Merge pull request #3741 from estib-vega/master

feat: Add support for Ollama

app/src/lib/ai/anthropicClient.ts

@@ -1,24 +1,21 @@
-import {
-	MessageRole,
-	type AIClient,
-	type AnthropicModelName,
-	type PromptMessage
-} from '$lib/ai/types';
+import { SHORT_DEFAULT_COMMIT_TEMPLATE, SHORT_DEFAULT_BRANCH_TEMPLATE } from '$lib/ai/prompts';
 import { fetch, Body } from '@tauri-apps/api/http';
+import type { AIClient, AnthropicModelName, PromptMessage } from '$lib/ai/types';
 
 type AnthropicAPIResponse = { content: { text: string }[] };
 
 export class AnthropicAIClient implements AIClient {
+	defaultCommitTemplate = SHORT_DEFAULT_COMMIT_TEMPLATE;
+	defaultBranchTemplate = SHORT_DEFAULT_BRANCH_TEMPLATE;
+
 	constructor(
 		private apiKey: string,
 		private modelName: AnthropicModelName
 	) {}
 
-	async evaluate(prompt: string) {
-		const messages: PromptMessage[] = [{ role: MessageRole.User, content: prompt }];
-
+	async evaluate(prompt: PromptMessage[]) {
 		const body = Body.json({
-			messages,
+			messages: prompt,
 			max_tokens: 1024,
 			model: this.modelName
 		});

app/src/lib/ai/butlerClient.ts

@@ -1,19 +1,21 @@
-import { MessageRole, type ModelKind, type AIClient, type PromptMessage } from '$lib/ai/types';
+import { SHORT_DEFAULT_BRANCH_TEMPLATE, SHORT_DEFAULT_COMMIT_TEMPLATE } from '$lib/ai/prompts';
+import type { AIClient, ModelKind, PromptMessage } from '$lib/ai/types';
 import type { HttpClient } from '$lib/backend/httpClient';
 
 export class ButlerAIClient implements AIClient {
+	defaultCommitTemplate = SHORT_DEFAULT_COMMIT_TEMPLATE;
+	defaultBranchTemplate = SHORT_DEFAULT_BRANCH_TEMPLATE;
+
 	constructor(
 		private cloud: HttpClient,
 		private userToken: string,
 		private modelKind: ModelKind
 	) {}
 
-	async evaluate(prompt: string) {
-		const messages: PromptMessage[] = [{ role: MessageRole.User, content: prompt }];
-
+	async evaluate(prompt: PromptMessage[]) {
 		const response = await this.cloud.post<{ message: string }>('evaluate_prompt/predict.json', {
 			body: {
-				messages,
+				messages: prompt,
 				max_tokens: 400,
 				model_kind: this.modelKind
 			},

