discoclaw 1.2.4 → 2.0.0

package/dist/voice/voice-prompt-builder.js

@@ -125,6 +125,21 @@ function estimateSection(chars) {
         included: safeChars > 0,
     };
 }
+function buildVoiceContextSections(parts) {
+    const sections = [];
+    sections.push(buildPromptPreamble(parts.identity, { skipTrackedTools: true }));
+    if (parts.actionsSection) {
+        sections.push(parts.actionsSection);
+    }
+    if (parts.voiceSystemPrompt) {
+        sections.push(parts.voiceSystemPrompt);
+    }
+    sections.push(VOICE_STYLE_INSTRUCTION);
+    if (parts.durableMemory) {
+        sections.push(`---\nDurable memory (user-specific notes):\n${parts.durableMemory}`);
+    }
+    return sections;
+}
 export function buildVoicePromptSectionEstimates(parts) {
     const charsBySection = {
         rootPolicy: ROOT_POLICY_CHARS,
@@ -162,28 +177,22 @@ export function buildVoicePromptSectionEstimates(parts) {
  * 8. User text
  */
 export function buildVoicePrompt(parts) {
-    const sections = [];
-    // 1. Root policy + identity.
-    sections.push(buildPromptPreamble(parts.identity, { skipTrackedTools: true }));
-    // 2. Actions section.
-    if (parts.actionsSection) {
-        sections.push(parts.actionsSection);
-    }
-    // 3. Voice system prompt (user-configurable).
-    if (parts.voiceSystemPrompt) {
-        sections.push(parts.voiceSystemPrompt);
-    }
-    // 4. Voice style instruction.
-    sections.push(VOICE_STYLE_INSTRUCTION);
-    // 5. Durable memory.
-    if (parts.durableMemory) {
-        sections.push(`---\nDurable memory (user-specific notes):\n${parts.durableMemory}`);
-    }
+    const sections = buildVoiceContextSections(parts);
     // 6. Separator + user text.
     sections.push(VOICE_INTERNAL_CONTEXT_SEPARATOR);
     sections.push(parts.userText);
     return sections.join('\n\n');
 }
+/**
+ * Build the static session instruction used by live voice providers.
+ *
+ * This contains the same persistent voice context as the classic per-turn
+ * prompt, excluding the current user utterance and the internal-context
+ * separator that only makes sense for single-shot prompt assembly.
+ */
+export function buildVoiceSystemInstruction(parts) {
+    return buildVoiceContextSections(parts).join('\n\n');
+}
 /**
  * Build a follow-up prompt for voice action result processing.
  *

package/dist/voice/voice-prompt-builder.test.js

@@ -1,7 +1,7 @@
 import { describe, it, expect, beforeEach, afterEach } from 'vitest';
 import fs from 'node:fs/promises';
 import path from 'node:path';
-import { extractSections, extractSoulEssentials, extractUserEssentials, loadVoiceIdentity, buildVoicePrompt, buildVoiceFollowUpPrompt, buildVoicePromptSectionEstimates, VOICE_INTERNAL_CONTEXT_SEPARATOR, VOICE_IDENTITY_MAX_CHARS, } from './voice-prompt-builder.js';
+import { extractSections, extractSoulEssentials, extractUserEssentials, loadVoiceIdentity, buildVoicePrompt, buildVoiceSystemInstruction, buildVoiceFollowUpPrompt, buildVoicePromptSectionEstimates, VOICE_INTERNAL_CONTEXT_SEPARATOR, VOICE_IDENTITY_MAX_CHARS, } from './voice-prompt-builder.js';
 import { VOICE_STYLE_INSTRUCTION } from './voice-style-prompt.js';
 import { ROOT_POLICY, TRACKED_DEFAULTS_PREAMBLE, TRACKED_TOOLS_PREAMBLE, buildPromptPreamble, } from '../discord/prompt-common.js';
 // ---------------------------------------------------------------------------
@@ -180,6 +180,21 @@ describe('loadVoiceIdentity', () => {
         expect(identityIdx).toBeLessThan(userIdx);
     });
 });
+describe('buildVoiceSystemInstruction', () => {
+    it('builds the static voice context without the user-turn separator', () => {
+        const result = buildVoiceSystemInstruction({
+            identity: 'identity block',
+            durableMemory: '',
+            actionsSection: '',
+            voiceSystemPrompt: 'custom voice system prompt',
+        });
+        expect(result).toContain(buildPromptPreamble('identity block', { skipTrackedTools: true }));
+        expect(result).toContain('custom voice system prompt');
+        expect(result).toContain(VOICE_STYLE_INSTRUCTION);
+        expect(result).not.toContain(VOICE_INTERNAL_CONTEXT_SEPARATOR);
+        expect(result).not.toContain('Current user message:');
+    });
+});
 // ---------------------------------------------------------------------------
 // buildVoicePrompt
 // ---------------------------------------------------------------------------

package/docs/configuration.md

@@ -37,7 +37,7 @@ For npm-managed daemon installs, readiness is currently constrained by service e
 Model/runtime state is intentionally split across three storage modes:
 
