Skip to content

Stop one-shot AI from picking Haiku or the first available model #1178

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
arul28 merged 3 commits into from
Aug 29, 2026
Merged

Stop one-shot AI from picking Haiku or the first available model #1178

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
fd2c5d9
Stop one-shot AI from picking Haiku or the first available model.
arul28 Aug 29, 2026
8790210
Keep session-model aliases in the naming pool.
arul28 Aug 29, 2026
c9685fa
Fix session-metadata test typing that failed desktop typecheck.
arul28 Aug 29, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
46 changes: 42 additions & 4 deletions apps/desktop/src/main/services/ai/aiIntegrationService.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -77,7 +77,7 @@
}));

vi.mock("../chat/cursorCloudCreateOptions", async (importOriginal) => {
const actual = await importOriginal<typeof import("../chat/cursorCloudCreateOptions")>();

Check warning on line 80 in apps/desktop/src/main/services/ai/aiIntegrationService.test.ts

View workflow job for this annotation

GitHub Actions / lint-desktop 

`import()` type annotations are forbidden
return {
...actual,
resolveCursorCloudCreateCloudExtras: (
Expand All @@ -99,7 +99,7 @@
}));

import { createDynamicCursorCliModelDescriptor, getLocalProviderDefaultEndpoint } from "../../../shared/modelRegistry";
import { createAiIntegrationService } from "./aiIntegrationService";
import { createAiIntegrationService, missingFeatureModelMessage } from "./aiIntegrationService";

type ServiceFactoryOptions = {
aiConfig?: Record<string, unknown>;
Expand Down Expand Up @@ -354,9 +354,12 @@
}));
});

it("resolves a default task model when model is omitted", async () => {
it("uses the feature model override when executeTask omits model", async () => {
const { service } = makeService({
aiConfig: { features: { orchestrator: true } },
aiConfig: {
features: { orchestrator: true },
featureModelOverrides: { orchestrator: "openai/gpt-5.4" },
},
});

await service.executeTask({
Expand All @@ -368,7 +371,42 @@

expect(mockState.runProviderTask).toHaveBeenCalledTimes(1);
const firstCall = mockState.runProviderTask.mock.calls[0]?.[0] as Record<string, unknown>;
expect(firstCall.descriptor).toMatchObject({ id: expect.any(String) });
expect(firstCall.descriptor).toMatchObject({ id: "openai/gpt-5.4" });
});

it("skips AI instead of picking a default model when no setting is configured", async () => {
const { service } = makeService({
aiConfig: { features: { orchestrator: true } },
});

await expect(
service.executeTask({
feature: "orchestrator",
taskType: "implementation",
prompt: "Implement feature",
cwd: "/tmp"
})
).rejects.toThrow(missingFeatureModelMessage("orchestrator"));
expect(mockState.runProviderTask).not.toHaveBeenCalled();
});

it("requires an explicit model for session intelligence tasks", async () => {
const { service } = makeService({
aiConfig: {
features: { terminal_summaries: true },
featureModelOverrides: { terminal_summaries: "openai/gpt-5.4" },
},
});

await expect(
service.executeTask({
feature: "terminal_summaries",
taskType: "session_title",
prompt: "Title this chat",
cwd: "/tmp",
})
).rejects.toThrow(/Session intelligence task 'session_title' requires an explicit model/);
expect(mockState.runProviderTask).not.toHaveBeenCalled();
});

it("fails in guest mode when no providers are available", async () => {
Expand Down
158 changes: 66 additions & 92 deletions apps/desktop/src/main/services/ai/aiIntegrationService.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,6 @@ import type {
import {
decodeOpenCodeRegistryId,
replaceDynamicPiModelDescriptors,
getDefaultModelDescriptor,
getModelById,
getAvailableModels,
getLocalProviderDefaultEndpoint,
Expand Down Expand Up @@ -215,11 +214,6 @@ export type ExecuteAiTaskResult = {
durationMs: number;
};

type RuntimeTaskDefaults = {
modelId: string;
timeoutMs: number;
};

