and embedding creation is never hard blocked on a paid plan.

For detailed information about the different pricing options and features, please visit the pricing page.

Best practices

To keep embedding costs predictable:

1. Pick the smallest model that works: all-minilm costs half as much per token as nomic-embed-text and stores 384 values per vector instead of 768.
2. Embed once, store the vector: reuse the stored embeddings field instead of regenerating it on every search.
3. Batch your calls: pass several strings to one createTextEmbeddings call to cut request overhead.
4. Monitor usage: track embedding tokens per model in the Appwrite Console.
5. Set a budget cap: bound overage on Pro and Scale before it surprises you.