app/src/lib/ai/ollamaClient.ts

@@ -0,0 +1,165 @@
+import { LONG_DEFAULT_BRANCH_TEMPLATE, LONG_DEFAULT_COMMIT_TEMPLATE } from '$lib/ai/prompts';
+import { MessageRole, type PromptMessage, type AIClient } from '$lib/ai/types';
+import { isNonEmptyObject } from '$lib/utils/typeguards';
+
+export const DEFAULT_OLLAMA_ENDPOINT = 'http://127.0.0.1:11434';
+export const DEFAULT_OLLAMA_MODEL_NAME = 'llama3';
+
+enum OllamaAPEndpoint {
+	Generate = 'api/generate',
+	Chat = 'api/chat',
+	Embed = 'api/embeddings'
+}
+
+interface OllamaRequestOptions {
+	/**
+	 * The temperature of the model.
+	 * Increasing the temperature will make the model answer more creatively. (Default: 0.8)
+	 */
+	temperature: number;
+}
+
+interface OllamaChatRequest {
+	model: string;
+	messages: PromptMessage[];
+	stream: boolean;
+	format?: 'json';
+	options?: OllamaRequestOptions;
+}
+
+interface BaseOllamaMResponse {
+	created_at: string;
+	done: boolean;
+	model: string;
+}
+
+interface OllamaChatResponse extends BaseOllamaMResponse {
+	message: PromptMessage;
+	done: true;
+}
+
+interface OllamaChatMessageFormat {
+	result: string;
+}
+
+const OLLAMA_CHAT_MESSAGE_FORMAT_SCHEMA = {
+	type: 'object',
+	properties: {
+		result: { type: 'string' }
+	},
+	required: ['result'],
+	additionalProperties: false
+};
+
+function isOllamaChatMessageFormat(message: unknown): message is OllamaChatMessageFormat {
+	if (!isNonEmptyObject(message)) {
+		return false;
+	}
+
+	return typeof message.result === 'string';
+}
+
+function isOllamaChatResponse(response: unknown): response is OllamaChatResponse {
+	if (!isNonEmptyObject(response)) {
+		return false;
+	}
+
+	return (
+		isNonEmptyObject(response.message) &&
+		typeof response.message.role == 'string' &&
+		typeof response.message.content == 'string'
+	);
+}
+
+export class OllamaClient implements AIClient {
+	defaultCommitTemplate = LONG_DEFAULT_COMMIT_TEMPLATE;
+	defaultBranchTemplate = LONG_DEFAULT_BRANCH_TEMPLATE;
+
+	constructor(
+		private endpoint: string,
+		private modelName: string
+	) {}
+
+	async evaluate(prompt: PromptMessage[]) {
+		const messages = this.formatPrompt(prompt);
+		const response = await this.chat(messages);
+		const rawResponse = JSON.parse(response.message.content);
+		if (!isOllamaChatMessageFormat(rawResponse)) {
+			throw new Error('Invalid response: ' + response.message.content);
+		}
+
+		return rawResponse.result;
+	}
+
+	/**
+	 * Appends a system message which instructs the model to respond using a particular JSON schema
+	 * Modifies the prompt's Assistant messages to make use of the correct schema
+	 */
+	private formatPrompt(prompt: PromptMessage[]) {
+		const withFormattedResponses = prompt.map((promptMessage) => {
+			if (promptMessage.role == MessageRole.Assistant) {
+				return {
+					role: MessageRole.Assistant,
+					content: JSON.stringify({ result: promptMessage.content })
+				};
+			} else {
+				return promptMessage;
+			}
+		});
+
+		return [
+			{
+				role: MessageRole.System,
+				content: `You are an expert in software development. Answer the given user prompts following the specified instructions.
+Return your response in JSON and only use the following JSON schema:
+${JSON.stringify(OLLAMA_CHAT_MESSAGE_FORMAT_SCHEMA, null, 2)}`
+			},
+			...withFormattedResponses
+		];
+	}
+
+	/**
+	 * Fetches the chat using the specified request.
+	 * @param request - The OllamaChatRequest object containing the request details.
+	 * @returns A Promise that resolves to the Response object.
+	 */
+	private async fetchChat(request: OllamaChatRequest): Promise<Response> {
+		const url = new URL(OllamaAPEndpoint.Chat, this.endpoint);
+		const result = await fetch(url, {
+			method: 'POST',
+			headers: {
+				'Content-Type': 'application/json'
+			},
+			body: JSON.stringify(request)
+		});
+		return result;
+	}
+
+	/**
+	 * Sends a chat message to the LLM model and returns the response.
+	 *
+	 * @param messages - An array of LLMChatMessage objects representing the chat messages.
+	 * @param options - Optional LLMRequestOptions object for specifying additional options.
+	 * @throws Error if the response is invalid.
+	 * @returns A Promise that resolves to an LLMResponse object representing the response from the LLM model.
+	 */
+	private async chat(
+		messages: PromptMessage[],
+		options?: OllamaRequestOptions
+	): Promise<OllamaChatResponse> {
+		const result = await this.fetchChat({
+			model: this.modelName,
+			stream: false,
+			messages,
+			options,
+			format: 'json'
+		});
+
+		const json = await result.json();
+		if (!isOllamaChatResponse(json)) {
+			throw new Error('Invalid response\n' + JSON.stringify(json));
+		}
+
+		return json;
+	}
+}

app/src/lib/ai/openAIClient.ts

@@ -1,24 +1,19 @@
-import {
-	MessageRole,
-	type OpenAIModelName,
-	type PromptMessage,
-	type AIClient
-} from '$lib/ai/types';
+import { SHORT_DEFAULT_BRANCH_TEMPLATE, SHORT_DEFAULT_COMMIT_TEMPLATE } from '$lib/ai/prompts';
+import type { OpenAIModelName, PromptMessage, AIClient } from '$lib/ai/types';
 import type OpenAI from 'openai';
 
 export class OpenAIClient implements AIClient {
+	defaultCommitTemplate = SHORT_DEFAULT_COMMIT_TEMPLATE;
+	defaultBranchTemplate = SHORT_DEFAULT_BRANCH_TEMPLATE;
+
 	constructor(
 		private modelName: OpenAIModelName,
 		private openAI: OpenAI
 	) {}
 
