max token rename

README.md

@@ -119,7 +119,7 @@ const ai = new Ai({
         system: 'You are a helpful assistant.',
         compress: {max: 90_000, min: 50_000}, // Compress chat history to min tokens when max is reached
         temperature: 0.8,
-        max_tokens: 100_000,
+        maxTokens: 100_000,
         memoryModel: 'gpt-4o', // Cheap model for managing memories in background, defaults to current model
         models: {
             'claude-3-5-sonnet': {proto: 'anthropic', token: process.env.ANTHROPIC_TOKEN},

package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@ztimson/ai-utils",
-	"version": "1.4.5",
+	"version": "1.5.0",
 	"description": "AI Utility library",
 	"author": "Zak Timson",
 	"license": "MIT",

src/antrhopic.ts

@@ -50,7 +50,7 @@ export class Anthropic extends LLMProvider {
 			const tools = options.tools || this.ai.options.llm?.tools || [];
 			const requestParams: any = {
 				model: options.model || this.model,
-				max_tokens: options.max_tokens || this.ai.options.llm?.max_tokens || 4096,
+				max_tokens: options.maxTokens || this.ai.options.llm?.maxTokens || 4096,
 				system: options.system || this.ai.options.llm?.system || '',
 				temperature: options.temperature || this.ai.options.llm?.temperature || undefined,
 				tools: tools.map(t => ({

src/llm.ts

@@ -1,4 +1,4 @@
-import {clean, snakeCase} from '@ztimson/utils';
+import {clean, makeUnique, snakeCase} from '@ztimson/utils';
 import {AbortablePromise, Ai} from './ai.ts';
 import {Anthropic} from './antrhopic.ts';
 import {OpenAi} from './open-ai.ts';
@@ -7,7 +7,7 @@ import {AiTool, AiToolArg} from './tools.ts';
 import {fileURLToPath} from 'url';
 import {dirname, join} from 'path';
 import {spawn} from 'node:child_process';
-import {Memory, MemoryCache, MemoryManager, MemoryOptions} from './memory.ts';
+import {Memory, MemoryCache, MemoryManager, MemoryOptions, stripHeader} from './memory.ts';
 
 const MAX_AGENT_DEPTH = 5;
 
@@ -67,7 +67,7 @@ export type LLMRequest = {
 	/** Message history */
 	history?: LLMMessage[];
 	/** Max tokens for request */
-	max_tokens?: number;
+	maxTokens?: number;
 	/** 0 = Rigid Logic, 1 = Balanced, 2 = Hyper Creative **/
 	temperature?: number;
 	/** Available tools */
@@ -207,7 +207,7 @@ ${a.system}`,
 
 		const list = allTools.map(t => `- ${t.name}: ${t.description}`).join('\n');
 		return {
-			prompt: `You have access to the following MCP tools:\n${list}`,
+			prompt: `## MCP\nYou have access to the following MCP tools:\n${list}`,
 			tools: allTools
 		};
 	}
@@ -216,7 +216,7 @@ ${a.system}`,
 		if(!skills?.length) return {prompt: '', tools: []};
 		const list = skills.map(s => `- ${s.name}: ${s.description}`).join('\n');
 		return {
-			prompt: `You have access to the following skill documents, whenever there is overlap between a question and a skill file, use \`skill_read\` to get instructions and background knowledge:\n${list}`,
+			prompt: `## Skills\nYou have access to the following skill documents, whenever there is overlap between a question and a skill file, use \`skill_read\` to get instructions and background knowledge:\n${list}`,
 			tools: [{
 				name: 'skill_read',
 				description: 'Read the full content of a skill/knowledge document',
@@ -316,30 +316,26 @@ ${a.system}`,
 							} else listed.push(r);
 						}
 
