memcode-gateway

Documentation

Overview

memcode-api is the hosted inference gateway: the DELIBERATELY SEPARATE service that owns what must never ship in the public CLI binary — backend routing, provider API keys, and metering. The CLI keeps the agent loop, tools, and prompt doctrine local and calls up here for every model turn.

The wire is the OpenAI-compat surface (one-wire): POST /v1/chat/completions + GET /v1/models behind org-key auth, with the memcode extensions riding as ignorable headers/fields (internal/compat). Behind the translation: hybrid routing (Fireworks cheap lane + the strong-tier vendors), BYOK, entitlement gating, and the server-side ledger.