function readCursorSdkAgentName(agent: object): string {
if (!("name" in agent)) return "";
const name: unknown = (agent as { name: unknown }).name;
Expand All @@ -236,67 +230,49 @@ const DEFAULT_AI_FEATURE_FLAGS: Record<AiFeatureKey, boolean> = {
initial_context: true,
};

const DEFAULT_CLAUDE_TASK_MODEL_ID = getDefaultModelDescriptor("claude")?.id ?? "anthropic/claude-sonnet-5";
const DEFAULT_CODEX_TASK_MODEL_ID = getDefaultModelDescriptor("codex")?.id ?? "openai/gpt-5.6-sol";

const TASK_DEFAULTS: Record<AiTaskType, RuntimeTaskDefaults> = {
planning: {
modelId: DEFAULT_CLAUDE_TASK_MODEL_ID,
timeoutMs: 45_000
},
implementation: {
modelId: DEFAULT_CODEX_TASK_MODEL_ID,
timeoutMs: 120_000
},
review: {
modelId: DEFAULT_CLAUDE_TASK_MODEL_ID,
timeoutMs: 30_000
},
conflict_resolution: {
modelId: DEFAULT_CLAUDE_TASK_MODEL_ID,
timeoutMs: 60_000
},
commit_message: {
modelId: "anthropic/claude-haiku-4-5",
timeoutMs: 20_000
},
narrative: {
modelId: "anthropic/claude-haiku-4-5",
timeoutMs: 45_000
},
pr_description: {
modelId: "anthropic/claude-haiku-4-5",
timeoutMs: 30_000
},
terminal_summary: {
modelId: "anthropic/claude-haiku-4-5",
timeoutMs: 20_000
},
session_title: {
modelId: "anthropic/claude-haiku-4-5",
timeoutMs: 20_000
},
session_summary: {
modelId: "anthropic/claude-haiku-4-5",
timeoutMs: 45_000
},
handoff_summary: {
modelId: "anthropic/claude-haiku-4-5",
timeoutMs: 45_000
},
continuity_summary: {
modelId: "anthropic/claude-haiku-4-5",
timeoutMs: 45_000
},
context_compaction: {
modelId: "anthropic/claude-haiku-4-5",
timeoutMs: 120_000
},
initial_context: {
modelId: DEFAULT_CLAUDE_TASK_MODEL_ID,
timeoutMs: 120_000
const SESSION_INTELLIGENCE_TASK_TYPES: ReadonlySet<AiTaskType> = new Set([
"session_title",
"session_summary",
"handoff_summary",
"continuity_summary",
]);

/** These one-shots must receive a caller-chosen model. They never read feature pickers. */
const EXPLICIT_MODEL_ONLY_TASK_TYPES: ReadonlySet<AiTaskType> = new Set([
...SESSION_INTELLIGENCE_TASK_TYPES,
"context_compaction",
]);

export function readConfiguredFeatureModel(aiConfig: unknown, feature: AiFeatureKey): string | null {
if (!isRecord(aiConfig)) return null;
const overrides = isRecord(aiConfig.featureModelOverrides) ? aiConfig.featureModelOverrides : {};
const raw = overrides[feature];
const modelId = typeof raw === "string" ? raw.trim() : "";
return modelId.length ? modelId : null;
}

export function missingFeatureModelMessage(feature: AiFeatureKey): string {
switch (feature) {
case "commit_messages":
return "Choose a Commit Messages model in Settings or type a commit message manually.";
case "pr_descriptions":
return "Choose a PR Descriptions model in Settings or write the description manually.";
case "terminal_summaries":
return "Choose a Summaries model in Settings.";
case "conflict_proposals":
return "Choose a Conflict Proposals model in Settings.";
case "narratives":
return "Choose a Narratives model in Settings.";
case "orchestrator":
return "Choose an Orchestrator model in Settings.";
case "initial_context":
return "Choose an Initial Context model in Settings.";
default: {
const _exhaustive: never = feature;
return _exhaustive;
}
}
};
}

const CODEX_FALLBACK_MODELS: AgentModelDescriptor[] = listModelDescriptorsForProvider("codex")
.map((descriptor) => ({ id: descriptor.id, label: descriptor.displayName }));
Expand Down Expand Up @@ -1509,49 +1485,42 @@ export function createAiIntegrationService(args: {
);
};

