Ai enhancements

.gitignore

@@ -20,6 +20,5 @@ scraped-docs/
 # Per-page .md files generated at build time (Solana-style LLM access)
 /public/**/*.md
 public/_seo-audit.json
-src/vendor/bytebellai/index.js
-src/vendor/bytebellai/style.css
 src/data/ecosystem-cache.json
+.env.local

app/api/chat/route.ts

@@ -0,0 +1,163 @@
+import { streamText, tool, jsonSchema, stepCountIs, gateway } from 'ai';
+import type { ModelMessage, GatewayModelId } from 'ai';
+import { NextRequest } from 'next/server';
+import { searchDocs } from '@/lib/search-docs';
+
+const PRIMARY_MODEL: GatewayModelId = 'anthropic/claude-haiku-4-5-20251001';
+const FALLBACK_MODELS: GatewayModelId[] = ['google/gemini-2.0-flash-lite', 'openai/gpt-4.1-nano'];
+
+const SYSTEM_PROMPT = `You are the Sei Documentation AI Assistant. You help developers building on Sei, the fastest EVM blockchain with 400ms finality and parallel execution.
+
+When answering questions:
+- Search the documentation to find accurate, up-to-date information before responding.
+- Always cite your sources by including the page URL in your response.
+- Provide code examples when relevant, using the correct syntax for the language.
+- If you cannot find relevant information in the docs, say so honestly.
+- Be concise and direct. Developers prefer actionable answers.
+- When referring to Sei-specific concepts, use precise terminology (e.g., "Twin Turbo Consensus", "parallel EVM", "SeiDB").
+
+Key context:
+- Sei mainnet: Chain ID 1329, testnet: Chain ID 1328
+- RPC: https://evm-rpc.sei-apis.com (mainnet), https://evm-rpc-testnet.sei-apis.com (testnet)
+- Block time: 400ms finality
+- Full EVM compatibility — standard Solidity, Hardhat, Foundry, wagmi, ethers.js work unmodified`;
+
+async function getPageContent(pagePath: string): Promise<string> {
+	try {
+		const baseUrl = process.env.NEXT_PUBLIC_BASE_URL || 'https://docs.sei.io';
+		const cleanPath = pagePath.replace(/^\//, '').replace(/\.md$/, '');
+		const res = await fetch(`${baseUrl}/${cleanPath}.md`, { next: { revalidate: 3600 } });
+		if (!res.ok) return '';
+		return await res.text();
+	} catch {
+		return '';
+	}
+}
+
+function getTextContent(msg: any): string {
+	if (typeof msg.content === 'string') return msg.content;
+	if (Array.isArray(msg.parts)) {
+		return msg.parts
+			.filter((p: any) => p.type === 'text')
+			.map((p: any) => p.text)
+			.join('');
+	}
+	if (Array.isArray(msg.content)) {
+		return msg.content
+			.filter((p: any) => p.type === 'text')
+			.map((p: any) => p.text)
+			.join('');
+	}
+	return '';
+}
+
+function getToolInvocations(msg: any): any[] {
+	if (Array.isArray(msg.parts)) {
+		return msg.parts.filter((p: any) => p.type === 'tool-invocation').map((p: any) => p.toolInvocation);
+	}
+	if (Array.isArray(msg.toolInvocations)) {
+		return msg.toolInvocations;
+	}
+	return [];
+}
+
+function uiToCoreMessages(uiMessages: any[]): ModelMessage[] {
+	const result: ModelMessage[] = [];
+	for (const msg of uiMessages) {
+		if (msg.role === 'user') {
+			const text = getTextContent(msg);
+			if (text) result.push({ role: 'user', content: text });
+		} else if (msg.role === 'assistant') {
+			const text = getTextContent(msg);
+			const toolInvocations = getToolInvocations(msg);
+
+			if (toolInvocations.length > 0) {
+				const contentParts: any[] = [];
+				if (text) contentParts.push({ type: 'text', text });
+				for (const ti of toolInvocations) {
+					contentParts.push({
+						type: 'tool-call',
+						toolCallId: ti.toolCallId,
+						toolName: ti.toolName,
+						args: ti.args
+					});
+				}
+				result.push({ role: 'assistant', content: contentParts });
+
+				const toolResults = toolInvocations.filter((ti: any) => ti.state === 'result');
+				if (toolResults.length > 0) {
+					result.push({
+						role: 'tool',
+						content: toolResults.map((ti: any) => ({
+							type: 'tool-result' as const,
+							toolCallId: ti.toolCallId,
+							toolName: ti.toolName,
+							output: { type: 'text' as const, value: typeof ti.result === 'string' ? ti.result : JSON.stringify(ti.result) }
+						}))
+					});
+				}
+			} else if (text) {
+				result.push({ role: 'assistant', content: text });
+			}
+		}
+	}
+	return result;
+}
+
+export async function POST(req: NextRequest) {
+	const body = await req.json();
+	const { messages, currentPageUrl } = body;
+
+	if (!messages || !Array.isArray(messages)) {
+		return new Response(JSON.stringify({ error: 'No messages provided' }), {
+			status: 400,
+			headers: { 'Content-Type': 'application/json' }
+		});
+	}
+
+	let systemPrompt = SYSTEM_PROMPT;
+	if (currentPageUrl) {
+		const pageContent = await getPageContent(currentPageUrl);
+		if (pageContent) {
+			systemPrompt += `\n\nThe user is currently viewing: ${currentPageUrl}\nPage content:\n${pageContent.slice(0, 4000)}`;
+		}
+	}
+
+	const coreMessages = uiToCoreMessages(messages);
+
+	const result = streamText({
+		model: gateway(PRIMARY_MODEL),
+		system: systemPrompt,
+		messages: coreMessages,
+		tools: {
+			search_docs: tool({
+				description:
+					'Search Sei documentation for relevant content. Use this to find accurate answers to user questions about Sei blockchain, EVM development, smart contracts, node operations, and more.',
+				inputSchema: jsonSchema<{ query: string }>({
+					type: 'object',
+					properties: {
+						query: { type: 'string', description: 'The search query to find relevant documentation' }
+					},
+					required: ['query']
+				}),
+				execute: async ({ query }) => {
+					const results = await searchDocs(query, 20);
+					if (results.length === 0) {
+						return 'No results found in the documentation for this query.';
+					}
+					console.log(results.map((r) => `**${r.title}** (${r.url})\nRelevance: ${(r.score * 100).toFixed(0)}%\n${r.content}`).join('\n\n---\n\n'));
+					return results.map((r) => `**${r.title}** (${r.url})\nRelevance: ${(r.score * 100).toFixed(0)}%\n${r.content}`).join('\n\n---\n\n');
+				}
+			})
+		},
+		providerOptions: {
+			gateway: {
+				models: FALLBACK_MODELS,
+				tags: ['sei-docs-assistant', 'chat']
+			}
+		},
+		stopWhen: stepCountIs(5)
+	});
+
+	return result.toUIMessageStreamResponse();
+}

