BMO/server/index.ts

// Tiny HTTP server for BMO's brain. POST /api/chat streams a reply as
// server-sent events; /api/brain reads and sets the API key from the side
// menu. In production it also serves the built SPA from dist/.
//
// The API key stays on this side; the browser never sees it.

import { createReadStream, existsSync, statSync } from "node:fs";
import { createServer, type IncomingMessage, type ServerResponse } from "node:http";
import path from "node:path";
import { fileURLToPath } from "node:url";
import Anthropic from "@anthropic-ai/sdk";
import { currentKey, keyHint, looksLikeKey, storeKey } from "./keystore.ts";
import { BMO_SYSTEM_PROMPT, MOODS, type Mood } from "./persona.ts";
import { BMO_TOOLS, describeAction, describeScene, parseScene, toAction } from "./tools.ts";

const PORT = Number(process.env.BMO_PORT ?? 4310);
// Haiku while testing: fast and cheap. Set BMO_MODEL=claude-opus-5-5 for the big brain.
const MODEL = process.env.BMO_MODEL?.trim() || "claude-haiku-5-5";
// Server-side refusal fallback only exists on the larger models; Haiku rejects it.
const SUPPORTS_FALLBACK = /^claude-(opus-5|opus-5-5|fable-5-1|sonnet-5-5)$/.test(MODEL);
const MAX_HISTORY = 40;
const MAX_MESSAGE_CHARS = 4000;

const here = path.dirname(fileURLToPath(import.meta.url));
const distDir = path.resolve(here, "../dist");

// Rebuilt whenever the key changes from the side menu.
let client = new Anthropic();

type ChatTurn = { role: "user" | "assistant"; content: string };

function hasCredentials() {
  return Boolean(currentKey() || process.env.ANTHROPIC_AUTH_TOKEN);
}

/**
 * Only same-machine pages may call the API. Browsers always send Origin on
 * cross-site POSTs, so this stops other websites from spending your credits
 * or swapping your key through this local server.
 */
function isTrustedRequest(req: IncomingMessage) {
  const origin = req.headers.origin;
  if (origin) {
    let host: string;
    try {
      host = new URL(origin).hostname;
    } catch {
      return false;
    }
    if (host !== "localhost" && host !== "127.0.0.1" && host !== "[::1]") return false;
  }
  return (req.headers["content-type"] ?? "").startsWith("application/json");
}

function sendJson(res: ServerResponse, status: number, body: unknown) {
  res.writeHead(status, { "content-type": "application/json" });
  res.end(JSON.stringify(body));
}

function brainStatus() {
  return { connected: hasCredentials(), keyHint: keyHint(), model: MODEL };
}

async function handleSetKey(req: IncomingMessage, res: ServerResponse) {
  let body: unknown;
  try {
    body = await readJson(req);
  } catch {
    return sendJson(res, 400, { error: "Bad request." });
  }
  const raw = (body as { apiKey?: unknown } | null)?.apiKey;
  if (raw === null) {
    storeKey(null);
    client = new Anthropic();
    return sendJson(res, 200, brainStatus());
  }
  const key = typeof raw === "string" ? raw.trim() : "";
  if (!looksLikeKey(key)) {
    return sendJson(res, 400, { error: "That doesn't look like an Anthropic API key. It should start with sk-ant-." });
  }
  // Try the key before saving it, with a free metadata call.
  const candidate = new Anthropic({ apiKey: key });
  try {
    await candidate.models.retrieve(MODEL);
  } catch (err) {
    if (err instanceof Anthropic.AuthenticationError) {
      return sendJson(res, 400, { error: "Anthropic rejected that key. Double-check you copied all of it." });
    }
    if (err instanceof Anthropic.PermissionDeniedError || err instanceof Anthropic.NotFoundError) {
      return sendJson(res, 400, { error: `That key can't use ${MODEL}.` });
    }
    if (err instanceof Anthropic.APIConnectionError) {
      return sendJson(res, 502, { error: "Couldn't reach Anthropic to check the key. Are you online?" });
    }
    // Rate limits or server hiccups don't mean the key is bad; save it anyway.
    console.warn("[bmo] key check inconclusive:", err);
  }
  storeKey(key);
  client = candidate;
  console.log(`[bmo] API key saved (${keyHint(key)})`);
  sendJson(res, 200, brainStatus());
}