-	async evaluate(prompt: string) {
-		const messages: PromptMessage[] = [{ role: MessageRole.User, content: prompt }];
-
+	async evaluate(prompt: PromptMessage[]) {
 		const response = await this.openAI.chat.completions.create({
-			// @ts-expect-error There is a type mismatch where it seems to want a "name" paramater
-			// that isn't required https://github.com/openai/openai-openapi/issues/118#issuecomment-1847667988
-			messages,
+			messages: prompt,
 			model: this.modelName,
 			max_tokens: 400
 		});

app/src/lib/ai/prompts.ts

@@ -0,0 +1,107 @@
+import { type PromptMessage, MessageRole } from '$lib/ai/types';
+
+export const SHORT_DEFAULT_COMMIT_TEMPLATE: PromptMessage[] = [
+	{
+		role: MessageRole.User,
+		content: `Please could you write a commit message for my changes.
+Only respond with the commit message. Don't give any notes.
+Explain what were the changes and why the changes were done.
+Focus the most important changes.
+Use the present tense.
+Use a semantic commit prefix.
+Hard wrap lines at 72 characters.
+Ensure the title is only 50 characters.
+Do not start any lines with the hash symbol.
+%{brief_style}
+%{emoji_style}
+
+Here is my git diff:
+%{diff}`
+	}
+];
+
+export const LONG_DEFAULT_COMMIT_TEMPLATE: PromptMessage[] = [
+	{
+		role: MessageRole.User,
+		content: `Please could you write a commit message for my changes.
+Explain what were the changes and why the changes were done.
+Focus the most important changes.
+Use the present tense.
+Use a semantic commit prefix.
+Hard wrap lines at 72 characters.
+Ensure the title is only 50 characters.
+Do not start any lines with the hash symbol.
+Only respond with the commit message.
+
+Here is my git diff:
+diff --git a/src/utils/typing.ts b/src/utils/typing.ts
+index 1cbfaa2..7aeebcf 100644
+--- a/src/utils/typing.ts
++++ b/src/utils/typing.ts
+@@ -35,3 +35,10 @@ export function isNonEmptyObject(something: unknown): something is UnknownObject
+     (Object.keys(something).length > 0 || Object.getOwnPropertySymbols(something).length > 0)
+   );
+ }
++
++export function isArrayOf<T>(
++  something: unknown,
++  check: (value: unknown) => value is T
++): something is T[] {
++  return Array.isArray(something) && something.every(check);
++}`
+	},
+	{
+		role: MessageRole.Assistant,
+		content: `Typing utilities: Check for array of type
+
+Added an utility function to check whether a given value is an array of a specific type.`
+	},
+	...SHORT_DEFAULT_COMMIT_TEMPLATE
+];
+
+export const SHORT_DEFAULT_BRANCH_TEMPLATE: PromptMessage[] = [
+	{
+		role: MessageRole.User,
+		content: `Please could you write a branch name for my changes.
+A branch name represent a brief description of the changes in the diff (branch).
+Branch names should contain no whitespace and instead use dashes to separate words.
+Branch names should contain a maximum of 5 words.
+Only respond with the branch name.
+
+Here is my git diff:
+%{diff}`
+	}
+];
+
+export const LONG_DEFAULT_BRANCH_TEMPLATE: PromptMessage[] = [
+	{
+		role: MessageRole.User,
+		content: `Please could you write a branch name for my changes.
+A branch name represent a brief description of the changes in the diff (branch).
+Branch names should contain no whitespace and instead use dashes to separate words.
+Branch names should contain a maximum of 5 words.
+Only respond with the branch name.
+
+Here is my git diff:
+diff --git a/src/utils/typing.ts b/src/utils/typing.ts
+index 1cbfaa2..7aeebcf 100644
+--- a/src/utils/typing.ts
++++ b/src/utils/typing.ts
+@@ -35,3 +35,10 @@ export function isNonEmptyObject(something: unknown): something is UnknownObject
+     (Object.keys(something).length > 0 || Object.getOwnPropertySymbols(something).length > 0)
+   );
+ }
++
++export function isArrayOf<T>(
++  something: unknown,
++  check: (value: unknown) => value is T
++): something is T[] {
++  return Array.isArray(something) && something.every(check);
++}`
+	},
+	{
+		role: MessageRole.Assistant,
+		content: `utils-typing-is-array-of-type`
+	},
+	...SHORT_DEFAULT_BRANCH_TEMPLATE
+];