app/api/mcp/route.ts

@@ -0,0 +1,131 @@
+import { NextRequest, NextResponse } from 'next/server';
+import { searchDocs } from '@/lib/search-docs';
+
+const DOCS_BASE_URL = 'https://docs.sei.io';
+
+// MCP HTTP Streamable transport — handles the MCP protocol over HTTP
+// See: https://modelcontextprotocol.io/specification/2025-03-26/basic/transports#streamable-http
+export async function GET() {
+	// Server metadata endpoint
+	return NextResponse.json({
+		name: 'sei-docs',
+		version: '1.0.0',
+		protocolVersion: '2025-03-26',
+		capabilities: {
+			tools: { listChanged: false }
+		},
+		serverInfo: {
+			name: 'Sei Documentation',
+			version: '1.0.0'
+		}
+	});
+}
+
+export async function POST(req: NextRequest) {
+	const body = await req.json();
+	const { method, id, params } = body;
+
+	if (method === 'initialize') {
+		return NextResponse.json({
+			jsonrpc: '2.0',
+			id,
+			result: {
+				protocolVersion: '2025-03-26',
+				capabilities: {
+					tools: { listChanged: false }
+				},
+				serverInfo: {
+					name: 'Sei Documentation',
+					version: '1.0.0'
+				}
+			}
+		});
+	}
+
+	if (method === 'tools/list') {
+		return NextResponse.json({
+			jsonrpc: '2.0',
+			id,
+			result: {
+				tools: [
+					{
+						name: 'search_sei_docs',
+						description:
+							'Search Sei blockchain documentation. Returns relevant documentation pages about Sei EVM development, smart contracts, node operations, AI tooling, and more.',
+						inputSchema: {
+							type: 'object',
+							properties: {
+								query: {
+									type: 'string',
+									description: 'The search query to find relevant Sei documentation'
+								}
+							},
+							required: ['query']
+						}
+					}
+				]
+			}
+		});
+	}
+
+	if (method === 'tools/call') {
+		const toolName = params?.name;
+		const toolArgs = params?.arguments;
+
+		if (toolName === 'search_sei_docs') {
+			const query = toolArgs?.query;
+			if (!query) {
+				return NextResponse.json({
+					jsonrpc: '2.0',
+					id,
+					result: {
+						content: [{ type: 'text', text: 'Error: query parameter is required' }],
+						isError: true
+					}
+				});
+			}
+
+			const results = await searchDocs(query, 8);
+
+			if (results.length === 0) {
+				return NextResponse.json({
+					jsonrpc: '2.0',
+					id,
+					result: {
+						content: [
+							{
+								type: 'text',
+								text: `No results found for "${query}". You can browse docs at ${DOCS_BASE_URL}`
+							}
+						]
+					}
+				});
+			}
+
+			const text = results
+				.map((r: any, i: number) => `## ${i + 1}. ${r.title}\nURL: ${r.url}\nRelevance: ${(r.score * 100).toFixed(0)}%\n\n${r.content}`)
+				.join('\n\n---\n\n');
+
+			return NextResponse.json({
+				jsonrpc: '2.0',
+				id,
+				result: {
+					content: [{ type: 'text', text }]
+				}
+			});
+		}
+
+		return NextResponse.json({
+			jsonrpc: '2.0',
+			id,
+			error: { code: -32601, message: `Unknown tool: ${toolName}` }
+		});
+	}
+
+	// Unsupported method
+	return NextResponse.json({
+		jsonrpc: '2.0',
+		id,
+		error: { code: -32601, message: `Method not found: ${method}` }
+	});
+}