function readJson(req: IncomingMessage): Promise<unknown> {
  return new Promise((resolve, reject) => {
    let body = "";
    req.on("data", (chunk) => {
      body += chunk;
      if (body.length > 200_000) reject(new Error("body too large"));
    });
    req.on("end", () => {
      try {
        resolve(JSON.parse(body || "{}"));
      } catch (err) {
        reject(err);
      }
    });
    req.on("error", reject);
  });
}

function parseHistory(input: unknown): ChatTurn[] | null {
  if (!input || typeof input !== "object" || !Array.isArray((input as { messages?: unknown }).messages)) {
    return null;
  }
  const raw = (input as { messages: unknown[] }).messages;
  const turns: ChatTurn[] = [];
  for (const m of raw) {
    if (!m || typeof m !== "object") return null;
    const { role, content } = m as { role?: unknown; content?: unknown };
    if ((role !== "user" && role !== "assistant") || typeof content !== "string") return null;
    turns.push({ role, content: content.slice(0, MAX_MESSAGE_CHARS) });
  }
  // Trim to the most recent turns and make sure the first one is from the user.
  const recent = turns.slice(-MAX_HISTORY);
  while (recent.length && recent[0]!.role !== "user") recent.shift();
  if (!recent.length || recent[recent.length - 1]!.role !== "user") return null;
  return recent;
}

function sse(res: ServerResponse, event: string, data: unknown) {
  res.write(`event: ${event}\ndata: ${JSON.stringify(data)}\n\n`);
}

/**
 * Splits the leading "[mood] " tag off the streamed reply. Text is buffered
 * until the tag is resolved (or clearly absent), then flows through untouched.
 */
function createMoodSplitter(onMood: (mood: Mood) => void, onText: (text: string) => void) {
  let buffer = "";
  let resolved = false;
  return {
    push(chunk: string) {
      if (resolved) return onText(chunk);
      buffer += chunk;
      const match = buffer.match(/^\s*\[([a-z]+)\]\s*/i);
      if (match) {
        const mood = match[1]!.toLowerCase();
        onMood((MOODS as readonly string[]).includes(mood) ? (mood as Mood) : "happy");
        resolved = true;
        const rest = buffer.slice(match[0].length);
        if (rest) onText(rest);
      } else if (buffer.length > 20 || !/^\s*\[?[a-z]*$/i.test(buffer)) {
        onMood("happy");
        resolved = true;
        onText(buffer);
      }
    },
    flush() {
      if (!resolved && buffer) {
        onMood("happy");
        onText(buffer);
      }
    },
  };
}

const MAX_TOOL_ROUNDS = 4;

async function handleChat(req: IncomingMessage, res: ServerResponse) {
  let body: unknown;
  try {
    body = await readJson(req);
  } catch {
    body = null;
  }
  const history = parseHistory(body);
  if (!history) {
    return sendJson(res, 400, { error: "Expected { messages: [{ role, content }] } ending with a user turn." });
  }
  const scene = parseScene((body as { scene?: unknown }).scene);

  res.writeHead(200, {
    "content-type": "text/event-stream",
    "cache-control": "no-cache",
    connection: "keep-alive",
  });

  if (!hasCredentials()) {
    sse(res, "mood", "oops");
    sse(res, "text", "BMO can't reach its brain yet! Open the ⋮ menu and paste your API key under Brain.");
    sse(res, "done", {});
    res.end();
    return;
  }

  // The live scene rides along as a mid-conversation system message after the
  // newest user turn, so the cached system prompt + history prefix stays stable.
  const messages: Anthropic.Beta.BetaMessageParam[] = [...history];
  if (scene) messages.push({ role: "system", content: describeScene(scene) });

  let aborted = false;
  let current: { abort: () => void } | null = null;
  req.on("close", () => {
    aborted = true;
    current?.abort();
  });

  // Text from later rounds (after a tool call) continues the same bubble,
  // separated from the earlier words by a space.
  let wroteText = false;
  const writeText = (text: string, startsSegment = true) => {
    if (!text) return;
    if (startsSegment && wroteText) text = " " + text.trimStart();
    wroteText = true;
    sse(res, "text", text);
  };

  try {
    for (let round = 0; round < MAX_TOOL_ROUNDS && !aborted; round++) {
      let roundText = "";
      const splitter = createMoodSplitter(
        (mood) => sse(res, "mood", mood),
        (text) => {
          writeText(roundText ? text : text.trimStart(), !roundText);
          roundText += text;
        },
      );

