Problem
There is no provider in the catalog that serves models hosted inside Türkiye. Teams that need in-country inference currently have to hand-write a custom provider block in cyberstrike.json, which means no model list, no cost data, and no reasoning variants.
Proposal
Add LLMTR as a bundled provider. It is an OpenAI-compatible gateway (https://llmtr.com/v1, Bearer auth) with 188 chat models. What it adds that existing providers don't:
- Türkiye-hosted models — Trendyol Asure 12B, Magibu 11B v8, Muse Glimmer 30B and local Qwen/Gemma builds, none of which are reachable through any currently bundled provider
- Lower pricing on global models than their list rate
Notes
LLMTR is not published on models.dev, so the provider entry has to be built from the gateway's own public /v1/models endpoint (no key required) and merged into the catalog. provider.ts rejects providers missing from the model list, so a custom loader alone is not enough.
I have a working branch with tests, verified live against the gateway (completion, streaming, tool calling, reasoning efforts, a Türkiye-hosted model, and a full agent run). Happy to open the PR if this is wanted — flagging first per the issue-first policy.
Disclosure: I work at LLMTR.
Problem
There is no provider in the catalog that serves models hosted inside Türkiye. Teams that need in-country inference currently have to hand-write a custom provider block in
cyberstrike.json, which means no model list, no cost data, and no reasoning variants.Proposal
Add LLMTR as a bundled provider. It is an OpenAI-compatible gateway (
https://llmtr.com/v1,Bearerauth) with 188 chat models. What it adds that existing providers don't:Notes
LLMTR is not published on models.dev, so the provider entry has to be built from the gateway's own public
/v1/modelsendpoint (no key required) and merged into the catalog.provider.tsrejects providers missing from the model list, so a custom loader alone is not enough.I have a working branch with tests, verified live against the gateway (completion, streaming, tool calling, reasoning efforts, a Türkiye-hosted model, and a full agent run). Happy to open the PR if this is wanted — flagging first per the issue-first policy.
Disclosure: I work at LLMTR.