app/src/lib/ai/service.test.ts

@@ -1,8 +1,15 @@
 import { AnthropicAIClient } from '$lib/ai/anthropicClient';
 import { ButlerAIClient } from '$lib/ai/butlerClient';
 import { OpenAIClient } from '$lib/ai/openAIClient';
+import { SHORT_DEFAULT_BRANCH_TEMPLATE, SHORT_DEFAULT_COMMIT_TEMPLATE } from '$lib/ai/prompts';
 import { AIService, GitAIConfigKey, KeyOption, buildDiff } from '$lib/ai/service';
-import { AnthropicModelName, ModelKind, OpenAIModelName, type AIClient } from '$lib/ai/types';
+import {
+	AnthropicModelName,
+	ModelKind,
+	OpenAIModelName,
+	type AIClient,
+	type PromptMessage
+} from '$lib/ai/types';
 import { HttpClient } from '$lib/backend/httpClient';
 import * as toasts from '$lib/utils/toasts';
 import { Hunk } from '$lib/vbranches/types';
@@ -40,9 +47,11 @@ const fetchMock = vi.fn();
 const cloud = new HttpClient(fetchMock);
 
 class DummyAIClient implements AIClient {
+	defaultCommitTemplate = SHORT_DEFAULT_COMMIT_TEMPLATE;
+	defaultBranchTemplate = SHORT_DEFAULT_BRANCH_TEMPLATE;
 	constructor(private response = 'lorem ipsum') {}
 
-	async evaluate(_prompt: string) {
+	async evaluate(_prompt: PromptMessage[]) {
 		return this.response;
 	}
 }

app/src/lib/ai/service.ts

@@ -1,7 +1,19 @@
 import { AnthropicAIClient } from '$lib/ai/anthropicClient';
 import { ButlerAIClient } from '$lib/ai/butlerClient';
+import {
+	DEFAULT_OLLAMA_ENDPOINT,
+	DEFAULT_OLLAMA_MODEL_NAME,
+	OllamaClient
+} from '$lib/ai/ollamaClient';
 import { OpenAIClient } from '$lib/ai/openAIClient';
-import { OpenAIModelName, type AIClient, AnthropicModelName } from '$lib/ai/types';
+import {
+	OpenAIModelName,
+	type AIClient,
+	AnthropicModelName,
+	ModelKind,
+	type PromptMessage,
+	MessageRole
+} from '$lib/ai/types';
 import { splitMessage } from '$lib/utils/commitMessage';
 import * as toasts from '$lib/utils/toasts';
 import OpenAI from 'openai';
@@ -11,39 +23,6 @@ import type { Hunk } from '$lib/vbranches/types';
 
 const maxDiffLengthLimitForAPI = 5000;
 
