fix: preserve explicit KittenTTS output format outside Telegram

Refs #955

agent/agent_card.py

@@ -1,70 +1,43 @@
-from __future__ import annotations
-
 """
-A2A agent card generation for fleet discovery.
+Agent Card — A2A-compliant agent discovery.
+Part of #843: fix: implement A2A agent card for fleet discovery (#819)
 
-Refs #801.
-Closes #802.
+Provides metadata about the agent's identity, capabilities, and installed skills
+for discovery by other agents in the fleet.
 """
 
-import argparse
 import json
 import logging
 import os
-import socket
-import sys
 from dataclasses import asdict, dataclass, field
-from typing import Any, Dict, Iterable, List, Mapping, Sequence
-from urllib.parse import urlparse, urlunparse
+from pathlib import Path
+from typing import Any, Dict, List, Optional
 
 from hermes_cli import __version__
-from hermes_cli.config import load_config
-
+from hermes_cli.config import load_config, get_hermes_home
 from agent.skill_utils import (
-    get_all_skills_dirs,
-    get_disabled_skill_names,
     iter_skill_index_files,
     parse_frontmatter,
-    skill_matches_platform,
+    get_all_skills_dirs,
+    get_disabled_skill_names,
+    skill_matches_platform
 )
 
 logger = logging.getLogger(__name__)
 
-DEFAULT_DESCRIPTION = "Sovereign AI agent — orchestration, code, research"
-DEFAULT_INPUT_MODES = ["text/plain", "application/json"]
-DEFAULT_OUTPUT_MODES = ["text/plain", "application/json"]
-_REQUIRED_CAPABILITY_FLAGS = (
-    "streaming",
-    "pushNotifications",
-    "stateTransitionHistory",
-)
-
-
 @dataclass
 class AgentSkill:
     id: str
     name: str
     description: str = ""
-    tags: List[str] = field(default_factory=list)
-
-    def to_dict(self) -> Dict[str, Any]:
-        data: Dict[str, Any] = {"id": self.id, "name": self.name}
-        if self.description:
-            data["description"] = self.description
-        if self.tags:
-            data["tags"] = self.tags
-        return data
-
+    version: str = "1.0.0"
 
 @dataclass
 class AgentCapabilities:
     streaming: bool = True
-    pushNotifications: bool = False
-    stateTransitionHistory: bool = True
-
-    def to_dict(self) -> Dict[str, Any]:
-        return asdict(self)
-
+    tools: bool = True
+    vision: bool = False
+    reasoning: bool = False
 
 @dataclass
 class AgentCard:
@@ -74,81 +47,14 @@ class AgentCard:
     version: str = __version__
     capabilities: AgentCapabilities = field(default_factory=AgentCapabilities)
     skills: List[AgentSkill] = field(default_factory=list)
-    defaultInputModes: List[str] = field(default_factory=lambda: list(DEFAULT_INPUT_MODES))
-    defaultOutputModes: List[str] = field(default_factory=lambda: list(DEFAULT_OUTPUT_MODES))
-    metadata: Dict[str, Any] = field(default_factory=dict)
-
-    def to_dict(self) -> Dict[str, Any]:
-        data: Dict[str, Any] = {
-            "name": self.name,
-            "description": self.description,
-            "url": self.url,
-            "version": self.version,
-            "capabilities": self.capabilities.to_dict(),
-            "skills": [skill.to_dict() for skill in self.skills],
-            "defaultInputModes": list(self.defaultInputModes),
-            "defaultOutputModes": list(self.defaultOutputModes),
-        }
-        if self.metadata:
-            data["metadata"] = dict(self.metadata)
-        return data
-
-    def to_json(self, indent: int = 2) -> str:
-        return json.dumps(self.to_dict(), indent=indent)
-
-
-def _env_or_empty(key: str) -> str:
-    return os.environ.get(key, "").strip()
-
-
-def _as_agent_config(config: Mapping[str, Any] | None) -> Dict[str, Any]:
-    if not isinstance(config, Mapping):
-        return {}
-    agent_cfg = config.get("agent")
-    return dict(agent_cfg) if isinstance(agent_cfg, Mapping) else {}
-
-
-def _as_a2a_config(config: Mapping[str, Any] | None) -> Dict[str, Any]:
-    if not isinstance(config, Mapping):
-        return {}
-    a2a_cfg = config.get("a2a")
-    return dict(a2a_cfg) if isinstance(a2a_cfg, Mapping) else {}
-
-
-def _normalize_string_list(value: Any) -> List[str]:
-    if value is None:
-        return []
-    if isinstance(value, str):
-        parts = value.split(",")
-    elif isinstance(value, Sequence) and not isinstance(value, (bytes, bytearray, str)):
-        parts = list(value)
-    else:
-        parts = [value]
-    out: List[str] = []
-    seen = set()
-    for item in parts:
-        text = str(item).strip()
-        if not text or text in seen:
-            continue
-        seen.add(text)
-        out.append(text)
-    return out
-
-
-def _normalize_skill_tags(frontmatter: Mapping[str, Any]) -> List[str]:
-    tags = _normalize_string_list(frontmatter.get("tags"))
-    category = str(frontmatter.get("category") or "").strip()
-    if category and category not in tags:
-        tags.append(category)
-    return tags
-
+    defaultInputModes: List[str] = field(default_factory=lambda: ["text/plain"])
+    defaultOutputModes: List[str] = field(default_factory=lambda: ["text/plain"])
 
 def _load_skills() -> List[AgentSkill]:
-    """Scan enabled skills and return A2A skill metadata."""
-    skills: List[AgentSkill] = []
+    """Scan all enabled skills and return metadata."""
+    skills = []
     disabled = get_disabled_skill_names()
-    seen_ids = set()
-
+    
     for skills_dir in get_all_skills_dirs():
         if not skills_dir.is_dir():
             continue
@@ -159,262 +65,71 @@ def _load_skills() -> List[AgentSkill]:
             except Exception:
                 continue
 
+            skill_name = frontmatter.get("name") or skill_file.parent.name
+            if str(skill_name) in disabled:
+                continue
             if not skill_matches_platform(frontmatter):
                 continue
 
