AI Gateway

Purpose

Expose named AI capabilities to clients without giving them raw provider credentials.

How it works

Clients request named capabilities (e.g. ai.grammar_review), not raw provider access.

Flow: auth → capability check → reserve credits → provider call → capture/release → usage record.

Mock provider is default in development (AI_DEFAULT_PROVIDER=mock). OpenAI-compatible adapter uses OPENAI_API_KEY or a credential profile secret_reference.

Live prototype

Runs ai.grammar_review on the docs ClientSite. Production clients call POST /api/v1/server/ai/executions/ with a server secret (never from the browser). This control uses a docs-only proxy so you can see a real execution without exposing secrets.

Mock grammar review

Creates an AIExecution row. Mock output is expected in development.

Code to paste (trusted server / Python)

from mkp_client import MKPClient

client = MKPClient(base_url="https://YOUR_PLATFORM", client_id="skpub_...", client_secret="sk_...")
result = client.ai.execute(
    "ai.grammar_review",
    {"text": "Ths is a deme."},
    user_public_id="usr_...",  # optional depending on capability
)
print(result)

Related Pages

• Protected actions
• Wallet
• Python client