Interface Platform

The Interface platform is Evose's unified model abstraction. It harmonizes API differences across vendors (OpenAI, Anthropic, Azure, in-house) into a single calling interface.

What It Solves

Each vendor's API differs slightly:

Vendor	API path	Auth
OpenAI	`/v1/chat/completions`	Bearer
Anthropic	`/v1/messages`	x-api-key
Azure OpenAI	`/openai/deployments/<dep>/chat/completions?api-version=...`	api-key
Self-hosted	Any	Any

The business side shouldn't know about these differences. The Interface platform hides them.

Configure an Interface Platform

Field	Description
Platform name	`OpenAI US-West`
Vendor type	OpenAI / Anthropic / Azure / Google / OpenAI-compatible / Custom
Base URL	e.g. `https://api.openai.com/v1`
API Keys	Multiple keys (round-robin, canary, rate-limit sharing)

Multiple Base URLs

Multiple Base URLs are supported (geographic proximity / failover):

Platform: OpenAI
├─ Base URL 1: https://api.openai.com/v1            (US-West)
├─ Base URL 2: https://api-eu.openai.com/v1         (Europe)
└─ Base URL 3: https://internal-proxy.example.com/v1 (Internal proxy)

Strategies: Round Robin / primary-backup / geo-proximity (planned).

Multiple API Keys

Each key has:

Field	Description
Key value	Encrypted at rest
Description	Note its purpose
Weight	Round Robin weight
Status	Enabled / disabled
Rotation	One-click replace; smooth swap of old/new

Bidirectional Mapping

Model                  ↔        Interface Platform
──────────────                  ──────────────────
gpt-4-turbo  ─┬─ registered to ─→  OpenAI US-West
              └─ registered to ─→  Azure China
gpt-4o       ──── registered to ─→  OpenAI US-West
qwen-max     ──── registered to ─→  Aliyun

One model can be registered to multiple platforms (GPT-4 via both OpenAI and Azure). Evose picks a route by policy at call time.

Routing

When the business calls "gpt-4-turbo":

1. Find which platforms gpt-4-turbo is registered on → [OpenAI US-West, Azure China]
2. Pick by Round Robin / priority
3. Within the platform, pick a Base URL + Key
4. Call; on failure, Failover to the next platform

Next Steps

Register a model → Model definition
Self-host model instances → Model deployment
Set org default → Default model configuration