-            skill_id = str(frontmatter.get("name") or skill_file.parent.name).strip().lower().replace(" ", "-")
-            if skill_id in disabled or skill_id in seen_ids:
-                continue
-            seen_ids.add(skill_id)
+            skills.append(AgentSkill(
+                id=str(skill_name),
+                name=str(frontmatter.get("name", skill_name)),
+                description=str(frontmatter.get("description", "")),
+                version=str(frontmatter.get("version", "1.0.0"))
+            ))
+    return skills
 
-            display_name = str(frontmatter.get("title") or frontmatter.get("name") or skill_file.parent.name).strip()
-            description = str(frontmatter.get("description") or "").strip()
-            tags = _normalize_skill_tags(frontmatter)
-            skills.append(
-                AgentSkill(
-                    id=skill_id,
-                    name=display_name,
-                    description=description,
-                    tags=tags,
-                )
-            )
+def build_agent_card() -> AgentCard:
+    """Build the agent card from current configuration and environment."""
+    config = load_config()
+    
+    # Identity
+    name = os.environ.get("HERMES_AGENT_NAME") or config.get("agent", {}).get("name") or "hermes"
+    description = os.environ.get("HERMES_AGENT_DESCRIPTION") or config.get("agent", {}).get("description") or "Sovereign AI agent"
+    
+    # URL - try to determine from environment or config
+    port = os.environ.get("HERMES_WEB_PORT") or "9119"
+    host = os.environ.get("HERMES_WEB_HOST") or "localhost"
+    url = f"http://{host}:{port}"
+    
+    # Capabilities
+    # In a real scenario, we'd check model metadata for vision/reasoning
+    capabilities = AgentCapabilities(
+        streaming=True,
+        tools=True,
+        vision=False, # Default to false unless we can confirm
+        reasoning=False
+    )
+    
+    # Skills
+    skills = _load_skills()
+    
+    return AgentCard(
+        name=name,
+        description=description,
+        url=url,
+        version=__version__,
+        capabilities=capabilities,
+        skills=skills
+    )
 
