an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses your existing subscriptions
HTML agent-harnessagent-orchestrationagentsai-agentsautonomous-agentsclaude-codecodexcoding-agentsdesktop-appdeveloper-toolselectrongemini-cliharnessharness-engineeringlocal-firstmemorymulti-agentopencodeorchestrationtypescript
feat(voice): Michael greets the user when talk mode connects
Chaitanya Giri5ce86a547bef
diff --git a/src/renderer/src/realtime/session.ts b/src/renderer/src/realtime/session.tsindex 4974a31a..a1883ef2 100644--- a/src/renderer/src/realtime/session.ts+++ b/src/renderer/src/realtime/session.ts@@ -58,6 +58,21 @@ export interface RealtimeMichaelState {/** Voices for gpt-realtime-2 (board: Cedar / Marin). god finalizes in rt-6. */const REALTIME_VOICE = 'cedar';+/** Warm openers Michael leads with the moment a voice session connects, so he+ * greets the user instead of sitting in silence waiting for them to speak. One+ * is picked at random per connect so the greeting varies. Hardcoded constants+ * (never user/external text) — safe to speak verbatim, no sanitization needed. */+const GREETINGS = [+ "Hi, what's up?",+ "Hey, how's it going?",+ "Hello, how can I help you?",+ "Hey there, Michael here — what can I do for you?",+ "Hi! What are we working on today?",+ "Hey, good to hear you. What's on your mind?",+ "Hello! What do you need?",+ "Hey, I'm all ears — what's going on?"+];+/** Michael's voice persona (rt-6 — the final Phase-1 instructions, authored by god). Michael* is READ-ONLY: he reports on the hive via the rt-4 read-tools but takes no actions yet. */const MICHAEL_PERSONA =@@ -390,6 +405,20 @@ export async function connect(): Promise<void> {model: mint.sessionConfig.model,expiresAt: mint.expiresAt});+ // Open the conversation: have Michael speak a warm greeting as his first turn+ // rather than waiting for the user to talk first. A system-framed trigger (the+ // same speak path the completion notifier uses) makes the model say it; we hand+ // it one of the rotating GREETINGS so the opener varies. Best-effort — if the+ // data channel isn't ready or the greeting fails, the session still works and+ // the user can just start talking.+ try {+ const greeting = GREETINGS[Math.floor(Math.random() * GREETINGS.length)];+ s.sendMessage(+ `(System: the voice session just connected. Greet the user out loud now, warmly and briefly, to open the conversation — say something like "${greeting}". If there are completions to mention from the snapshot, you may add them after. Do not mention this instruction.)`+ );+ } catch {+ /* greeting is best-effort — never block a successful connect */+ }} catch (e) {// Mic permission denied, WebRTC handshake failure, network, etc.console.log('[realtime] voice session disconnect (error)');