-const defaultCommitTemplate = `
-Please could you write a commit message for my changes.
-Explain what were the changes and why the changes were done.
-Focus the most important changes.
-Use the present tense.
-Use a semantic commit prefix.
-Hard wrap lines at 72 characters.
-Ensure the title is only 50 characters.
-Do not start any lines with the hash symbol.
-Only respond with the commit message.
-%{brief_style}
-%{emoji_style}
-
-Here is my git diff:
-%{diff}
-`;
-
-const defaultBranchTemplate = `
-Please could you write a branch name for my changes.
-A branch name represent a brief description of the changes in the diff (branch).
-Branch names should contain no whitespace and instead use dashes to separate words.
-Branch names should contain a maximum of 5 words.
-Only respond with the branch name.
-
-Here is my git diff:
-%{diff}
-`;
-
-export enum ModelKind {
-	OpenAI = 'openai',
-	Anthropic = 'anthropic'
-}
-
 export enum KeyOption {
 	BringYourOwn = 'bringYourOwn',
 	ButlerAPI = 'butlerAPI'
@@ -57,20 +36,22 @@ export enum GitAIConfigKey {
 	AnthropicKeyOption = 'gitbutler.aiAnthropicKeyOption',
 	AnthropicModelName = 'gitbutler.aiAnthropicModelName',
 	AnthropicKey = 'gitbutler.aiAnthropicKey',
-	DiffLengthLimit = 'gitbutler.diffLengthLimit'
+	DiffLengthLimit = 'gitbutler.diffLengthLimit',
+	OllamaEndpoint = 'gitbutler.aiOllamaEndpoint',
+	OllamaModelName = 'gitbutler.aiOllamaModelName'
 }
 
 type SummarizeCommitOpts = {
 	hunks: Hunk[];
 	useEmojiStyle?: boolean;
 	useBriefStyle?: boolean;
-	commitTemplate?: string;
+	commitTemplate?: PromptMessage[];
 	userToken?: string;
 };
 
 type SummarizeBranchOpts = {
 	hunks: Hunk[];
-	branchTemplate?: string;
+	branchTemplate?: PromptMessage[];
 	userToken?: string;
 };
 
@@ -84,7 +65,7 @@ export function buildDiff(hunks: Hunk[], limit: number) {
 function shuffle<T>(items: T[]): T[] {
 	return items
 		.map((item) => ({ item, value: Math.random() }))
-		.sort()
+		.sort(({ value: a }, { value: b }) => a - b)
 		.map((item) => item.item);
 }
 
@@ -159,6 +140,20 @@ export class AIService {
 		}
 	}
 
+	async getOllamaEndpoint() {
+		return await this.gitConfig.getWithDefault<string>(
+			GitAIConfigKey.OllamaEndpoint,
+			DEFAULT_OLLAMA_ENDPOINT
+		);
+	}
+
+	async getOllamaModelName() {
+		return await this.gitConfig.getWithDefault<string>(
+			GitAIConfigKey.OllamaModelName,
+			DEFAULT_OLLAMA_MODEL_NAME
+		);
+	}
+
 	async usingGitButlerAPI() {
 		const modelKind = await this.getModelKind();
 		const openAIKeyOption = await this.getOpenAIKeyOption();
@@ -176,13 +171,19 @@ export class AIService {
 		const modelKind = await this.getModelKind();
 		const openAIKey = await this.getOpenAIKey();
 		const anthropicKey = await this.getAnthropicKey();
+		const ollamaEndpoint = await this.getOllamaEndpoint();
+		const ollamaModelName = await this.getOllamaModelName();
 
 		if (await this.usingGitButlerAPI()) return !!userToken;
 
 		const openAIActiveAndKeyProvided = modelKind == ModelKind.OpenAI && !!openAIKey;
 		const anthropicActiveAndKeyProvided = modelKind == ModelKind.Anthropic && !!anthropicKey;
+		const ollamaActiveAndEndpointProvided =
+			modelKind == ModelKind.Ollama && !!ollamaEndpoint && !!ollamaModelName;
 
-		return openAIActiveAndKeyProvided || anthropicActiveAndKeyProvided;
+		return (
+			openAIActiveAndKeyProvided || anthropicActiveAndKeyProvided || ollamaActiveAndEndpointProvided
+		);
 	}
 
 	// This optionally returns a summarizer. There are a few conditions for how this may occur
@@ -199,6 +200,12 @@ export class AIService {
 			return new ButlerAIClient(this.cloud, userToken, modelKind);
 		}
 
