fix(cli): only route auxiliary tasks to kilo-auto/small with kilo credentials

.changeset/small-model-fallback-requires-kilo-credentials.md

@@ -0,0 +1,5 @@
+---
+"@kilocode/cli": patch
+---
+
+Only route auxiliary tasks (session titles, commit messages, branch names) to the cloud kilo-auto/small model when kilo credentials are configured; otherwise fall back to the session's own model so offline and local-only setups keep working.

packages/opencode/src/kilocode/provider/provider.ts

@@ -299,6 +299,24 @@ export function kiloSmallModelPriority(providerID: string): string[] | undefined
   return undefined
 }
 
+/**
+ * True when the user has kilo credentials: a KILO_API_KEY env var, a stored
+ * auth entry, or an apiKey in the kilo provider config. Mirrors the hasKey
+ * check in the kilo custom loader. The kilo provider is autoloaded with an
+ * anonymous key even without credentials, so this gates the cloud
+ * kilo-auto/small fallback to users who can actually reach it.
+ */
+export function hasKiloCredentials(
+  cfg: { provider?: Record<string, { options?: { apiKey?: string } } | null> },
+  auth: unknown,
+  env: Record<string, string | undefined>,
+) {
+  if (env.KILO_API_KEY) return true
+  if (auth) return true
+  if (cfg.provider?.["kilo"]?.options?.apiKey) return true
+  return false
+}
+
 // ---------------------------------------------------------------------------
 // Fetch timeout wrappers
 // Replaces AbortSignal.timeout() with a cancellable setTimeout+AbortController

packages/opencode/src/provider/provider.ts

@@ -43,6 +43,7 @@ import {
   patchKiloProviderAuth,
   publicKiloProvider,
   kiloSmallModelPriority,
+  hasKiloCredentials,
   buildTimeoutSignal,
   requestTimeout,
   wrapFirstByte,
@@ -2062,9 +2063,19 @@ const layer = Layer.effect(
         if (candidates[0]) return candidates[0]
       }
 
-      // kilocode_change start - fall back to kilo's auto small model
-      const kiloFallback = s.providers[ProviderV2.ID.make("kilo")] ?? s.catalog[ProviderV2.ID.make("kilo")]
-      if (kiloFallback?.models["kilo-auto/small"]) return kiloFallback.models["kilo-auto/small"]
+      // kilocode_change start - fall back to kilo's auto small model only when the user actually has
+      // kilo credentials. The kilo provider is always autoloaded (anonymous key), so checking it
+      // unconditionally would route auxiliary tasks (session titles, commit messages, branch names)
+      // to the cloud for users without kilo access and break offline/local-only setups.
+      const kiloFallback = s.providers[ProviderV2.ID.make("kilo")]
+      if (kiloFallback?.models["kilo-auto/small"]) {
+        const hasCreds = hasKiloCredentials(
+          cfg,
+          yield* auth.get(ProviderV2.ID.make("kilo")).pipe(Effect.orDie),
+          yield* env.all(),
+        )
+        if (hasCreds) return kiloFallback.models["kilo-auto/small"]
+      }
       // kilocode_change end
 
