fix(vscode): use fim fallback for chat autocomplete

.changeset/chat-fim-fallback.md

@@ -0,0 +1,6 @@
+---
+"kilo-code": patch
+"@kilocode/kilo-gateway": patch
+---
+
+Use the matching FIM model for chat autocomplete when Next Edit is selected.

packages/kilo-gateway/src/autocomplete.ts

@@ -1,7 +1,7 @@
 export type AutocompleteProviderID = "kilo" | "mistral" | "inception"
 export type DirectAutocompleteProviderID = Exclude<AutocompleteProviderID, "kilo">
 
-export interface AutocompleteModelDef {
+interface AutocompleteModelBase {
   /** Stable combined value for internal comparisons. */
   readonly id: string
   /** Model value stored in settings and sent to the autocomplete API. */
@@ -18,14 +18,23 @@ export interface AutocompleteModelDef {
   readonly directProvider?: DirectAutocompleteProviderID
   /** Request temperature. */
   readonly temperature: number
-  /**
-   * Which gateway endpoint this model targets. Defaults to "fim" if omitted
-   * (back-compat with existing entries). Models with `kind: "edit"` route
-   * through `/kilo/edit` and use Mercury's Next Edit pipeline.
-   */
-  readonly kind?: "fim" | "edit"
 }
 
+export type AutocompleteModelDef = AutocompleteModelBase &
+  (
+    | {
+        /** Route through `/kilo/edit` using the Next Edit pipeline. */
+        readonly kind: "edit"
+        /** Stable combined ID of the FIM model used where Next Edit is unsupported. */
+        readonly fimModelID: string
+      }
+    | {
+        /** Route through the FIM endpoint. */
+        readonly kind?: "fim"
+        readonly fimModelID?: never
+      }
+  )
+
 const models: AutocompleteModelDef[] = [
   {
     id: "kilo/mistralai/codestral-2508",
@@ -57,6 +66,7 @@ const models: AutocompleteModelDef[] = [
     requestModel: "inception/mercury-edit-2",
     temperature: 0,
     kind: "edit",
+    fimModelID: "kilo/inception/mercury-edit-2",
   },
   {
     id: "mistral/codestral-2508",
@@ -91,6 +101,7 @@ const models: AutocompleteModelDef[] = [
     directProvider: "inception",
     temperature: 0,
     kind: "edit",
+    fimModelID: "inception/mercury-edit-2",
   },
 ]
 

packages/kilo-gateway/test/autocomplete.test.ts

@@ -18,3 +18,15 @@ describe("DEFAULT_AUTOCOMPLETE_MODEL", () => {
     expect(DEFAULT_AUTOCOMPLETE_MODEL.kind).toBe("edit")
   })
 })
+
+describe("Next Edit FIM models", () => {
+  test("reference a FIM model from the same provider", () => {
+    for (const model of AUTOCOMPLETE_MODELS) {
+      if (model.kind !== "edit") continue
+      const sibling = AUTOCOMPLETE_MODELS.find((candidate) => candidate.id === model.fimModelID)
+      expect(sibling).toBeDefined()
+      expect(sibling?.kind).not.toBe("edit")
+      expect(sibling?.providerID).toBe(model.providerID)
+    }
+  })
+})

packages/kilo-vscode/src/services/autocomplete/chat-autocomplete/ChatTextAreaAutocomplete.ts

@@ -7,7 +7,7 @@ import { VisibleCodeTracker } from "../context/VisibleCodeTracker"
 import { FileIgnoreController } from "../shims/FileIgnoreController"
 import type { KiloConnectionService } from "../../cli-backend"
 import { generateFim, hasValidCredentials } from "../fim"
-import { getAutocompleteModel } from "../../../shared/autocomplete-models"
+import { getAutocompleteModel, getAutocompleteModelById } from "../../../shared/autocomplete-models"
 import { finalizeChatSuggestion, buildChatPrefix } from "./chat-autocomplete-utils"
 
 interface ChatCompletionRequestMessage {
@@ -20,6 +20,12 @@ interface ChatCompletionResponseSender {
   postMessage(message: { type: "chatCompletionResult"; text: string; requestId: string }): void
 }
 
+export function getChatAutocompleteModel(provider?: string, model?: string) {
+  const info = getAutocompleteModel(provider, model)
+  if (info.kind !== "edit") return info
+  return getAutocompleteModelById(info.fimModelID)
+}
+
 /**
  * Chat textarea autocomplete with cached per-request objects.
  *
@@ -77,7 +83,7 @@ export class ChatTextAreaAutocomplete {
 
   async getCompletion(userText: string, visibleCodeContext?: VisibleCodeContext): Promise<{ suggestion: string }> {
     const cfg = vscode.workspace.getConfiguration("kilo-code.new.autocomplete")
-    const entry = getAutocompleteModel(cfg.get<string>("provider"), cfg.get<string>("model"))
+    const entry = getChatAutocompleteModel(cfg.get<string>("provider"), cfg.get<string>("model"))
     const startTime = Date.now()
 
     // Build context for telemetry

packages/kilo-vscode/tests/unit/chat-autocomplete-utils.test.ts

@@ -3,6 +3,7 @@ import {
   finalizeChatSuggestion,
   buildChatPrefix,
 } from "../../src/services/autocomplete/chat-autocomplete/chat-autocomplete-utils"
+import { getChatAutocompleteModel } from "../../src/services/autocomplete/chat-autocomplete/ChatTextAreaAutocomplete"
 
 describe("finalizeChatSuggestion", () => {
   it("returns empty string for empty input", () => {
@@ -98,3 +99,14 @@ describe("buildChatPrefix", () => {
     expect(result).toContain("hi")
   })
 })
+
+describe("getChatAutocompleteModel", () => {
+  it("uses the matching FIM model for Next Edit settings", () => {
+    expect(getChatAutocompleteModel("kilo", "inception/mercury-next-edit").id).toBe("kilo/inception/mercury-edit-2")
+    expect(getChatAutocompleteModel("inception", "mercury-next-edit").id).toBe("inception/mercury-edit-2")
+  })
+
+  it("keeps FIM settings unchanged", () => {
+    expect(getChatAutocompleteModel("kilo", "mistralai/codestral-2508").id).toBe("kilo/mistralai/codestral-2508")
+  })
+})