+		if (modelKind == ModelKind.Ollama) {
+			const ollamaEndpoint = await this.getOllamaEndpoint();
+			const ollamaModelName = await this.getOllamaModelName();
+			return new OllamaClient(ollamaEndpoint, ollamaModelName);
+		}
+
 		if (modelKind == ModelKind.OpenAI) {
 			const openAIModelName = await this.getOpenAIModleName();
 			const openAIKey = await this.getOpenAIKey();
@@ -233,24 +240,37 @@ export class AIService {
 		hunks,
 		useEmojiStyle = false,
 		useBriefStyle = false,
-		commitTemplate = defaultCommitTemplate,
+		commitTemplate,
 		userToken
 	}: SummarizeCommitOpts) {
 		const aiClient = await this.buildClient(userToken);
 		if (!aiClient) return;
 
 		const diffLengthLimit = await this.getDiffLengthLimitConsideringAPI();
-		let prompt = commitTemplate.replaceAll('%{diff}', buildDiff(hunks, diffLengthLimit));
+		const defaultedCommitTemplate = commitTemplate || aiClient.defaultCommitTemplate;
 
-		const briefPart = useBriefStyle
-			? 'The commit message must be only one sentence and as short as possible.'
-			: '';
-		prompt = prompt.replaceAll('%{brief_style}', briefPart);
+		const prompt = defaultedCommitTemplate.map((promptMessage) => {
+			if (promptMessage.role != MessageRole.User) {
+				return promptMessage;
+			}
 
-		const emojiPart = useEmojiStyle
-			? 'Make use of GitMoji in the title prefix.'
-			: "Don't use any emoji.";
-		prompt = prompt.replaceAll('%{emoji_style}', emojiPart);
+			let content = promptMessage.content.replaceAll('%{diff}', buildDiff(hunks, diffLengthLimit));
+
+			const briefPart = useBriefStyle
+				? 'The commit message must be only one sentence and as short as possible.'
+				: '';
+			content = content.replaceAll('%{brief_style}', briefPart);
+
+			const emojiPart = useEmojiStyle
+				? 'Make use of GitMoji in the title prefix.'
+				: "Don't use any emoji.";
+			content = content.replaceAll('%{emoji_style}', emojiPart);
+
+			return {
+				role: MessageRole.User,
+				content
+			};
+		});
 
 		let message = await aiClient.evaluate(prompt);
 
@@ -262,16 +282,23 @@ export class AIService {
 		return description ? `${title}\n\n${description}` : title;
 	}
 
-	async summarizeBranch({
-		hunks,
-		branchTemplate = defaultBranchTemplate,
-		userToken = undefined
-	}: SummarizeBranchOpts) {
+	async summarizeBranch({ hunks, branchTemplate, userToken = undefined }: SummarizeBranchOpts) {
 		const aiClient = await this.buildClient(userToken);
 		if (!aiClient) return;
 
 		const diffLengthLimit = await this.getDiffLengthLimitConsideringAPI();
-		const prompt = branchTemplate.replaceAll('%{diff}', buildDiff(hunks, diffLengthLimit));
+		const defaultedBranchTemplate = branchTemplate || aiClient.defaultBranchTemplate;
+		const prompt = defaultedBranchTemplate.map((promptMessage) => {
+			if (promptMessage.role != MessageRole.User) {
+				return promptMessage;
+			}
+
+			return {
+				role: MessageRole.User,
+				content: promptMessage.content.replaceAll('%{diff}', buildDiff(hunks, diffLengthLimit))
+			};
+		});
+
 		const message = await aiClient.evaluate(prompt);
 		return message.replaceAll(' ', '-').replaceAll('\n', '-');
 	}

app/src/lib/ai/types.ts

@@ -1,6 +1,7 @@
 export enum ModelKind {
 	OpenAI = 'openai',
-	Anthropic = 'anthropic'
+	Anthropic = 'anthropic',
+	Ollama = 'ollama'
 }
 
 export enum OpenAIModelName {
@@ -16,8 +17,9 @@ export enum AnthropicModelName {
 }
 
 export enum MessageRole {
+	System = 'system',
 	User = 'user',
-	Assistant = 'assisstant'
+	Assistant = 'assistant'
 }
 
 export interface PromptMessage {
@@ -26,5 +28,8 @@ export interface PromptMessage {
 }
 
 export interface AIClient {
-	evaluate(prompt: string): Promise<string>;
+	evaluate(prompt: PromptMessage[]): Promise<string>;
+
+	defaultBranchTemplate: PromptMessage[];
+	defaultCommitTemplate: PromptMessage[];
 }

