layahost

layahost is an AI classification and routing API that returns decisions rather than text. It classifies each message as one of predefined options and gives probabilities you can act on, returning one answer per call and avoiding parsing free text. The API is hosted in the EU (Helsinki, Finland), and text is processed in memory, not written to disk, and not used for training.

[Uses and main capabilities]

• Content moderation: block insults, harassment, and threats before publishing. Example: "You are all idiots and I hope you get fired." → toxic 80%.
• Spam and scam filtering: block fake prizes, phishing, and get-rich-quick messages. Example: "Earn $5000 a day from home! Message me on WhatsApp." → spam 84%.
• Intent recognition: determine customer needs before human handoff. Example: "Please cancel my subscription at the end of this month." → cancellation 94%.
• Routing and distribution: send messages to the corresponding team based on a custom department list. Example: "The API returns 502 errors since this morning." → technical 82%.
• Priority and escalation: notify personnel immediately for outages, and wait until working hours for the rest. Example: "Our whole checkout is down and we are losing orders right now!"
• Language detection: supports 35 languages, routing or replying according to the customer's language. Example: "Dzień dobry, mam pytanie o fakturę."
• PII detection and redaction: identify names, emails, phone numbers, bank cards, and ID numbers, and return a redacted copy before storage or forwarding. Example: text containing a name and card number → personal data 99%.
• Prompt injection screening: detect before an AI agent reads web pages, emails, or tool output. Example: text containing hidden instructions → injection 78%.
• Sentiment tracking: track customer sentiment across tickets, reviews, and chats.

[How to use]

• Calling /v1/decide is one decision; when calling /v1/systemone,Each question counts as one decision. Rejected or failed requests (4xx, 5xx) are free.
• Compatible with the Jev API: the same /v1/systemone and /v1/models endpoints, request and response structures, errors, and request IDs. The official Python and JavaScript SDKs can be used directly; just point them to layahost and use a layahost key; Jev model names such as jev-latest are accepted. The models are different, so the probabilities are not exactly the same.
• It can also serve as a replacement for the Google Perspective API (shut down on December 31, 2026) and the OpenAI moderation endpoint (beta). A complete API reference and an MCP server for AI agents are provided.
• Decision flow: orchestrate spam checks, intent, routing, urgency, and more on a canvas, where each answer leads to the next question or result. A single request runs the entire flow, and only the questions on the path are billed. You can start from templates (such as support inbox triage, comment moderation, lead qualification), test with real text before publishing, and branch borderline answers to a human.
• Measurement: upload hundreds of your own texts with expected answers, run templates, custom questions, or an entire flow, and view accuracy, error locations, and thresholds suited to your data. CSV, TSV, or JSON Lines are supported, up to 2,000 rows, with two sample sets included.
• Migration: the official Jev SDK reads configuration through environment variables, and no other code changes are needed.

[Use cases] Suitable for any task that ends in a decision: content moderation, spam and scam filtering, customer intent recognition, ticket routing, priority escalation, language detection, PII redaction, prompt injection screening, sentiment analysis.

[Limitations and notes]

• Decisions are text-based only and do not process images; if there is an image, send its text description (such as a title, alt text, or OCR output).
• On some benchmarks, the Laya models lag behind Jev. Before switching production traffic, it is recommended to test on your own data and check the thresholds you depend on.
• The default is 600 requests per minute per key, with up to 32 questions and 32,000 characters per request.

[Billing and data]

• New accounts receive $1 in free credit after verifying their email (about 667,000 decisions), with no credit card required.
• Monthly plans start at $9 and include 1 million decisions; or prepay starting at $10, with 10% added for amounts above $100 and 500 dollarsand above adds 20%, with no subscription and credits that do not expire. Each million decisions costs $15 (prepaid); after monthly payment, each additional million decisions costs $10, and after the Pro plan, each additional million decisions costs $5. Decision fees are independent of text length.
• You can fine-tune on your own labels (such as product categories, internal teams, policy rules) to generate private Laya checkpoints, using the same API and SDK, visible only to your account. On Laya projects' typed decision benchmarks, fine-tuning improved accuracy from 36% to 77%.
• The servers are in Helsinki, Finland (EU). Logs retain decisions and metadata for 7 days; request text is retained only when enabled. DPA and company invoicing are supported.