Add Custom Model Support and Configurable API URL

src/noteGenerator.ts

@@ -105,24 +105,30 @@ export function noteGenerator(
 	}
 
 	const buildMessages = async (node: CanvasNode) => {
-		const encoding = getEncoding(settings)
-
 		const messages: openai.ChatCompletionRequestMessage[] = []
+		const isCustomModel = settings.apiModel === CHAT_MODELS.CUSTOMIZE.name
+		const encoding = isCustomModel ? null : getEncoding(settings)
 		let tokenCount = 0
 
 		// Note: We are not checking for system prompt longer than context window.
 		// That scenario makes no sense, though.
 		const systemPrompt = await getSystemPrompt(node)
 		if (systemPrompt) {
-			tokenCount += encoding.encode(systemPrompt).length
+			if (!isCustomModel && encoding) {
+				tokenCount += encoding.encode(systemPrompt).length
+			}
+			messages.push({
+				role: 'system',
+				content: systemPrompt
+			})
 		}
 
 		const visit = async (node: CanvasNode, depth: number) => {
 			if (settings.maxDepth && depth > settings.maxDepth) return false
 
 			const nodeData = node.getData()
 			let nodeText = (await readNodeContent(node))?.trim() || ''
-			const inputLimit = getTokenLimit(settings)
+			const inputLimit = isCustomModel ? Infinity : getTokenLimit(settings)
 
 			let shouldContinue = true
 			if (!nodeText) {
@@ -140,34 +146,36 @@ export function noteGenerator(
 			} else {
 				if (isSystemPromptNode(nodeText)) return true
 
-				let nodeTokens = encoding.encode(nodeText)
+				let nodeTokens = isCustomModel || !encoding ? null : encoding.encode(nodeText)
 				let keptNodeTokens: number
 
-				if (tokenCount + nodeTokens.length > inputLimit) {
+				if (!isCustomModel && nodeTokens && tokenCount + nodeTokens.length > inputLimit) {
 					// will exceed input limit
-
 					shouldContinue = false
 
 					// Leaving one token margin, just in case
-					const keepTokens = nodeTokens.slice(0, inputLimit - tokenCount - 1)
-					const truncateTextTo = encoding.decode(keepTokens).length
-					logDebug(
-						`Truncating node text from ${nodeText.length} to ${truncateTextTo} characters`
+					const keepTokens = Math.max(
+						0,
+						inputLimit - tokenCount - 1
 					)
-					nodeText = nodeText.slice(0, truncateTextTo)
-					keptNodeTokens = keepTokens.length
+					if (encoding) {
+						const keepBytes = encoding
+							.decode(nodeTokens.slice(0, keepTokens))
+							.length
+						nodeText = nodeText.slice(0, keepBytes)
+					}
+					keptNodeTokens = keepTokens
 				} else {
-					keptNodeTokens = nodeTokens.length
+					keptNodeTokens = nodeTokens?.length || 0
 				}
 
-				tokenCount += keptNodeTokens
-
-				const role: openai.ChatCompletionRequestMessageRoleEnum =
-					nodeData.chat_role === 'assistant' ? 'assistant' : 'user'
+				if (!isCustomModel) {
+					tokenCount += keptNodeTokens
+				}
 
 				messages.unshift({
 					content: nodeText,
-					role
+					role: 'user'
 				})
 			}
 
@@ -177,13 +185,6 @@ export function noteGenerator(
 		await visitNodeAndAncestors(node, visit)
 
 		if (messages.length) {
-			if (systemPrompt) {
-				messages.unshift({
-					content: systemPrompt,
-					role: 'system'
-				})
-			}
-
 			return { messages, tokenCount }
 		} else {
 			return { messages: [], tokenCount: 0 }
@@ -242,9 +243,10 @@ export function noteGenerator(
 					settings.apiModel,
 					messages,
 					{
-						max_tokens: settings.maxResponseTokens || undefined,
-						temperature: settings.temperature
-					}
+						temperature: settings.temperature,
+						max_tokens: settings.maxResponseTokens || undefined
+					},
+					settings.customModelName
 				)
 
 				if (generated == null) {
@@ -287,13 +289,19 @@ export function noteGenerator(
 }
 
 function getEncoding(settings: ChatStreamSettings) {
+	if (settings.apiModel === CHAT_MODELS.CUSTOMIZE.name) {
+		return null
+	}
 	const model: ChatModelSettings | undefined = chatModelByName(settings.apiModel)
 	return encodingForModel(
 		(model?.encodingFrom || model?.name || DEFAULT_SETTINGS.apiModel) as TiktokenModel
 	)
 }
 
 function getTokenLimit(settings: ChatStreamSettings) {
+	if (settings.apiModel === CHAT_MODELS.CUSTOMIZE.name) {
+		return Infinity
+	}
 	const model = chatModelByName(settings.apiModel) || CHAT_MODELS.GPT_35_TURBO_0125
 	return settings.maxInputTokens
 		? Math.min(settings.maxInputTokens, model.tokenLimit)

src/openai/chatGPT.ts

@@ -1,7 +1,11 @@
 import { request, RequestUrlParam } from 'obsidian'
 import { openai } from './chatGPT-types'
 