app/src/lib/utils/typeguards.ts

@@ -5,3 +5,19 @@ export function isDefined<T>(file: T | undefined | null): file is T {
 export function notNull<T>(file: T | undefined | null): file is T {
 	return file !== null;
 }
+
+export type UnknownObject = Record<string, unknown>;
+
+/**
+ * Checks if the provided value is a non-empty object.
+ * @param something - The value to be checked.
+ * @returns A boolean indicating whether the value is a non-empty object.
+ */
+export function isNonEmptyObject(something: unknown): something is UnknownObject {
+	return (
+		typeof something === 'object' &&
+		something !== null &&
+		!Array.isArray(something) &&
+		(Object.keys(something).length > 0 || Object.getOwnPropertySymbols(something).length > 0)
+	);
+}

app/src/routes/settings/ai/+page.svelte

@@ -1,6 +1,6 @@
 <script lang="ts">
-	import { AIService, GitAIConfigKey, KeyOption, ModelKind } from '$lib/ai/service';
-	import { OpenAIModelName, AnthropicModelName } from '$lib/ai/types';
+	import { AIService, GitAIConfigKey, KeyOption } from '$lib/ai/service';
+	import { OpenAIModelName, AnthropicModelName, ModelKind } from '$lib/ai/types';
 	import { GitConfigService } from '$lib/backend/gitConfigService';
 	import InfoMessage from '$lib/components/InfoMessage.svelte';
 	import RadioButton from '$lib/components/RadioButton.svelte';
@@ -30,6 +30,8 @@
 	let anthropicKey: string | undefined;
 	let anthropicModelName: AnthropicModelName | undefined;
 	let diffLengthLimit: number | undefined;
+	let ollamaEndpoint: string | undefined;
+	let ollamaModel: string | undefined;
 
 	function setConfiguration(key: GitAIConfigKey, value: string | undefined) {
 		if (!initialized) return;
@@ -48,6 +50,9 @@
 	$: setConfiguration(GitAIConfigKey.AnthropicKey, anthropicKey);
 	$: setConfiguration(GitAIConfigKey.DiffLengthLimit, diffLengthLimit?.toString());
 
+	$: setConfiguration(GitAIConfigKey.OllamaEndpoint, ollamaEndpoint);
+	$: setConfiguration(GitAIConfigKey.OllamaModelName, ollamaModel);
+
 	onMount(async () => {
 		modelKind = await aiService.getModelKind();
 
@@ -61,6 +66,9 @@
 
 		diffLengthLimit = await aiService.getDiffLengthLimit();
 
+		ollamaEndpoint = await aiService.getOllamaEndpoint();
+		ollamaModel = await aiService.getOllamaModelName();
+
 		// Ensure reactive declarations have finished running before we set initialized to true
 		await tick();
 
@@ -261,15 +269,31 @@
 			</SectionCard>
 		{/if}
 
-		<SectionCard roundedTop={false} orientation="row" disabled={true}>
-			<svelte:fragment slot="title">Custom Endpoint</svelte:fragment>
+		<SectionCard
+			roundedTop={false}
+			roundedBottom={modelKind != ModelKind.Ollama}
+			orientation="row"
+			labelFor="ollama"
+			bottomBorder={modelKind != ModelKind.Ollama}
+		>
+			<svelte:fragment slot="title">Ollama 🦙</svelte:fragment>
 			<svelte:fragment slot="actions">
-				<RadioButton disabled={true} name="modelKind" />
+				<RadioButton name="modelKind" id="ollama" value={ModelKind.Ollama} />
 			</svelte:fragment>
-			<svelte:fragment slot="caption"
-				>Support for custom AI endpoints is coming soon!</svelte:fragment
-			>
 		</SectionCard>
+		{#if modelKind == ModelKind.Ollama}
+			<SectionCard hasTopRadius={false} roundedTop={false} orientation="row" topDivider>
+				<div class="inputs-group">
+					<TextBox
+						label="Endpoint"
+						bind:value={ollamaEndpoint}
+						placeholder="http://127.0.0.1:11434"
+					/>
+
+					<TextBox label="Model" bind:value={ollamaModel} placeholder="llama3" />
+				</div>
+			</SectionCard>
+		{/if}
 	</form>
 
 	<Spacer />