AI EpesiCloud AI Platform
Launch console →
Centralized AI infrastructure

One AI layer for every EpesiCloud product.

CRM, Engage, Pulse, Analytics, Affinity, Payroll, HR, Call Center — every applet stops wiring itself to OpenAI, Groq, Cloudflare or Gemini directly. Instead it calls a stable text.generate or conversation.chat service, and the platform decides the provider and model, tracks every token, and falls back automatically.

How it works

Applications never name a model.

A caller invokes a service by its stable key. The router matches required capabilities against every enabled model, filters to the ones this tenant has an active credential for, then walks a priority-ordered fallback chain until one answers — recording the whole thing.

ApplicationAI Service (keyed: "text.generate")Model Router → capability match (model_capabilities ∩ required) → tenant filter (enabled_models, provider_accounts=active) → routing policy (priority + fallback chain)Provider adapter (Groq · Cloudflare · Gemini · …)executions + token_usage recorded (reserve → finalize)
What's live

Built for real, today.

Text generation

Single-turn completions for summaries, extraction and classification — one call, the router picks the model.

Conversation chat

Multi-turn threads with full history, backed by the same routing and accounting as everything else.

Automatic fallback

A priority chain across providers. If the primary model errors or hits a limit, the next one answers — transparently.

Token accounting

Every request is reserved then finalized into executions and a per-tenant usage ledger with real cost.

Capability routing

Models and capabilities are first-class. Ask for a capability, not a vendor — the registry does the matching.

API keys for applets

Issue scoped, rate-limited keys so any product can reach the platform instead of a raw provider.

Images, video, audio, document processing, knowledge bases, agents, evaluations and workflows are modeled in the schema and shipping next — no migration needed to turn them on.

Providers

One registry, many vendors.

Provider-specific quirks live in thin adapters, never in the schema. Add a vendor by adding an adapter — callers never change.

Groq Cloudflare Workers AI Google Gemini OpenAI Anthropic Mistral Cohere Ollama
11
normalized schema layers
8
providers registered
0
vendor-specific columns

Stop wiring every applet to a provider.

Point them at one platform. Swap models without a redeploy.

Launch the console →