-export const OPENAI_COMPLETIONS_URL = `https://api.openai.com/v1/chat/completions`
+export const DEFAULT_OPENAI_COMPLETIONS_URL = `https://api.openai.com/v1/chat/completions`
+
+export const getOpenAICompletionsURL = (configuredUrl?: string): string => {
+	return configuredUrl || DEFAULT_OPENAI_COMPLETIONS_URL
+}
 
 export type ChatModelSettings = {
 	name: string,
@@ -62,6 +66,10 @@ export const CHAT_MODELS = {
 	GPT_4_32K_0613: {
 		name: 'gpt-4-32k-0613',
 		tokenLimit: 32768
+	},
+	CUSTOMIZE: {
+		name: 'customize',
+		tokenLimit: 0
 	}
 }
 
@@ -89,15 +97,19 @@ export async function getChatGPTCompletion(
 	messages: openai.CreateChatCompletionRequest['messages'],
 	settings?: Partial<
 		Omit<openai.CreateChatCompletionRequest, 'messages' | 'model'>
-	>
+	>,
+	customModelName?: string
 ): Promise<string | undefined> {
 	const headers = {
 		Authorization: `Bearer ${apiKey}`,
 		'Content-Type': 'application/json'
 	}
+	const actualModel = model === CHAT_MODELS.CUSTOMIZE.name && customModelName 
+		? customModelName 
+		: (model ?? CHAT_MODELS.GPT_35_TURBO.name)
 	const body: openai.CreateChatCompletionRequest = {
 		messages,
-		model,
+		model: actualModel,
 		...settings
 	}
 	const requestParam: RequestUrlParam = {

src/settings/ChatStreamSettings.ts

@@ -1,4 +1,4 @@
-import { CHAT_MODELS, OPENAI_COMPLETIONS_URL } from 'src/openai/chatGPT'
+import { CHAT_MODELS, DEFAULT_OPENAI_COMPLETIONS_URL } from 'src/openai/chatGPT'
 
 export interface ChatStreamSettings {
 	/**
@@ -16,6 +16,11 @@ export interface ChatStreamSettings {
 	 */
 	apiModel: string
 
+	/**
+	 * Custom model name when using customize option
+	 */
+	customModelName: string
+
 	/**
 	 * The temperature to use when generating responses (0-2). 0 means no randomness.
 	 */
@@ -57,8 +62,9 @@ Use step-by-step reasoning. Be brief.
 
 export const DEFAULT_SETTINGS: ChatStreamSettings = {
 	apiKey: '',
-	apiUrl: OPENAI_COMPLETIONS_URL,
+	apiUrl: DEFAULT_OPENAI_COMPLETIONS_URL,
 	apiModel: CHAT_MODELS.GPT_35_TURBO.name,
+	customModelName: '',
 	temperature: 1,
 	systemPrompt: DEFAULT_SYSTEM_PROMPT,
 	debug: false,

src/settings/SettingsTab.ts

@@ -1,6 +1,7 @@
 import { App, PluginSettingTab, Setting } from 'obsidian'
 import { ChatStreamPlugin } from 'src/ChatStreamPlugin'
 import { getModels } from './ChatStreamSettings'
+import { CHAT_MODELS } from 'src/openai/chatGPT'
 
 export class SettingsTab extends PluginSettingTab {
 	plugin: ChatStreamPlugin
@@ -26,9 +27,26 @@ export class SettingsTab extends PluginSettingTab {
 				cb.onChange(async (value) => {
 					this.plugin.settings.apiModel = value
 					await this.plugin.saveSettings()
+					// Trigger refresh to show/hide custom model input
+					this.display()
 				})
 			})
 
+		if (this.plugin.settings.apiModel === CHAT_MODELS.CUSTOMIZE.name) {
+			new Setting(containerEl)
+				.setName('Custom Model Name')
+				.setDesc('Enter the name of your custom model')
+				.addText((text) => {
+					text
+						.setPlaceholder('e.g., gpt-4-1106-vision-preview')
+						.setValue(this.plugin.settings.customModelName)
+						.onChange(async (value) => {
+							this.plugin.settings.customModelName = value
+							await this.plugin.saveSettings()
+						})
+				})
+		}
+
 		new Setting(containerEl)
 			.setName('API key')
 			.setDesc('The API key to use when making requests - Get from OpenAI')