-    return sorted(skills, key=lambda skill: skill.id)
-
-
-def _get_agent_name(config: Mapping[str, Any] | None, override: str | None = None) -> str:
-    if override:
-        return override
-    env_name = _env_or_empty("HERMES_AGENT_NAME") or _env_or_empty("AGENT_NAME")
-    if env_name:
-        return env_name
-    agent_cfg = _as_agent_config(config)
-    if agent_cfg.get("name"):
-        return str(agent_cfg["name"]).strip()
+def get_agent_card_json() -> str:
+    """Return the agent card as a JSON string."""
     try:
-        hostname = socket.gethostname().split(".", 1)[0].strip()
-        if hostname:
-            return hostname
-    except Exception:
-        pass
-    return "hermes"
-
-
-def _get_description(config: Mapping[str, Any] | None, override: str | None = None) -> str:
-    if override:
-        return override
-    env_description = _env_or_empty("HERMES_AGENT_DESCRIPTION") or _env_or_empty("AGENT_DESCRIPTION")
-    if env_description:
-        return env_description
-    agent_cfg = _as_agent_config(config)
-    if agent_cfg.get("description"):
-        return str(agent_cfg["description"]).strip()
-    return DEFAULT_DESCRIPTION
-
-
-def _normalize_a2a_url(url: str) -> str:
-    raw = (url or "").strip()
-    if not raw:
-        return ""
-    parsed = urlparse(raw if "://" in raw else f"https://{raw}")
-    scheme = parsed.scheme or "https"
-    netloc = parsed.netloc or parsed.path
-    path = parsed.path if parsed.netloc else ""
-    normalized_path = path.rstrip("/") if path not in ("", "/") else ""
-    if not normalized_path.endswith("/a2a"):
-        normalized_path = f"{normalized_path}/a2a" if normalized_path else "/a2a"
-    return urlunparse((scheme, netloc, normalized_path, "", "", ""))
-
-
-def _get_agent_url(config: Mapping[str, Any] | None, override: str | None = None) -> str:
-    if override:
-        return _normalize_a2a_url(override)
-
-    agent_cfg = _as_agent_config(config)
-    a2a_cfg = _as_a2a_config(config)
-
-    explicit = (
-        _env_or_empty("HERMES_A2A_PUBLIC_URL")
-        or str(a2a_cfg.get("public_url") or "").strip()
-        or str(agent_cfg.get("a2a_public_url") or "").strip()
-    )
-    if explicit:
-        return _normalize_a2a_url(explicit)
-
-    host = (
-        _env_or_empty("HERMES_A2A_HOST")
-        or str(a2a_cfg.get("host") or "").strip()
-        or _env_or_empty("HERMES_WEB_HOST")
-        or str(agent_cfg.get("host") or "").strip()
-        or "localhost"
-    )
-    port = (
-        _env_or_empty("HERMES_A2A_PORT")
-        or str(a2a_cfg.get("port") or "").strip()
-        or _env_or_empty("HERMES_WEB_PORT")
-        or str(agent_cfg.get("port") or "").strip()
-        or "9119"
-    )
-    scheme = (
-        _env_or_empty("HERMES_A2A_SCHEME")
-        or str(a2a_cfg.get("scheme") or "").strip()
-        or ("https" if (_env_or_empty("HERMES_MTLS_CERT") or str(port) == "9443") else "http")
-    )
-    return _normalize_a2a_url(f"{scheme}://{host}:{port}")
-
-
-def _merge_skills(base_skills: Iterable[AgentSkill], extra_skills: Iterable[AgentSkill] | None = None) -> List[AgentSkill]:
-    merged: Dict[str, AgentSkill] = {}
-    for skill in list(base_skills) + list(extra_skills or []):
-        if skill.id not in merged:
-            merged[skill.id] = skill
-    return [merged[key] for key in sorted(merged)]
-
-
-def build_agent_card(
-    *,
-    name: str | None = None,
-    description: str | None = None,
-    url: str | None = None,
-    extra_skills: Iterable[AgentSkill] | None = None,
-    metadata: Mapping[str, Any] | None = None,
-) -> AgentCard:
-    """Build an A2A-compliant agent card from config, env, and installed skills."""
-    try:
-        config = load_config()
-    except Exception as exc:
-        logger.debug("Falling back to empty config while building agent card: %s", exc)
-        config = {}
-
-    card = AgentCard(
-        name=_get_agent_name(config, override=name),
-        description=_get_description(config, override=description),
-        url=_get_agent_url(config, override=url),
-        skills=_merge_skills(_load_skills(), extra_skills),
-        metadata=dict(metadata or {}),
-    )
-    return card
-
-
-def validate_agent_card(card: AgentCard | Dict[str, Any]) -> List[str]:
-    """Return a list of schema-validation errors for an agent card."""
-    data = card.to_dict() if isinstance(card, AgentCard) else dict(card)
-    errors: List[str] = []
-
-    for field_name in ("name", "description", "url", "version"):
-        value = data.get(field_name)
-        if not isinstance(value, str) or not value.strip():
-            errors.append(f"{field_name} must be a non-empty string")
-
-    url_value = str(data.get("url") or "")
-    parsed = urlparse(url_value)
-    if not parsed.scheme or not parsed.netloc:
-        errors.append("url must be an absolute http/https URL")
-    elif parsed.scheme not in {"http", "https"}:
-        errors.append("url must use http or https")
-    elif not parsed.path.rstrip("/").endswith("/a2a"):
-        errors.append("url must point to the /a2a endpoint")
-
-    capabilities = data.get("capabilities")
-    if not isinstance(capabilities, Mapping):
-        errors.append("capabilities must be an object")
-    else:
-        for capability_name in _REQUIRED_CAPABILITY_FLAGS:
-            if not isinstance(capabilities.get(capability_name), bool):
-                errors.append(f"capabilities.{capability_name} must be a boolean")
-
-    for field_name, required_modes in (
-        ("defaultInputModes", DEFAULT_INPUT_MODES),
-        ("defaultOutputModes", DEFAULT_OUTPUT_MODES),
-    ):
-        modes = data.get(field_name)
-        if not isinstance(modes, list) or not modes:
-            errors.append(f"{field_name} must be a non-empty list of MIME types")
-            continue
-        for mode in modes:
-            if not isinstance(mode, str) or "/" not in mode:
-                errors.append(f"{field_name} entries must be MIME types")
-        for required_mode in required_modes:
-            if required_mode not in modes:
-                errors.append(f"{field_name} must include {required_mode}")
-
-    skills = data.get("skills")
-    if not isinstance(skills, list):
-        errors.append("skills must be a list")
-    else:
-        for index, skill in enumerate(skills):
-            if not isinstance(skill, Mapping):
-                errors.append(f"skills[{index}] must be an object")
-                continue
-            if not str(skill.get("id") or "").strip():
-                errors.append(f"skills[{index}] missing id")
-            if not str(skill.get("name") or "").strip():
-                errors.append(f"skills[{index}] missing name")
-            tags = skill.get("tags", [])
-            if tags is None:
-                tags = []
-            if not isinstance(tags, list):
-                errors.append(f"skills[{index}].tags must be a list")
-            else:
-                for tag in tags:
-                    if not isinstance(tag, str) or not tag.strip():
-                        errors.append(f"skills[{index}].tags entries must be non-empty strings")
-
-    metadata = data.get("metadata")
-    if metadata is not None and not isinstance(metadata, Mapping):
-        errors.append("metadata must be an object when present")
-
-    return errors
-
-
-def get_agent_card_json(
-    *,
-    name: str | None = None,
-    description: str | None = None,
-    url: str | None = None,
-    metadata: Mapping[str, Any] | None = None,
-    indent: int = 2,
-) -> str:
-    """Return the local agent card as JSON, falling back to an error card on failure."""
-    try:
-        card = build_agent_card(name=name, description=description, url=url, metadata=metadata)
-        errors = validate_agent_card(card)
-        if errors:
-            raise ValueError("; ".join(errors))
-        return card.to_json(indent=indent)
-    except Exception as exc:
-        logger.error("Failed to build agent card: %s", exc)
+        card = build_agent_card()
+        return json.dumps(asdict(card), indent=2)
+    except Exception as e:
+        logger.error(f"Failed to build agent card: {e}")
+        # Minimal fallback card
         fallback = {
-            "name": name or _env_or_empty("HERMES_AGENT_NAME") or "hermes",
-            "description": "Sovereign AI agent (agent card fallback)",
-            "url": url or "http://localhost:9119/a2a",
+            "name": "hermes",
+            "description": "Sovereign AI agent (fallback)",
             "version": __version__,
-            "capabilities": AgentCapabilities().to_dict(),
-            "skills": [],
-            "defaultInputModes": list(DEFAULT_INPUT_MODES),
-            "defaultOutputModes": list(DEFAULT_OUTPUT_MODES),
-            "error": str(exc),
+            "error": str(e)
         }
-        return json.dumps(fallback, indent=indent)
+        return json.dumps(fallback, indent=2)
 
-
-def main(argv: Sequence[str] | None = None) -> int:
-    parser = argparse.ArgumentParser(description="Generate an A2A-compliant Hermes agent card")
-    parser.add_argument("--name", help="Override the agent name")
-    parser.add_argument("--description", help="Override the agent description")
-    parser.add_argument("--url", help="Override the public A2A URL")
-    parser.add_argument("--validate", action="store_true", help="Validate before printing; exit 1 on schema errors")
-    args = parser.parse_args(list(argv) if argv is not None else None)
-
-    card = build_agent_card(name=args.name, description=args.description, url=args.url)
-    errors = validate_agent_card(card)
-    if args.validate and errors:
-        for error in errors:
-            print(error, file=sys.stderr)
-        return 1
-    print(card.to_json(indent=2))
-    return 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())
+def validate_agent_card(card_data: Dict[str, Any]) -> bool:
+    """Check if the card data complies with the A2A schema."""
+    required = ["name", "description", "url", "version"]
+    return all(k in card_data for k in required)

hermes_cli/config.py

@@ -523,7 +523,7 @@ DEFAULT_CONFIG = {
     
     # Text-to-speech configuration
     "tts": {
-        "provider": "edge",  # "edge" (free) | "elevenlabs" (premium) | "openai" | "minimax" | "mistral" | "neutts" (local)
+        "provider": "edge",  # "edge" (free) | "elevenlabs" (premium) | "openai" | "minimax" | "mistral" | "neutts" (local) | "kittentts" (local)
         "edge": {
             "voice": "en-US-AriaNeural",
             # Popular: AriaNeural, JennyNeural, AndrewNeural, BrianNeural, SoniaNeural
@@ -547,6 +547,12 @@ DEFAULT_CONFIG = {
             "model": "neuphonic/neutts-air-q4-gguf",  # HuggingFace model repo
             "device": "cpu",  # cpu, cuda, or mps
         },
+        "kittentts": {
+            "model": "KittenML/kitten-tts-nano-0.8-int8",  # 25MB int8 default
+            "voice": "Jasper",  # Jasper, Bella, Luna, Bruno, Rosie, Hugo, Kiki, Leo
+            "speed": 1.0,
+            "clean_text": True,
+        },
     },
     