 - `models.json` stores persisted model strings per role (`chat`, `fast`, `plan-run`, `voice`, forge roles, cron roles, etc.).
-- `runtime-overrides.json` stores persisted runtime-only overlays such as `fastRuntime` and `voiceRuntime` (plus non-model keys such as `ttsVoice`).
+- `runtime-overrides.json` stores persisted runtime-only overlays such as `fastRuntime` and `voiceRuntime`.
 - Live chat runtime swaps stay in memory only. `!models set chat <runtime>` changes the active chat runtime immediately, but there is no persisted `chatRuntime` overlay.
 
 On first run, `models.json` is scaffolded from the instance startup defaults. After that:
@@ -322,24 +322,19 @@ The same forum-boundary rule applies to tasks: `DISCOCLAW_TASKS_FORUM` is the di
 
 ## Voice
 
-See [docs/voice.md](voice.md) for the full setup guide and provider details.
+See [docs/voice.md](voice.md) for the full Gemini Live setup guide.
 
 | Variable | Default | Description |
 |----------|---------|-------------|
 | `DISCOCLAW_VOICE_ENABLED` | `false` | Master switch for voice subsystem |
 | `DISCOCLAW_VOICE_AUTO_JOIN` | `false` | Auto-join voice channels when users enter |
 | `ANTHROPIC_API_KEY` | — | Anthropic API key (required for direct Messages API voice responses) |
+| `GEMINI_API_KEY` | — | Gemini API key required for Gemini Live voice |
 | `DISCOCLAW_VOICE_MODEL` | follows startup chat model | Model override for voice responses |
 | `DISCOCLAW_VOICE_SYSTEM_PROMPT` | — | System prompt override for voice (max 4000 chars) |
-| `DISCOCLAW_STT_PROVIDER` | `deepgram` | Speech-to-text provider: `deepgram`, `whisper`, `openai` |
-| `DISCOCLAW_TTS_PROVIDER` | `cartesia` | Text-to-speech provider: `cartesia`, `deepgram`, `kokoro`, `openai` |
+| `DISCOCLAW_GEMINI_SESSION_ROTATION_MS` | `780000` | Proactive Gemini Live session rotation interval in milliseconds |
 | `DISCOCLAW_VOICE_HOME_CHANNEL` | — | Voice channel name or ID for prompt context |
 | `DISCOCLAW_VOICE_LOG_CHANNEL` | `voice-log` | Text channel for transcript mirror |
-| `DEEPGRAM_API_KEY` | — | Deepgram API key (required for Deepgram STT/TTS) |
-| `DEEPGRAM_STT_MODEL` | `nova-3-general` | Deepgram STT model |
-| `DEEPGRAM_TTS_VOICE` | `aura-2-asteria-en` | Deepgram TTS voice |
-| `DEEPGRAM_TTS_SPEED` | `1.3` | Deepgram TTS playback speed multiplier (0.5–1.5) |
-| `CARTESIA_API_KEY` | — | Cartesia API key (required for Cartesia TTS) |
 
 ## Webhook
 

package/docs/official-docs.md

@@ -6,8 +6,8 @@ Completeness pass for this index was cross-checked against:
 