      const stream = client.beta.messages.stream({
        model: MODEL,
        max_tokens: 4000,
        output_config: { effort: "low" },
        // Server-side fallback: if a safety classifier declines, the API retries on
        // a fallback model inside the same call instead of BMO going silent.
        ...(SUPPORTS_FALLBACK ? { betas: ["server-side-fallback-2026-07-01"], fallbacks: "default" as const } : {}),
        cache_control: { type: "ephemeral" },
        system: BMO_SYSTEM_PROMPT,
        tools: BMO_TOOLS,
        tool_choice: { type: "auto" },
        messages,
      });
      current = stream;

      for await (const event of stream) {
        if (event.type === "content_block_delta" && event.delta.type === "text_delta") {
          splitter.push(event.delta.text);
        }
      }
      splitter.flush();

      let final: Anthropic.Beta.BetaMessage;
      try {
        final = await stream.finalMessage();
      } catch (err) {
        // A tool input that couldn't be parsed at all: skip the actions, keep the words.
        if (err instanceof Anthropic.APIError) throw err;
        console.warn("[bmo] unparseable tool input:", err);
        break;
      }

      if (final.stop_reason === "refusal") {
        sse(res, "mood", "oops");
        writeText("Hmm, BMO can't talk about that one.");
        break;
      }
      const toolUses = final.content.filter((b): b is Anthropic.Beta.BetaToolUseBlock => b.type === "tool_use");
      // A tool call cut off at max_tokens may carry a truncated input; don't run it.
      if (final.stop_reason !== "tool_use" || toolUses.length === 0) break;

      const results: Anthropic.Beta.BetaToolResultBlockParam[] = toolUses.map((use) => {
        const action = toAction(use.name, use.input);
        if ("error" in action) {
          return { type: "tool_result", tool_use_id: use.id, content: action.error, is_error: true };
        }
        sse(res, "action", action);
        return { type: "tool_result", tool_use_id: use.id, content: describeAction(action) };
      });
      messages.push({ role: "assistant", content: final.content });
      messages.push({ role: "user", content: results });
    }
    sse(res, "done", {});
  } catch (err) {
    if (!res.writableEnded && !aborted) {
      let message = "BMO's brain hiccupped. Try again?";
      if (err instanceof Anthropic.AuthenticationError) message = "BMO's API key doesn't work. Check it in the ⋮ menu under Brain!";
      else if (err instanceof Anthropic.RateLimitError) message = "BMO is talking too fast. Give me a second!";
      else if (err instanceof Anthropic.APIError) message = `BMO's brain said error ${err.status}.`;
      console.error("[bmo] chat error:", err);
      sse(res, "mood", "oops");
      writeText(message);
      sse(res, "done", {});
    }
  } finally {
    res.end();
  }
}

const MIME: Record<string, string> = {
  ".html": "text/html; charset=utf-8",
  ".js": "text/javascript",
  ".css": "text/css",
  ".svg": "image/svg+xml",
  ".png": "image/png",
  ".ico": "image/x-icon",
  ".json": "application/json",
};

function serveStatic(req: IncomingMessage, res: ServerResponse) {
  const urlPath = decodeURIComponent(new URL(req.url ?? "/", "http://x").pathname);
  let file = path.join(distDir, urlPath);
  if (!file.startsWith(distDir) || !existsSync(file) || statSync(file).isDirectory()) {
    file = path.join(distDir, "index.html");
  }
  if (!existsSync(file)) {
    res.writeHead(404).end("Build the app first: npm run build");
    return;
  }
  res.writeHead(200, { "content-type": MIME[path.extname(file)] ?? "application/octet-stream" });
  createReadStream(file).pipe(res);
}

createServer((req, res) => {
  if (req.url?.startsWith("/api/") && req.method === "POST" && !isTrustedRequest(req)) {
    return sendJson(res, 403, { error: "Forbidden." });
  }
  if (req.method === "POST" && req.url === "/api/chat") {
    void handleChat(req, res);
    return;
  }
  if (req.method === "GET" && req.url === "/api/brain") {
    return sendJson(res, 200, brainStatus());
  }
  if (req.method === "POST" && req.url === "/api/brain") {
    void handleSetKey(req, res);
    return;
  }
  if (req.method === "GET" && req.url === "/api/health") {
    return sendJson(res, 200, { ok: true, ...brainStatus() });
  }
  serveStatic(req, res);
}).listen(PORT, "127.0.0.1", () => {
  console.log(`[bmo] brain listening at http://127.0.0.1:${PORT} (model ${MODEL}, key ${hasCredentials() ? "found" : "missing"})`);
});