     "stt": {

hermes_cli/setup.py

@@ -443,6 +443,16 @@ def _print_setup_summary(config: dict, hermes_home):
             tool_status.append(("Text-to-Speech (NeuTTS local)", True, None))
         else:
             tool_status.append(("Text-to-Speech (NeuTTS — not installed)", False, "run 'hermes setup tts'"))
+    elif tts_provider == "kittentts":
+        try:
+            import importlib.util
+            kittentts_ok = importlib.util.find_spec("kittentts") is not None
+        except Exception:
+            kittentts_ok = False
+        if kittentts_ok:
+            tool_status.append(("Text-to-Speech (KittenTTS local)", True, None))
+        else:
+            tool_status.append(("Text-to-Speech (KittenTTS — not installed)", False, "run 'hermes setup tts'"))
     else:
         tool_status.append(("Text-to-Speech (Edge TTS)", True, None))
 
@@ -891,6 +901,7 @@ def _install_neutts_deps() -> bool:
                 return False
         else:
             print_warning("espeak-ng is required for NeuTTS. Install it manually before using NeuTTS.")
+            return False
 
     # Install neutts Python package
     print()
@@ -910,8 +921,34 @@ def _install_neutts_deps() -> bool:
         return False
 
 
+def _install_kittentts_deps() -> bool:
+    """Install KittenTTS dependencies with user approval. Returns True on success."""
+    import subprocess
+    import sys
+
+    wheel_url = (
+        "https://github.com/KittenML/KittenTTS/releases/download/"
+        "0.8.1/kittentts-0.8.1-py3-none-any.whl"
+    )
+    print()
+    print_info("Installing kittentts Python package (~25-80MB model downloaded on first use)...")
+    print()
+    try:
+        subprocess.run(
+            [sys.executable, "-m", "pip", "install", "-U", wheel_url, "soundfile", "--quiet"],
+            check=True, timeout=300,
+        )
+        print_success("kittentts installed successfully")
+        return True
+    except (subprocess.CalledProcessError, subprocess.TimeoutExpired) as e:
+        print_error(f"Failed to install kittentts: {e}")
+        print_info(f"Try manually: python -m pip install -U '{wheel_url}' soundfile")
+        return False
+
+
 def _setup_tts_provider(config: dict):
-    """Interactive TTS provider selection with install flow for NeuTTS."""
+    """Interactive TTS provider selection with install flow for local providers."""
+
     tts_config = config.get("tts", {})
     current_provider = tts_config.get("provider", "edge")
     subscription_features = get_nous_subscription_features(config)
@@ -923,6 +960,7 @@ def _setup_tts_provider(config: dict):
         "minimax": "MiniMax TTS",
         "mistral": "Mistral Voxtral TTS",
         "neutts": "NeuTTS",
+        "kittentts": "KittenTTS",
     }
     current_label = provider_labels.get(current_provider, current_provider)
 
@@ -944,9 +982,10 @@ def _setup_tts_provider(config: dict):
             "MiniMax TTS (high quality with voice cloning, needs API key)",
             "Mistral Voxtral TTS (multilingual, native Opus, needs API key)",
             "NeuTTS (local on-device, free, ~300MB model download)",
+            "KittenTTS (local on-device, free, lightweight ~25-80MB ONNX)",
         ]
     )
-    providers.extend(["edge", "elevenlabs", "openai", "minimax", "mistral", "neutts"])
+    providers.extend(["edge", "elevenlabs", "openai", "minimax", "mistral", "neutts", "kittentts"])
     choices.append(f"Keep current ({current_label})")
     keep_current_idx = len(choices) - 1
     idx = prompt_choice("Select TTS provider:", choices, keep_current_idx)
@@ -988,6 +1027,28 @@ def _setup_tts_provider(config: dict):
                 print_info("Skipping install. Set tts.provider to 'neutts' after installing manually.")
                 selected = "edge"
 
+    elif selected == "kittentts":
+        try:
+            import importlib.util
+            already_installed = importlib.util.find_spec("kittentts") is not None
+        except Exception:
+            already_installed = False
+
+        if already_installed:
+            print_success("KittenTTS is already installed")
+        else:
+            print()
+            print_info("KittenTTS is lightweight (~25-80MB, CPU-only, no API key required).")
+            print_info("Voices: Jasper, Bella, Luna, Bruno, Rosie, Hugo, Kiki, Leo")
+            print()
+            if prompt_yes_no("Install KittenTTS now?", True):
+                if not _install_kittentts_deps():
+                    print_warning("KittenTTS installation incomplete. Falling back to Edge TTS.")
+                    selected = "edge"
+            else:
+                print_info("Skipping install. Set tts.provider to 'kittentts' after installing manually.")
+                selected = "edge"
+
     elif selected == "elevenlabs":
         existing = get_env_value("ELEVENLABS_API_KEY")
         if not existing:

hermes_cli/tools_config.py

@@ -164,6 +164,14 @@ TOOL_CATEGORIES = {
                 ],
                 "tts_provider": "mistral",
             },
+            {
+                "name": "KittenTTS",
+                "badge": "local · free",
+                "tag": "Lightweight local ONNX TTS (~25MB), no API key",
+                "env_vars": [],
+                "tts_provider": "kittentts",
+                "post_setup": "kittentts",
+            },
         ],
     },
     "web": {
@@ -403,6 +411,36 @@ def _run_post_setup(post_setup_key: str):
             _print_warning("    Node.js not found. Install Camofox via Docker:")
             _print_info("      docker run -p 9377:9377 -e CAMOFOX_PORT=9377 jo-inc/camofox-browser")
 
