fix(llmtr): round per-million cost and drop the residency claim

Scaling the gateway's per-token prices by a million left binary float
noise behind - $1e-7/token landed on 0.09999999999999999 instead of
0.1 - and those values are what a session's cost is rendered from.

Also drops the data-residency wording from the README: the gateway
hosts models in Türkiye, which is the claim worth making, and the
compliance framing is the operator's to make, not the client's.

Verified live against the gateway: completion, streaming, tool calling,
catalog-derived reasoning efforts, a Türkiye-hosted model, and a full
agent run through the CLI.

Author: knowhycodata

README.md

@@ -112,7 +112,7 @@ CyberStrike integrates with the entire AI ecosystem through 23 bundled SDK provi
 | **Cerebras**              | LLaMA on Cerebras        | Fastest inference available             |
 | **Cohere**                | Command R+               | RAG-optimized models                    |
 | **OpenRouter**            | 300+ models              | Single API, any model                   |
-| **LLMTR**                 | 180+ models              | Türkiye-hosted models, KVKK residency   |
+| **LLMTR**                 | 180+ models              | Türkiye-hosted models, local pricing    |
 | **Together AI**           | Open-source models       | Fine-tuning support                     |
 | **DeepInfra**             | Open-source models       | Pay-per-token, no GPU needed            |
 | **Perplexity**            | Sonar models             | Search-augmented generation             |

README.tr.md

@@ -100,7 +100,7 @@ Diger yapay zeka araclari siradaki ne yapacaginizi soylemenizi bekler. CyberStri
 
 **Herhangi Bir LLM, Bagimlilik Yok**
 
-Kutudan ciktigi gibi 15+ saglayici: Anthropic, OpenAI, Google, Amazon Bedrock, Azure, Groq, Mistral, OpenRouter, LLMTR (Turkiye'de barindirilan modeller ve KVKK veri yerelligi) — hatta OpenAI uyumlu uc noktalar araciligiyla yerel modeller bile. Claude, GPT, Gemini veya kendi barindiginiz LLM ile calistirin. Modeller iyilestikce ve ucuzladikca, CyberStrike de onlarla birlikte gelisir.
+Kutudan ciktigi gibi 15+ saglayici: Anthropic, OpenAI, Google, Amazon Bedrock, Azure, Groq, Mistral, OpenRouter, LLMTR (Turkiye'de barindirilan modeller) — hatta OpenAI uyumlu uc noktalar araciligiyla yerel modeller bile. Claude, GPT, Gemini veya kendi barindiginiz LLM ile calistirin. Modeller iyilestikce ve ucuzladikca, CyberStrike de onlarla birlikte gelisir.
 
 </td>
 <td width="50%">

packages/cyberstrike/src/provider/llmtr.ts

@@ -67,9 +67,13 @@ export namespace LLMTR {
   }
 
   // Catalog prices are per token; models.dev costs are per million tokens.
+  // Scaling by a million leaves binary float noise ($1e-7/token lands on
+  // 0.09999999999999999 rather than 0.1), so round back to a precision far
+  // finer than any real price - cost is rendered to the user per session.
   function price(input: string | undefined) {
     const value = Number(input)
-    return Number.isFinite(value) ? value * 1_000_000 : 0
+    if (!Number.isFinite(value)) return 0
+    return Math.round(value * 1_000_000 * 1e6) / 1e6
   }
 
   function model(entry: Entry): ModelsDev.Model {

packages/cyberstrike/test/provider/llmtr.test.ts

@@ -63,9 +63,11 @@ describe("LLMTR catalog", () => {
 
   test("converts per-token pricing to per-million-token cost", () => {
     const model = provider.models["llmtr/trendyol-asure-12b"]
-    expect(model.cost?.input).toBeCloseTo(0.1, 10)
-    expect(model.cost?.output).toBeCloseTo(0.5, 10)
-    expect(model.cost?.cache_read).toBeCloseTo(0.025, 10)
+    // Exact, not close-to: scaling per-token prices by a million must not leave
+    // float noise behind, since these values are rendered as session cost.
+    expect(model.cost?.input).toBe(0.1)
+    expect(model.cost?.output).toBe(0.5)
+    expect(model.cost?.cache_read).toBe(0.025)
   })
 
   test("carries limits, modalities and capabilities across", () => {