Documentation

Everything you need to run the ai-process stack: the KTMS gateway, the light-tokenmeter usage meter and the ktms-admin console. Each service also ships a full reference in its repository README.

Architecture & deployment

How the three services fit together, with a Docker Compose file for the whole stack.

KTMS gateway

gRPC and OpenAI-compatible APIs, effort routing, fallbacks, limits, batches, speech and images.

Local GPU node pool

Self-registering inference nodes, health states, Smooth WRR balancing and cloud fallback.

Quick start: local node

Connect Ollama, vLLM, TGI or llama.cpp on your own GPU to KTMS in a few commands.

light-tokenmeter

Idempotent usage events, fast COPY batches, daily summaries and greedy vs worthy analysis.

ktms-admin console

Google sign-in with allowlists, models, rules and keys, a prompt playground and a spend dashboard.

Full references

• KTMS README — configuration reference, error contract, API stability, operations.
• light-tokenmeter README — HTTP and gRPC API reference, configuration, security.
• ktms-admin README — Google OAuth setup, deployment, security model.