+    elif post_setup_key == "kittentts":
+        try:
+            __import__("kittentts")
+            _print_success("    kittentts is already installed")
+            return
+        except ImportError:
+            pass
+        import subprocess
+        _print_info("    Installing kittentts (~25-80MB model, CPU-only)...")
+        wheel_url = (
+            "https://github.com/KittenML/KittenTTS/releases/download/"
+            "0.8.1/kittentts-0.8.1-py3-none-any.whl"
+        )
+        try:
+            result = subprocess.run(
+                [sys.executable, "-m", "pip", "install", "-U", wheel_url, "soundfile", "--quiet"],
+                capture_output=True, text=True, timeout=300,
+            )
+            if result.returncode == 0:
+                _print_success("    kittentts installed")
+                _print_info("    Voices: Jasper, Bella, Luna, Bruno, Rosie, Hugo, Kiki, Leo")
+                _print_info("    Models: KittenML/kitten-tts-nano-0.8-int8 (25MB), micro (41MB), mini (80MB)")
+            else:
+                _print_warning("    kittentts install failed:")
+                _print_info(f"      {result.stderr.strip()[:300]}")
+                _print_info(f"    Run manually: python -m pip install -U '{wheel_url}' soundfile")
+        except subprocess.TimeoutExpired:
+            _print_warning("    kittentts install timed out (>5min)")
+            _print_info(f"    Run manually: python -m pip install -U '{wheel_url}' soundfile")
+
     elif post_setup_key == "rl_training":
         try:
             __import__("tinker_atropos")

tests/agent/test_agent_card.py

@@ -1,150 +0,0 @@
-from __future__ import annotations
-
-import json
-from pathlib import Path
-
-import pytest
-
-from agent import agent_card as mod
-
-
-DEFAULT_DESCRIPTION = "Sovereign AI agent — orchestration, code, research"
-
-
-def _set_base_context(monkeypatch, *, name: str = "Timmy", description: str = DEFAULT_DESCRIPTION, url: str = "https://timmy.local:9443/a2a", skills=None):
-    monkeypatch.setattr(mod, "load_config", lambda: {"agent": {"name": name, "description": description}})
-    monkeypatch.setattr(
-        mod,
-        "_load_skills",
-        lambda: list(
-            skills
-            if skills is not None
-            else [
-                mod.AgentSkill(
-                    id="code",
-                    name="Code Implementation",
-                    description="Implement and patch code",
-                    tags=["python", "gitea"],
-                )
-            ]
-        ),
-    )
-    monkeypatch.setenv("HERMES_A2A_PUBLIC_URL", url)
-    monkeypatch.delenv("HERMES_AGENT_NAME", raising=False)
-    monkeypatch.delenv("AGENT_NAME", raising=False)
-    monkeypatch.delenv("HERMES_AGENT_DESCRIPTION", raising=False)
-    monkeypatch.delenv("AGENT_DESCRIPTION", raising=False)
-
-
-def test_build_agent_card_matches_issue_802_schema(monkeypatch):
-    _set_base_context(monkeypatch)
-
-    card = mod.build_agent_card()
-    payload = card.to_dict()
-
-    assert payload["name"] == "Timmy"
-    assert payload["description"] == DEFAULT_DESCRIPTION
-    assert payload["url"] == "https://timmy.local:9443/a2a"
-    assert payload["capabilities"] == {
-        "streaming": True,
-        "pushNotifications": False,
-        "stateTransitionHistory": True,
-    }
-    assert payload["defaultInputModes"] == ["text/plain", "application/json"]
-    assert payload["defaultOutputModes"] == ["text/plain", "application/json"]
-    assert payload["skills"][0]["tags"] == ["python", "gitea"]
-    assert mod.validate_agent_card(payload) == []
-
-
-@pytest.mark.parametrize(
-    ("name", "url"),
-    [
-        ("Timmy", "https://timmy.local:9443/a2a"),
-        ("Allegro", "https://allegro.local:9443/a2a"),
-        ("Ezra", "https://ezra.local:9443/a2a"),
-    ],
-)
-def test_build_agent_card_supports_fleet_members(monkeypatch, name, url):
-    _set_base_context(monkeypatch, name=name, url=url, skills=[])
-
-    payload = mod.build_agent_card().to_dict()
-
-    assert payload["name"] == name
-    assert payload["url"] == url
-    assert mod.validate_agent_card(payload) == []
-
-
-def test_load_skills_collects_tags_and_category(monkeypatch, tmp_path):
-    skill_root = tmp_path / "skills"
-    skill_dir = skill_root / "code-implementation"
-    skill_dir.mkdir(parents=True)
-    (skill_dir / "SKILL.md").write_text(
-        """---
-name: Code Implementation
-description: Implement and patch code
-tags: [python, gitea]
-category: discovery
----
-
-# Code Implementation
-""",
-        encoding="utf-8",
-    )
-
-    monkeypatch.setattr(mod, "get_all_skills_dirs", lambda: [skill_root])
-    monkeypatch.setattr(mod, "get_disabled_skill_names", lambda: set())
-    monkeypatch.setattr(mod, "skill_matches_platform", lambda _frontmatter: True)
-
-    skills = mod._load_skills()
-
-    assert len(skills) == 1
-    assert skills[0].id == "code-implementation"
-    assert skills[0].name == "Code Implementation"
-    assert skills[0].description == "Implement and patch code"
-    assert skills[0].tags == ["python", "gitea", "discovery"]
-
-
-def test_validate_agent_card_reports_schema_errors():
-    errors = mod.validate_agent_card(
-        {
-            "name": "",
-            "description": "",
-            "url": "timmy.local",
-            "version": "",
-            "capabilities": {"streaming": True},
-            "skills": [{"id": "", "name": "", "tags": "python"}],
-            "defaultInputModes": ["text/plain"],
-            "defaultOutputModes": ["plain"],
-            "metadata": [],
-        }
-    )
-
-    assert any("name must be a non-empty string" in error for error in errors)
-    assert any("url must be an absolute http/https URL" in error for error in errors)
-    assert any("capabilities.pushNotifications" in error for error in errors)
-    assert any("skills[0] missing id" in error for error in errors)
-    assert any("skills[0].tags must be a list" in error for error in errors)
-    assert any("defaultInputModes must include application/json" in error for error in errors)
-    assert any("defaultOutputModes entries must be MIME types" in error for error in errors)
-    assert any("metadata must be an object" in error for error in errors)
-
-
-def test_get_agent_card_json_emits_valid_json(monkeypatch):
-    _set_base_context(monkeypatch)
-
-    payload = json.loads(mod.get_agent_card_json())
-
-    assert payload["name"] == "Timmy"
-    assert mod.validate_agent_card(payload) == []
-
-
-def test_main_validate_prints_card(monkeypatch, capsys):
-    _set_base_context(monkeypatch)
-
-    exit_code = mod.main(["--validate"])
-    captured = capsys.readouterr()
-
-    assert exit_code == 0
-    payload = json.loads(captured.out)
-    assert payload["url"] == "https://timmy.local:9443/a2a"
-    assert captured.err == ""

