diff --git a/src/lib/llm/utils/config.ts b/src/lib/llm/utils/config.ts
index 661136af..9d5162d4 100644
--- a/src/lib/llm/utils/config.ts
+++ b/src/lib/llm/utils/config.ts
@@ -18,6 +18,8 @@ export const DEFAULT_MODELS: Record<LLMProviderType, string> = {
   custom: "gpt-3.5-turbo",
 };
 
+const VALID_PROVIDERS = Object.keys(DEFAULT_MODELS) as LLMProviderType[];
+
 export const DEFAULT_API_URLS: Record<string, string> = {
   ollama: "http://localhost:11434/v1",
   openai: "https://api.openai.com/v1",
@@ -27,20 +29,14 @@ export const DEFAULT_API_URLS: Record<string, string> = {
 // Environment Resolution
 // ============================================================================
 
-function getEnvVar(key: string): string | undefined {
-  return process.env[key];
-}
-
 function resolveProvider(): LLMProviderType {
-  const provider = getEnvVar("LLM_PROVIDER")?.toLowerCase();
+  const provider = process.env.LLM_PROVIDER?.toLowerCase();
 
   if (!provider) {
     return DEFAULT_PROVIDER;
   }
 
-  const validProviders: LLMProviderType[] = ["gemini", "openai", "ollama", "custom"];
-
-  if (!validProviders.includes(provider as LLMProviderType)) {
+  if (!VALID_PROVIDERS.includes(provider as LLMProviderType)) {
     console.error(`[LLM] Invalid provider "${provider}", falling back to "${DEFAULT_PROVIDER}"`);
     return DEFAULT_PROVIDER;
   }
@@ -49,7 +45,7 @@ function resolveProvider(): LLMProviderType {
 }
 
 function resolveApiKey(provider: LLMProviderType): string | undefined {
-  const apiKey = getEnvVar("LLM_API_KEY");
+  const apiKey = process.env.LLM_API_KEY;
 
   // Ollama doesn't require API key
   if (!apiKey && provider === "ollama") {
@@ -60,13 +56,13 @@ function resolveApiKey(provider: LLMProviderType): string | undefined {
 }
 
 function resolveModel(provider: LLMProviderType): string {
-  const model = getEnvVar("LLM_MODEL");
+  const model = process.env.LLM_MODEL;
   return model || DEFAULT_MODELS[provider];
 }
 
 function resolveApiUrl(provider: LLMProviderType): string | undefined {
   // Primary: LLM_API_URL
-  const apiUrl = getEnvVar("LLM_API_URL");
+  const apiUrl = process.env.LLM_API_URL;
   if (apiUrl) {
     return apiUrl;
   }
@@ -116,10 +112,9 @@ export function resolveConfig(overrides?: Partial<LLMConfig>): LLMConfig {
  */
 export function validateConfig(config: LLMConfig): void {
   // Validate provider
-  const validProviders: LLMProviderType[] = ["gemini", "openai", "ollama", "custom"];
-  if (!validProviders.includes(config.provider)) {
+  if (!VALID_PROVIDERS.includes(config.provider)) {
     throw new LLMConfigError(
-      `Invalid provider: ${config.provider}. Valid options: ${validProviders.join(", ")}`,
+      `Invalid provider: ${config.provider}. Valid options: ${VALID_PROVIDERS.join(", ")}`,
       config.provider,
     );
   }
diff --git a/tests/unit/env-documentation.test.ts b/tests/unit/env-documentation.test.ts
index e84dbc70..30a08343 100644
--- a/tests/unit/env-documentation.test.ts
+++ b/tests/unit/env-documentation.test.ts
@@ -30,11 +30,10 @@
  *
  * What stays out of scope: a name that reaches `process.env[...]` as a function
  * argument or parameter, because resolving it needs a call graph and a regex
- * that faked one would report a name that is not a name. Two reads are in that
- * position today -- `getEnvVar("LLM_PROVIDER")` in `src/lib/llm/utils/config.ts`
- * and `process.env[envVar]` in `src/lib/seed/credential-resolver.ts` -- leaving
- * `HOSTNAME`, `MY_DB_PASSWORD` and the four `LLM_*` names undiscovered. That is
- * the whole remaining gap, not an aside.
+ * that faked one would report a name that is not a name. The only remaining
+ * read in that position is `process.env[envVar]` in
+ * `src/lib/seed/credential-resolver.ts`, leaving `HOSTNAME` and `MY_DB_PASSWORD`
+ * undiscovered. The four direct `LLM_*` reads are covered.
  */
 import { describe, expect, test } from "bun:test";
 import { readdirSync, readFileSync, statSync } from "node:fs";
@@ -175,16 +174,20 @@ describe("environment variable documentation", () => {
   });
 
   test("#609 boundary: a name reached through a function argument stays undiscovered", () => {
-    // getEnvVar("LLM_PROVIDER") in src/lib/llm/utils/config.ts and
-    // process.env[envVar] in src/lib/seed/credential-resolver.ts need a call
+    // process.env[envVar] in src/lib/seed/credential-resolver.ts needs a call
     // graph. When that stops being true, this test is the reminder to widen the
     // extractor rather than a silent gain.
     const names = new Set(readNames());
     // Control: a negative-only test passes on an empty set, so anchor it to a
     // name the extractor must always find before trusting the absences below.
     expect(names.has("JWT_SECRET")).toBe(true);
-    for (const name of ["LLM_PROVIDER", "LLM_API_KEY", "LLM_MODEL", "LLM_API_URL", "MY_DB_PASSWORD"]) {
-      expect(names.has(name)).toBe(false);
+    expect(names.has("MY_DB_PASSWORD")).toBe(false);
+  });
+
+  test("LLM configuration reads are visible to the documentation guard", () => {
+    const names = new Set(readNames());
+    for (const name of ["LLM_PROVIDER", "LLM_API_KEY", "LLM_MODEL", "LLM_API_URL"]) {
+      expect(names.has(name), name).toBe(true);
     }
   });
 
diff --git a/tests/unit/llm/config.test.ts b/tests/unit/llm/config.test.ts
index 29bd27ff..a1101897 100644
--- a/tests/unit/llm/config.test.ts
+++ b/tests/unit/llm/config.test.ts
@@ -8,7 +8,7 @@ import {
   DEFAULT_MODELS,
   DEFAULT_API_URLS,
 } from "@/lib/llm/utils/config";
-import { LLMConfigError } from "@/lib/llm/types";
+import { LLMConfigError, type LLMProviderType } from "@/lib/llm/types";
 
 // ============================================================================
 // Environment Variable Helpers
@@ -129,6 +129,22 @@ describe("resolveConfig", () => {
 // ============================================================================
 
 describe("validateConfig", () => {
+  test("accepts every provider in DEFAULT_MODELS and rejects an unknown provider", () => {
+    for (const provider of Object.keys(DEFAULT_MODELS) as LLMProviderType[]) {
+      expect(() =>
+        validateConfig({
+          provider,
+          model: DEFAULT_MODELS[provider],
+          apiKey: "test-key",
+          apiUrl: "https://example.test/v1",
+        }),
+      ).not.toThrow();
+    }
+    expect(() => validateConfig({ provider: "unknown-provider" as LLMProviderType, model: "test-model" })).toThrow(
+      LLMConfigError,
+    );
+  });
+
   test("valid gemini config with key passes", () => {
     expect(() =>
       validateConfig({