       return undefined

packages/opencode/test/kilocode/provider/provider.test.ts

@@ -0,0 +1,84 @@
+import { afterEach, expect } from "bun:test"
+import { LayerNode } from "@opencode-ai/core/effect/layer-node"
+import { ProviderV2 } from "@opencode-ai/core/provider"
+import { Effect } from "effect"
+import { Env } from "@/env"
+import { Plugin } from "@/plugin/index"
+import { Provider } from "@/provider/provider"
+import { disposeAllInstances } from "../../fixture/fixture"
+import { testEffect } from "../../lib/effect"
+
+const originalEnv = new Map<string, string | undefined>()
+
+const rememberEnv = (key: string) => {
+  if (!originalEnv.has(key)) originalEnv.set(key, process.env[key])
+}
+
+const clearEnv = (key: string) =>
+  Effect.gen(function* () {
+    rememberEnv(key)
+    delete process.env[key]
+    yield* Env.use.remove(key)
+  })
+
+afterEach(async () => {
+  for (const [key, value] of originalEnv) {
+    if (value === undefined) delete process.env[key]
+    else process.env[key] = value
+  }
+  originalEnv.clear()
+  await disposeAllInstances()
+})
+
+const it = testEffect(LayerNode.compile(LayerNode.group([Provider.node, Env.node, Plugin.node])))
+
+it.instance(
+  "getSmallModel returns undefined without kilo credentials when model IDs lack family metadata",
+  Effect.gen(function* () {
+    for (const key of ["KILO_API_KEY", "KILO_AUTH_CONTENT", "KILO_CONFIG_CONTENT"]) {
+      yield* clearEnv(key)
+    }
+    const model = yield* Provider.use.getSmallModel(ProviderV2.ID.make("test-provider"))
+    expect(model).toBeUndefined()
+  }),
+  {
+    config: {
+      provider: {
+        "test-provider": {
+          name: "Test Provider",
+          npm: "@ai-sdk/openai-compatible",
+          models: {
+            "gpt-5-nano": { release_date: "2026-01-01" },
+          },
+          options: { apiKey: "test-key" },
+        },
+        kilo: null,
+      },
+    },
+  },
+)
+
+it.instance(
+  "getSmallModel falls back to Kilo auto when the kilo provider is configured",
+  Effect.gen(function* () {
+    const model = yield* Provider.use.getSmallModel(ProviderV2.ID.make("test-provider"))
+    expect(model).toMatchObject({ providerID: "kilo", id: "kilo-auto/small" })
+  }),
+  {
+    config: {
+      provider: {
+        "test-provider": {
+          name: "Test Provider",
+          npm: "@ai-sdk/openai-compatible",
+          models: {
+            "gpt-5-nano": { release_date: "2026-01-01" },
+          },
+          options: { apiKey: "test-key" },
+        },
+        kilo: {
+          options: { apiKey: "kilo-key" },
+        },
+      },
+    },
+  },
+)

packages/opencode/test/kilocode/session-prompt-compaction-safety.test.ts

@@ -425,7 +425,8 @@ describe("SessionPrompt compaction safety", () => {
         Effect.fnUntraced(function* ({ llm }) {
           const prompt = yield* SessionPrompt.Service
           const sessions = yield* Session.Service
-          const chat = yield* sessions.create({})
+          // Explicit title so the auxiliary title-generation call is skipped and llm.calls stays exact.
+          const chat = yield* sessions.create({ title: "Pending request replay" })
           const old = yield* user(chat.id, "old request")
           yield* assistant(chat.id, old.id, {
             tokens: { input: 95_000, output: 100, reasoning: 0, cache: { read: 0, write: 0 } },
@@ -466,7 +467,11 @@ describe("SessionPrompt compaction safety", () => {
             const prompt = yield* SessionPrompt.Service
             const sessions = yield* Session.Service
             const tools = finish !== "stop"
-            const chat = yield* sessions.create({ permission: [{ permission: "*", pattern: "*", action: "allow" }] })
+            // Explicit title so the auxiliary title-generation call is skipped and llm.calls stays exact.
+            const chat = yield* sessions.create({
+              title: "Completed work no-replay",
+              permission: [{ permission: "*", pattern: "*", action: "allow" }],
+            })
             yield* llm.push(
               (tools ? reply().tool("glob", { pattern: "*.txt" }) : reply().text("answer"))
                 .finish(finish)
@@ -500,7 +505,8 @@ describe("SessionPrompt compaction safety", () => {
           const prompt = yield* SessionPrompt.Service
           const sessions = yield* Session.Service
           const compaction = yield* SessionCompaction.Service
-          const chat = yield* sessions.create({})
+          // Explicit title so the auxiliary title-generation call is skipped and llm.calls stays exact.
+          const chat = yield* sessions.create({ title: "Saved marker replay" })
           const old = yield* user(chat.id, "old request")
           yield* assistant(chat.id, old.id)
           const request = yield* user(chat.id, "run the tool")

packages/opencode/test/provider/provider.test.ts

@@ -769,30 +769,6 @@ it.instance(
   },
 )
 
-it.instance(
-  // kilocode_change start - Kilo always has an auto-routed small-model fallback
-  "getSmallModel falls back to Kilo auto when model IDs lack family metadata",
-  Effect.gen(function* () {
-    const model = yield* Provider.use.getSmallModel(ProviderV2.ID.make("test-provider"))
-    expect(model).toMatchObject({ providerID: "kilo", id: "kilo-auto/small" })
-  }),
-  // kilocode_change end
-  {
-    config: {
-      provider: {
-        "test-provider": {
-          name: "Test Provider",
-          npm: "@ai-sdk/openai-compatible",
-          models: {
-            "gpt-5-nano": { release_date: "2026-01-01" },
-          },
-          options: { apiKey: "test-key" },
-        },
-      },
-    },
-  },
-)
-
 it.instance("getSmallModel skips inferred models for Azure", () =>
   Effect.gen(function* () {
     yield* set("AZURE_RESOURCE_NAME", "test-resource")