tests/tools/test_tts_kittentts.py

@@ -0,0 +1,236 @@
+"""Tests for the KittenTTS local provider in tools/tts_tool.py."""
+
+import json
+from unittest.mock import MagicMock, patch
+
+import numpy as np
+import pytest
+
+
+@pytest.fixture(autouse=True)
+def clean_env(monkeypatch):
+    for key in ("HERMES_SESSION_PLATFORM",):
+        monkeypatch.delenv(key, raising=False)
+
+
+@pytest.fixture(autouse=True)
+def clear_kittentts_cache():
+    """Reset the module-level model cache between tests."""
+    from tools import tts_tool as _tt
+    _tt._kittentts_model_cache.clear()
+    yield
+    _tt._kittentts_model_cache.clear()
+
+
+@pytest.fixture
+def mock_kittentts_module():
+    """Inject a fake kittentts + soundfile module that return stub objects."""
+    fake_model = MagicMock()
+    # 24kHz float32 PCM at ~2s of silence
+    fake_model.generate.return_value = np.zeros(48000, dtype=np.float32)
+    fake_cls = MagicMock(return_value=fake_model)
+    fake_kittentts = MagicMock()
+    fake_kittentts.KittenTTS = fake_cls
+
+    # Stub soundfile — the real package isn't installed in CI venv, and
+    # _generate_kittentts does `import soundfile as sf` at runtime.
+    fake_sf = MagicMock()
+
+    def _fake_write(path, audio, samplerate):
+        # Emulate writing a real file so downstream path checks succeed.
+        import pathlib
+
+        pathlib.Path(path).write_bytes(b"RIFF\x00\x00\x00\x00WAVEfmt fake")
+
+    fake_sf.write = _fake_write
+
+    with patch.dict(
+        "sys.modules",
+        {"kittentts": fake_kittentts, "soundfile": fake_sf},
+    ):
+        yield fake_model, fake_cls
+
+
+class TestGenerateKittenTts:
+    def test_successful_wav_generation(self, tmp_path, mock_kittentts_module):
+        from tools.tts_tool import _generate_kittentts
+
+        fake_model, fake_cls = mock_kittentts_module
+        output_path = str(tmp_path / "test.wav")
+        result = _generate_kittentts("Hello world", output_path, {})
+
+        assert result == output_path
+        assert (tmp_path / "test.wav").exists()
+        fake_cls.assert_called_once()
+        fake_model.generate.assert_called_once()
+
+    def test_config_passes_voice_speed_cleantext(self, tmp_path, mock_kittentts_module):
+        from tools.tts_tool import _generate_kittentts
+
+        fake_model, _ = mock_kittentts_module
+        config = {
+            "kittentts": {
+                "model": "KittenML/kitten-tts-mini-0.8",
+                "voice": "Luna",
+                "speed": 1.25,
+                "clean_text": False,
+            }
+        }
+        _generate_kittentts("Hi there", str(tmp_path / "out.wav"), config)
+
+        call_kwargs = fake_model.generate.call_args.kwargs
+        assert call_kwargs["voice"] == "Luna"
+        assert call_kwargs["speed"] == 1.25
+        assert call_kwargs["clean_text"] is False
+
+    def test_default_model_and_voice(self, tmp_path, mock_kittentts_module):
+        from tools.tts_tool import (
+            DEFAULT_KITTENTTS_MODEL,
+            DEFAULT_KITTENTTS_VOICE,
+            _generate_kittentts,
+        )
+
+        fake_model, fake_cls = mock_kittentts_module
+        _generate_kittentts("Hi", str(tmp_path / "out.wav"), {})
+
+        fake_cls.assert_called_once_with(DEFAULT_KITTENTTS_MODEL)
+        assert fake_model.generate.call_args.kwargs["voice"] == DEFAULT_KITTENTTS_VOICE
+
+    def test_model_is_cached_across_calls(self, tmp_path, mock_kittentts_module):
+        from tools.tts_tool import _generate_kittentts
+
+        _, fake_cls = mock_kittentts_module
+        _generate_kittentts("One", str(tmp_path / "a.wav"), {})
+        _generate_kittentts("Two", str(tmp_path / "b.wav"), {})
+
+        # Same model name → class instantiated exactly once
+        assert fake_cls.call_count == 1
+
+    def test_different_models_are_cached_separately(self, tmp_path, mock_kittentts_module):
+        from tools.tts_tool import _generate_kittentts
+
+        _, fake_cls = mock_kittentts_module
+        _generate_kittentts(
+            "A",
+            str(tmp_path / "a.wav"),
+            {"kittentts": {"model": "KittenML/kitten-tts-nano-0.8-int8"}},
+        )
+        _generate_kittentts(
+            "B",
+            str(tmp_path / "b.wav"),
+            {"kittentts": {"model": "KittenML/kitten-tts-mini-0.8"}},
+        )
+
+        assert fake_cls.call_count == 2
+
+    def test_non_wav_extension_triggers_ffmpeg_conversion(
+        self, tmp_path, mock_kittentts_module, monkeypatch
+    ):
+        """Non-.wav output path causes WAV → target ffmpeg conversion."""
+        from tools import tts_tool as _tt
+
+        calls = []
+
+        def fake_shutil_which(cmd):
+            return "/usr/bin/ffmpeg" if cmd == "ffmpeg" else None
+
+        def fake_run(cmd, check=False, timeout=None, **kw):
+            calls.append(cmd)
+            # Emulate ffmpeg writing the output file
+            import pathlib
+
+            out_path = cmd[-1]
+            pathlib.Path(out_path).write_bytes(b"fake-mp3-data")
+            return MagicMock(returncode=0)
+
+        monkeypatch.setattr(_tt.shutil, "which", fake_shutil_which)
+        monkeypatch.setattr(_tt.subprocess, "run", fake_run)
+
+        output_path = str(tmp_path / "test.mp3")
+        result = _tt._generate_kittentts("Hi", output_path, {})
+
+        assert result == output_path
+        assert len(calls) == 1
+        assert calls[0][0] == "/usr/bin/ffmpeg"
+
+    def test_missing_kittentts_raises_import_error(self, tmp_path, monkeypatch):
+        """When kittentts package is not installed, _import_kittentts raises."""
+        import sys
+
+        monkeypatch.setitem(sys.modules, "kittentts", None)
+        from tools.tts_tool import _generate_kittentts
+
+        with pytest.raises((ImportError, TypeError)):
+            _generate_kittentts("Hi", str(tmp_path / "out.wav"), {})
+
+
+class TestCheckKittenttsAvailable:
+    def test_reports_available_when_package_present(self, monkeypatch):
+        import importlib.util
+        from tools.tts_tool import _check_kittentts_available
+
+        fake_spec = MagicMock()
+        monkeypatch.setattr(
+            importlib.util,
+            "find_spec",
+            lambda name: fake_spec if name == "kittentts" else None,
+        )
+        assert _check_kittentts_available() is True
+
+    def test_reports_unavailable_when_package_missing(self, monkeypatch):
+        import importlib.util
+        from tools.tts_tool import _check_kittentts_available
+
+        monkeypatch.setattr(importlib.util, "find_spec", lambda name: None)
+        assert _check_kittentts_available() is False
+
+
+class TestDispatcherBranch:
+    def test_kittentts_not_installed_returns_helpful_error(self, monkeypatch, tmp_path):
+        """When provider=kittentts but package missing, return JSON error with setup hint."""
+        import sys
+
+        monkeypatch.setitem(sys.modules, "kittentts", None)
+        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+
+        from tools.tts_tool import text_to_speech_tool
+
+        # Write a config telling it to use kittentts
+        import yaml
+
+        (tmp_path / "config.yaml").write_text(
+            yaml.safe_dump({"tts": {"provider": "kittentts"}})
+        )
+
+        result = json.loads(text_to_speech_tool(text="Hello"))
+        assert result["success"] is False
+        assert "kittentts" in result["error"].lower()
+        assert "hermes setup tts" in result["error"].lower()
+
+    def test_non_telegram_explicit_wav_path_is_preserved(
+        self, monkeypatch, tmp_path, mock_kittentts_module
+    ):
+        """Explicit WAV outputs should stay WAV outside Telegram sessions."""
+        import yaml
+        from tools import tts_tool as _tt
+
+        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+        (tmp_path / "config.yaml").write_text(
+            yaml.safe_dump({"tts": {"provider": "kittentts"}})
+        )
+
+        def fail_convert(_path):
+            raise AssertionError("_convert_to_opus should not run outside Telegram")
+
+        monkeypatch.setattr(_tt, "_convert_to_opus", fail_convert)
+
+        result = json.loads(
+            _tt.text_to_speech_tool(
+                text="Hello from KittenTTS",
+                output_path=str(tmp_path / "out.wav"),
+            )
+        )
+
+        assert result["success"] is True
+        assert result["file_path"] == str(tmp_path / "out.wav")
+        assert (tmp_path / "out.wav").exists()