const resolveModelForTask = async (
const getConfiguredFeatureModel = (feature: AiFeatureKey): string | null => {
return readConfiguredFeatureModel(extractAiConfig(projectConfigService.get()), feature);
};

const resolveModelForTask = (
taskType: AiTaskType,
modelIdHint?: string,
authHint?: DetectedAuth[],
): Promise<string> => {
feature?: AiFeatureKey,
): string => {
const snapshot = projectConfigService.get();
const aiConfig = extractAiConfig(snapshot);
const taskRouting = isRecord(aiConfig.taskRouting) ? aiConfig.taskRouting : {};
const taskOverride = isRecord(taskRouting[taskType]) ? (taskRouting[taskType] as Record<string, unknown>) : {};
const overrideModelId = toStringOrNull(taskOverride.model);
const requestedModelHint = modelIdHint ?? overrideModelId ?? undefined;

// If explicit model ID provided and valid, use it
if (requestedModelHint) {
const exact = getModelById(requestedModelHint);
if (exact) return exact.id;
}

// Resolve from alias (e.g. "sonnet" -> "anthropic/claude-sonnet-5")
if (requestedModelHint) {
const resolved = resolveModelAlias(requestedModelHint);
if (resolved) return resolved.id;
throw new Error(`Unknown model '${requestedModelHint}'.`);
}

// Check task defaults and map provider family to model ID.
const defaults = TASK_DEFAULTS[taskType];
const auth = authHint ?? await detectAuth();
const available = getAvailableModels(auth);

if (!available.length) {
throw new Error("No AI providers detected. Install Claude Code CLI, Codex CLI, or configure an API key.");
if (SESSION_INTELLIGENCE_TASK_TYPES.has(taskType)) {
throw new Error(`Session intelligence task '${taskType}' requires an explicit model.`);
}

const preferredDescriptor = getModelById(defaults.modelId) ?? resolveModelAlias(defaults.modelId);
if (preferredDescriptor) {
const exactMatch = available.find((candidate) => candidate.id === preferredDescriptor.id || candidate.shortId === preferredDescriptor.shortId);
if (exactMatch) return exactMatch.id;
const familyMatch = available.find((candidate) => candidate.family === preferredDescriptor.family);
if (familyMatch) return familyMatch.id;
if (taskType === "context_compaction") {
throw new Error("Context compaction requires an explicit model.");
}

// Fall back to first available
return available[0].id;
throw new Error(
feature
? missingFeatureModelMessage(feature)
: "Choose a model in Settings before running this AI task.",
);
};

const executeProviderTaskPath = async (
Expand Down Expand Up @@ -1632,15 +1601,19 @@ export function createAiIntegrationService(args: {
}

checkBudget(args.feature);
const requestedModel = toStringOrNull(args.model);
const featureModel = EXPLICIT_MODEL_ONLY_TASK_TYPES.has(args.taskType)
? null
: getConfiguredFeatureModel(args.feature);
const requestedModel = toStringOrNull(args.model) ?? featureModel;
const explicitDescriptor = requestedModel
? (getModelById(requestedModel) ?? resolveModelAlias(requestedModel))
: null;
if (requestedModel && !explicitDescriptor) {
throw new Error(`Unknown model '${requestedModel}'.`);
}

const resolvedModelId = explicitDescriptor?.id ?? await resolveModelForTask(args.taskType, requestedModel ?? undefined, auth);
const resolvedModelId = explicitDescriptor?.id
?? resolveModelForTask(args.taskType, requestedModel ?? undefined, args.feature);
logger.info("ai.task.begin", {
requestId,
taskType: args.taskType,
Expand Down Expand Up @@ -2089,6 +2062,7 @@ export function createAiIntegrationService(args: {

getAvailabilityAsync,
resolveModelForTask,
getConfiguredFeatureModel,
invalidateProviderReadinessCaches,

// Backward-compatible convenience methods used by migrated services.
Expand Down
Loading
Loading