 - `package.json`
 - `.context/runtime.md`
-- `src/voice/tts-factory.ts`
-- `src/voice/stt-factory.ts`
+- `src/voice/audio-pipeline.ts`
+- `src/voice/providers/gemini-live-provider.ts`
 - `src/cold-storage/embeddings.ts`
 - `src/cold-storage/openai-compat.ts`
 - `src/discord/actions-imagegen.ts`
@@ -24,8 +24,8 @@ Completeness pass for this index was cross-checked against:
 | Provider | What DiscoClaw uses | Official docs |
 |----------|----------------------|---------------|
 | Anthropic | Claude model families via `src/runtime/anthropic-rest.ts` and Claude Code CLI runtime | Models overview: <https://docs.anthropic.com/en/docs/about-claude/models/overview><br>Messages API: <https://platform.claude.com/docs/en/api/messages><br>Claude Code docs: <https://code.claude.com/docs/en/overview> |
-| OpenAI | OpenAI-compatible runtime, Codex runtime docs, OpenAI voice, embeddings, and image generation | Model IDs: <https://developers.openai.com/api/model-ids/><br>API reference overview: <https://platform.openai.com/docs/api-reference><br>Codex docs: <https://developers.openai.com/codex/><br>Codex app-server API: <https://developers.openai.com/codex/app-server> |
-| Google | Gemini API runtime and Gemini/Imagen image generation | Gemini models: <https://ai.google.dev/models/gemini><br>Gemini API docs: <https://ai.google.dev/gemini-api/docs> |
+| OpenAI | OpenAI-compatible runtime, Codex runtime docs, embeddings, and image generation | Model IDs: <https://developers.openai.com/api/model-ids/><br>API reference overview: <https://platform.openai.com/docs/api-reference><br>Codex docs: <https://developers.openai.com/codex/><br>Codex app-server API: <https://developers.openai.com/codex/app-server> |
+| Google | Gemini API runtime, Gemini Live voice, and Gemini/Imagen image generation | Gemini models: <https://ai.google.dev/models/gemini><br>Gemini API docs: <https://ai.google.dev/gemini-api/docs><br>Gemini Live API: <https://ai.google.dev/gemini-api/docs/live> |
 | OpenRouter | OpenRouter runtime through `src/runtime/openai-compat.ts` | Model list: <https://openrouter.ai/models><br>API docs: <https://openrouter.ai/docs/api/reference/overview> |
 
 ## Discord
@@ -53,11 +53,8 @@ Completeness pass for this index was cross-checked against:
 
 | Provider | Used in DiscoClaw | Official docs |
 |----------|-------------------|---------------|
-| Deepgram STT | `src/voice/stt-deepgram.ts` with Nova-3 streaming (`nova-3-general`) | STT API overview: <https://developers.deepgram.com/docs/speech-to-text><br>Streaming API: <https://developers.deepgram.com/reference/speech-to-text/listen-streaming><br>Nova-3 models: <https://developers.deepgram.com/docs/models-languages-overview> |
-| Deepgram TTS | `src/voice/tts-deepgram.ts` with Aura (`aura-2-asteria-en`) | TTS API overview: <https://developers.deepgram.com/docs/text-to-speech><br>Speak endpoint: <https://developers.deepgram.com/reference/text-to-speech/speak-streaming><br>Aura voices/models: <https://developers.deepgram.com/docs/tts-models> |
-| Cartesia TTS | `src/voice/tts-cartesia.ts` with Sonic-3 over WebSocket | API docs: <https://docs.cartesia.ai/api-reference><br>TTS WebSocket: <https://docs.cartesia.ai/api-reference/tts/websocket> |
-| OpenAI TTS | `src/voice/tts-openai.ts` (`/v1/audio/speech`, default `tts-1`) | Audio speech API reference: <https://platform.openai.com/docs/api-reference/audio/createSpeech> |
-| OpenAI STT | `src/voice/stt-openai.ts` (`/v1/audio/transcriptions`, `whisper-1`) | Audio transcription API reference: <https://platform.openai.com/docs/api-reference/audio/createTranscription> |
+| Gemini Live | `src/voice/audio-pipeline.ts` and the Gemini Live provider handle speech recognition, reasoning, and speech synthesis in one session | Live API overview: <https://ai.google.dev/gemini-api/docs/live><br>Realtime guide: <https://ai.google.dev/gemini-api/docs/live-guide> |
+| Anthropic Messages API (optional voice runtime) | `!models set voice claude-api` can switch voice response generation to direct Anthropic API calls while Discord audio transport stays on Gemini Live | API overview: <https://docs.anthropic.com/en/api/messages> |
 
 ## Image Generation
 

package/docs/runtime-switching.md

@@ -135,7 +135,7 @@ Do not print the full `.env` into Discord, terminal transcripts, or audit logs u
 
 Keep the files separate:
 - `models.json` stores model strings per role.
-- `runtime-overrides.json` stores runtime-only overlays such as `voiceRuntime`, `fastRuntime`, and `ttsVoice`.
+- `runtime-overrides.json` stores runtime-only overlays such as `voiceRuntime` and `fastRuntime`.
 - There is no `chatRuntime` key because chat runtime swaps do not persist.
 
 On first run, `models.json` is scaffolded from the startup defaults that instance booted with.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "discoclaw",
-  "version": "1.2.4",
+  "version": "2.0.0",
   "description": "Personal AI orchestrator that turns Discord into a persistent workspace",
   "license": "MIT",
   "keywords": [