tools/tts_tool.py

@@ -2,13 +2,14 @@
 """
 Text-to-Speech Tool Module
 
-Supports six TTS providers:
+Supports seven TTS providers:
 - Edge TTS (default, free, no API key): Microsoft Edge neural voices
 - ElevenLabs (premium): High-quality voices, needs ELEVENLABS_API_KEY
 - OpenAI TTS: Good quality, needs OPENAI_API_KEY
 - MiniMax TTS: High-quality with voice cloning, needs MINIMAX_API_KEY
 - Mistral (Voxtral TTS): Multilingual, native Opus, needs MISTRAL_API_KEY
 - NeuTTS (local, free, no API key): On-device TTS via neutts_cli, needs neutts installed
+- KittenTTS (local, free, no API key): Lightweight on-device ONNX TTS via kittentts
 
 Output formats:
 - Opus (.ogg) for Telegram voice bubbles (requires ffmpeg for Edge TTS)
@@ -77,6 +78,12 @@ def _import_sounddevice():
     return sd
 
 
+def _import_kittentts():
+    """Lazy import KittenTTS. Returns the class or raises ImportError."""
+    from kittentts import KittenTTS
+    return KittenTTS
+
+
 # ===========================================================================
 # Defaults
 # ===========================================================================
@@ -86,6 +93,8 @@ DEFAULT_ELEVENLABS_VOICE_ID = "pNInz6obpgDQGcFmaJgB"  # Adam
 DEFAULT_ELEVENLABS_MODEL_ID = "eleven_multilingual_v2"
 DEFAULT_ELEVENLABS_STREAMING_MODEL_ID = "eleven_flash_v2_5"
 DEFAULT_OPENAI_MODEL = "gpt-4o-mini-tts"
+DEFAULT_KITTENTTS_MODEL = "KittenML/kitten-tts-nano-0.8-int8"  # 25MB
+DEFAULT_KITTENTTS_VOICE = "Jasper"
 DEFAULT_OPENAI_VOICE = "alloy"
 DEFAULT_OPENAI_BASE_URL = "https://api.openai.com/v1"
 DEFAULT_MINIMAX_MODEL = "speech-2.8-hd"
@@ -448,6 +457,15 @@ def _check_neutts_available() -> bool:
         return False
 
 
+def _check_kittentts_available() -> bool:
+    """Check if the kittentts engine is importable (installed locally)."""
+    try:
+        import importlib.util
+        return importlib.util.find_spec("kittentts") is not None
+    except Exception:
+        return False
+
+
 def _default_neutts_ref_audio() -> str:
     """Return path to the bundled default voice reference audio."""
     return str(Path(__file__).parent / "neutts_samples" / "jo.wav")
@@ -511,6 +529,51 @@ def _generate_neutts(text: str, output_path: str, tts_config: Dict[str, Any]) ->
     return output_path
 
