Prompts for AI Assistants
Keystone. The Model Switchboard. (KTMS), light-tokenmeter and ktms-admin are set up through code,
gRPC calls, environment variables and deployment files: work a coding assistant (Claude Code,
Cursor, GitHub Copilot, …) does well once it knows the exact contracts. Each prompt here gives it
those contracts: the version-pinned sources to read first, the facts that matter, the task, and the
limits. Each page ends with a list to check the result against.
How to Use Them
- Open the assistant in the repository it should change: your service, your infrastructure
repository, or a new folder.
- Copy the prompt, replace the
<…> placeholders, and send it.
- Check the result against the page’s list before you merge or apply anything.
- The prompts pin release 1.0.0 (
v1.0.0 in the URLs). For a later release, replace the tag.
- Never paste real API keys, vendor keys or passwords into the chat. The prompts ask for
environment variables and secret references, and so should you.
- An assistant that cannot open URLs needs the files instead: download the sources listed below
into the repository and point it at them.
Prompts
| Prompt |
What you get |
| Integrate a service |
Your service calls Keystone over the OpenAI-compatible HTTP API or gRPC, with effort, attribution and errors handled in one client module. |
| Configure routing |
Models, vendor credentials, rules, client keys, key policies and spend limits as a repeatable script. |
| Deploy the stack |
Docker Compose or Kubernetes (Helm) configuration for KTMS, light-tokenmeter and ktms-admin, with secrets, TLS and monitoring. |
| Add a local GPU node |
A Mac or Linux machine serving a local model to Keystone, with cloud fallback. |
| Migrate from LiteLLM or vendor SDKs |
A move from LiteLLM Proxy or direct vendor SDK calls, with a mapping of models, keys and limits. |
| Usage and spend reports |
Reports and alerts on spend per project, user, model and job from light-tokenmeter. |
| Production review |
An audit of a running deployment against the production checklist, with alert rules. |
Sources the Prompts Use
| Source |
Pinned URL |
| Keystone (KTMS) README |
https://raw.githubusercontent.com/ai-process/ktms/v1.0.0/README.md |
Gateway API (ktms.v1) |
https://raw.githubusercontent.com/ai-process/ktms/v1.0.0/proto/ktms/v1/gateway.proto |
Admin API (ktms.v1) |
https://raw.githubusercontent.com/ai-process/ktms/v1.0.0/proto/ktms/v1/admin.proto |
| Helm chart values |
https://raw.githubusercontent.com/ai-process/ktms/v1.0.0/deploy/helm/ktms/values.yaml |
| Node installer |
https://raw.githubusercontent.com/ai-process/ktms/v1.0.0/scripts/install-node.sh |
| light-tokenmeter README |
https://raw.githubusercontent.com/ai-process/light-tokenmeter/v1.0.0/README.md |
| ktms-admin README |
https://raw.githubusercontent.com/ai-process/ktms-admin/v1.0.0/README.md |
| Full-stack Compose file |
https://raw.githubusercontent.com/ai-process/ktms-admin/v1.0.0/examples/compose.yaml |
ai-process