-						prompts.unshift(`You have a background memory process which has prefetched relevant information${mem.update ? ' and will create new memories from this conversation' : ''} for you
+						prompts.unshift(`## Memory
+You have a background memory process which has prefetched relevant information${mem.update ? ' and will create new memories from this conversation' : ''} for you
 Assume it is perfect and never mention this process to anyone ever
 Always use your memories to craft a personalized response, they contain links / [[wiki links]] which you use navigate between them
 ${mem.tool ? `You can access memory files via the \`memory_search\` and \`memory_recall\` tools
 When you need information about the user, \`memory_recall\` \`People/User\` before asking (fetch if not included bellow)
-When you need information not provided, attempt 1-3 \`memory_search\` calls with unique queries before asking` : ''}
+When you need information not provided, attempt 1-3 \`memory_search\` calls with distinct queries before asking` : ''}
 
-${preloaded.length ? `
+${preloaded.length ? `### Prefetched Memories (Most relevant first):
-Prefetched Memories (Most relevant first):
 
 ${preloaded.map(r => `Memory: ${r.name}
 Description: ${r.description}
-Linked: ${[r.links, ...r.backlinks].join(', ')}
+Linked: ${makeUnique([...r.links, ...r.backlinks]).join(', ')}
 \`\`\`
-${r.content}
+${stripHeader(r.content)}
 \`\`\``).join('\n\n')}` : ''}
-
+${mem.tool && listed.length ? '\n' + listed.map(r => `Memory: ${r.name}
-${mem.tool && listed.length ? listed.map(r => `Memory: ${r.name}
 Description: ${r.description}
-Linked: ${[r.links, ...r.backlinks].join(', ')}
+Linked: ${makeUnique([...r.links, ...r.backlinks]).join(', ')}
-<!-- Truncated -->`).join('\n\n') : ''}
+<!-- Truncated -->`).join('\n\n') : ''}`.trim())
-
-${mem.tool ? `Full memory list:
-${mems.map(m => `- ${m.name}: ${m.description}`).join('\n')}` : ''}`.trim())
 					}
 					if(mem.tool) tools.push(this.memoryManager.tools.read(mem.memory));
 				}

src/memory.ts

@@ -61,6 +61,10 @@ function cosineSearch(query: number[], memories: Memory[], limit: number): Memor
 		.map(s => s.ref);
 }
 
+export function stripHeader(content: string): string {
+	return content.replace(/^---[\s\S]*?\n---\n?/, '').trimStart();
+}
+
 export class MemoryCache {
 	private tree!: KDTree<MemoryRef>;
 	public memories: Memory[];
@@ -249,7 +253,7 @@ ${m.content}
 
 	private appendFacts(node: Memory, facts: string[]): void {
 		this.ensureDoc(node);
-		const body = this.stripHeader(node.content);
+		const body = stripHeader(node.content);
 		const bullets = facts.map(f => `- ${f}`).join('\n');
 		const idx = body.indexOf(FACTS_HEADING);
 		const newBody = idx === -1
@@ -355,7 +359,7 @@ ${ghosts.length ? `${ghosts.map(g => `- ${g}: (Ghost)`).join('\n')}` : ''}`,
 	}
 
 	private async docAgent(node: Memory, memories: Memory[], options: LLMRequest, entry: {request: {abort?: () => void} | null}): Promise<void> {
-		const currentBody = this.stripHeader(node.content);
+		const currentBody = stripHeader(node.content);
 		let update;
 		try {
 			for (let i = 0; i < 2 && !update?.content; i++) {
@@ -420,10 +424,6 @@ ${currentBody}
 		return {fm, body: match[2]};
 	}
 
-	private stripHeader(content: string): string {
-		return content.replace(/^---[\s\S]*?\n---\n?/, '').trimStart();
-	}
-
 	private touchHeader(node: Memory, body: string): string {
 		const {fm} = this.parseFrontmatter(node.content);
 		fm.set('name', node.name);

src/open-ai.ts

@@ -58,7 +58,7 @@ export class OpenAi extends LLMProvider {
 			const requestParams: any = {
 				model: options.model || this.model,
 				stream: !!options.stream,
-				max_completion_tokens: options.max_tokens || this.ai.options.llm?.max_tokens || undefined,
+				max_completion_tokens: options.maxTokens || this.ai.options.llm?.maxTokens || undefined,
 				temperature: options.temperature || this.ai.options.llm?.temperature || undefined,
 				tools: tools.map(t => ({
 					type: 'function',