What SchneeAI is
Is SchneeAI an LLM observability tool?
No. SchneeAI includes observability — structured usage events, cost dashboards, latency tracking — but the harder problems it solves are per-tenant isolation, credit and budget enforcement, raw retention (Vault), PII scanning, audit, and dataset building. If you only need observability, Langfuse or Helicone are good choices. If you need a platform to operate AI in production with governance, SchneeAI is built for that.
How is SchneeAI different from LiteLLM?
LiteLLM is a provider proxy — it handles the connection to OpenAI, Anthropic, Gemini, Groq, and others. SchneeAI uses LiteLLM as its provider connection layer. What SchneeAI adds on top is the platform: credits, budgets, prompt registry, Vault, audit, PII scanning, consent, and dataset building. LiteLLM is a building block; SchneeAI is the platform around it.
How is SchneeAI different from Portkey?
Portkey is a gateway-focused product with strong prompt management and observability. SchneeAI goes further on the operating side — credits, budgets, PII scanning on raw payloads, Vault retention with legal hold, consent management, and dataset building for fine-tuning. If you need a gateway, Portkey is a strong choice. If you need to operate AI across multiple services with full governance, SchneeAI is built for it.
Data and security
How does PII scanning work?
SchneeAI scans prompts and reasoning content before they leave your perimeter, across 17 categories — including credit cards (with Luhn check), Japan Mynumber (with check digit), US SSN, API keys (Google, OpenAI, Anthropic, Stripe, Slack), JWTs, PEM private keys, and connection strings. Findings are classified by severity and acted on per policy: flag, mask, or block. Critical findings block the call entirely. Read the technical details in PII scanning in production.
Do you store raw prompts and responses?
Yes, optionally, in the Vault — an encrypted raw blob store with retention windows, legal hold, and access audit. Retention is configurable per service, tenant, or feature. You can disable raw retention for high-sensitivity workloads while keeping structured usage records for billing and observability.
Do you train models on customer data?
No. We do not train models on customer prompts or responses. Raw payloads in the Vault are encrypted at rest with customer-scoped keys, and access is logged. Provider calls inherit the provider’s data handling — SchneeAI’s role is to give you the controls to keep sensitive content out of provider pipelines when your policy requires it.
Where is data stored?
Structured metadata (usage events, cost records, audit logs) lives in PostgreSQL. Raw prompts and responses live in object storage (S3/R2-compatible) with application-layer encryption. Both are scoped by service, tenant, and user identity. See the Sub-processors page for the current list of infrastructure providers.
Providers and routing
Which LLM providers are supported?
OpenAI, Anthropic, Google Gemini, and Groq are the primary providers, accessed through LiteLLM. Adding a new provider is a configuration change. Routing strategies let you pick models per feature, with failover when a provider is unavailable.
Can I bring my own provider keys?
For enterprise customers, yes. The default model uses provider keys managed by SchneeAI, with credits drawn down based on actual provider cost. Bring-your-own-key is available for teams that have existing provider commitments or need direct provider billing.
Pricing
How does pricing work?
SchneeAI uses a credit-based model — you commit to credits in advance, credits are drawn down based on actual Gateway usage, PromptOps volume, and Control Plane seats, and per-tenant / per-feature budget counters keep spend predictable. Public tier pricing is still being shaped with design partners; see the pricing page for current status and design-partner terms.
What happens when budgets are exhausted?
Budgets are a first-class concept in the platform. Per-service, per-tenant, and per-feature budgets can be configured with threshold actions (alert, throttle, block) so the platform fails safe when a limit is reached. Default behavior is alert-then-block, with operator override.
Getting started
Can I self-host?
Self-hosting is on the roadmap for enterprise customers. Today SchneeAI is operated as a managed platform. If self-hosting is a hard requirement for your team, email [email protected] to talk about the enterprise path.
How do I get started?
Email [email protected] with your use case. We’ll set up an onboarding call, provision your tenant, and walk through the first integration. Most teams ship their first AI feature through the Gateway within a week.
Is there an evaluation period for design partners?
Yes. Design partners get room to integrate and run a proof of concept before committing. Talk to us about your workload and we’ll make sure you have what you need to evaluate the platform properly.