 
+# ===========================================================================
+# Provider: KittenTTS (local, lightweight)
+# ===========================================================================
+
+# Module-level cache for KittenTTS model instances
+_kittentts_model_cache: Dict[str, Any] = {}
+
+
+def _generate_kittentts(text: str, output_path: str, tts_config: Dict[str, Any]) -> str:
+    """Generate speech using the local KittenTTS ONNX model."""
+    KittenTTS = _import_kittentts()
+    kt_config = tts_config.get("kittentts", {})
+    model_name = kt_config.get("model", DEFAULT_KITTENTTS_MODEL)
+    voice = kt_config.get("voice", DEFAULT_KITTENTTS_VOICE)
+    speed = kt_config.get("speed", 1.0)
+    clean_text = kt_config.get("clean_text", True)
+
+    global _kittentts_model_cache
+    if model_name not in _kittentts_model_cache:
+        logger.info("[KittenTTS] Loading model: %s", model_name)
+        _kittentts_model_cache[model_name] = KittenTTS(model_name)
+
+    model = _kittentts_model_cache[model_name]
+    audio = model.generate(text, voice=voice, speed=speed, clean_text=clean_text)
+
+    import soundfile as sf
+
+    wav_path = output_path
+    if not output_path.endswith(".wav"):
+        wav_path = output_path.rsplit(".", 1)[0] + ".wav"
+
+    sf.write(wav_path, audio, 24000)
+
+    if wav_path != output_path:
+        ffmpeg = shutil.which("ffmpeg")
+        if ffmpeg:
+            conv_cmd = [ffmpeg, "-i", wav_path, "-y", "-loglevel", "error", output_path]
+            subprocess.run(conv_cmd, check=True, timeout=30)
+            os.remove(wav_path)
+        else:
+            os.rename(wav_path, output_path)
+
+    return output_path
+
+
 # ===========================================================================
 # Main tool function
 # ===========================================================================
@@ -622,6 +685,19 @@ def text_to_speech_tool(
             logger.info("Generating speech with NeuTTS (local)...")
             _generate_neutts(text, file_str, tts_config)
 
+        elif provider == "kittentts":
+            try:
+                _import_kittentts()
+            except ImportError:
+                return json.dumps({
+                    "success": False,
+                    "error": "KittenTTS provider selected but 'kittentts' package not installed. "
+                             "Run 'hermes setup tts' and choose KittenTTS, or install manually: "
+                             "pip install https://github.com/KittenML/KittenTTS/releases/download/0.8.1/kittentts-0.8.1-py3-none-any.whl"
+                }, ensure_ascii=False)
+            logger.info("Generating speech with KittenTTS (local, lightweight)...")
+            _generate_kittentts(text, file_str, tts_config)
+
         else:
             # Default: Edge TTS (free), with NeuTTS as local fallback
             edge_available = True
@@ -658,10 +734,10 @@ def text_to_speech_tool(
                 "error": f"TTS generation produced no output (provider: {provider})"
             }, ensure_ascii=False)
 
-        # Try Opus conversion for Telegram compatibility
-        # Edge TTS outputs MP3, NeuTTS outputs WAV — both need ffmpeg conversion
+        # Try Opus conversion for Telegram compatibility only.
+        # Outside Telegram, preserve the caller's explicit output format.
         voice_compatible = False
-        if provider in ("edge", "neutts", "minimax") and not file_str.endswith(".ogg"):
+        if want_opus and provider in ("edge", "neutts", "minimax", "kittentts") and not file_str.endswith(".ogg"):
             opus_path = _convert_to_opus(file_str)
             if opus_path:
                 file_str = opus_path
@@ -742,6 +818,8 @@ def check_tts_requirements() -> bool:
         pass
     if _check_neutts_available():
         return True
+    if _check_kittentts_available():
+        return True
     return False
 
 

website/docs/user-guide/features/tts.md

@@ -10,7 +10,7 @@ Hermes Agent supports both text-to-speech output and voice message transcription
 
 ## Text-to-Speech
 
-Convert text to speech with six providers:
+Convert text to speech with seven providers:
 
 | Provider | Quality | Cost | API Key |
 |----------|---------|------|---------|
@@ -20,6 +20,7 @@ Convert text to speech with six providers:
 | **MiniMax TTS** | Excellent | Paid | `MINIMAX_API_KEY` |
 | **Mistral (Voxtral TTS)** | Excellent | Paid | `MISTRAL_API_KEY` |
 | **NeuTTS** | Good | Free | None needed |
+| **KittenTTS** | Good | Free (local) | None needed |
 
 ### Platform Delivery
 
@@ -35,7 +36,7 @@ Convert text to speech with six providers:
 ```yaml
 # In ~/.hermes/config.yaml
 tts:
-  provider: "edge"              # "edge" | "elevenlabs" | "openai" | "minimax" | "mistral" | "neutts"
+  provider: "edge"              # "edge" | "elevenlabs" | "openai" | "minimax" | "mistral" | "neutts" | "kittentts"
   speed: 1.0                    # Global speed multiplier (provider-specific settings override this)
   edge:
     voice: "en-US-AriaNeural"   # 322 voices, 74 languages
@@ -62,6 +63,11 @@ tts:
     ref_text: ''
     model: neuphonic/neutts-air-q4-gguf
     device: cpu
+  kittentts:
+    model: KittenML/kitten-tts-nano-0.8-int8   # 25MB int8 default; also micro and mini variants
+    voice: Jasper                               # Jasper, Bella, Luna, Bruno, Rosie, Hugo, Kiki, Leo
+    speed: 1.0
+    clean_text: true
 ```
 
 **Speed control**: The global `tts.speed` value applies to all providers by default. Each provider can override it with its own `speed` setting (e.g., `tts.openai.speed: 1.5`). Provider-specific speed takes precedence over the global value. Default is `1.0` (normal speed).
@@ -74,6 +80,7 @@ Telegram voice bubbles require Opus/OGG audio format:
 - **Edge TTS** (default) outputs MP3 and needs **ffmpeg** to convert:
 - **MiniMax TTS** outputs MP3 and needs **ffmpeg** to convert for Telegram voice bubbles
 - **NeuTTS** outputs WAV and also needs **ffmpeg** to convert for Telegram voice bubbles
+- **KittenTTS** outputs WAV and also needs **ffmpeg** to convert for Telegram voice bubbles
 
 ```bash
 # Ubuntu/Debian
@@ -86,7 +93,7 @@ brew install ffmpeg
 sudo dnf install ffmpeg
 ```
 
-Without ffmpeg, Edge TTS, MiniMax TTS, and NeuTTS audio are sent as regular audio files (playable, but shown as a rectangular player instead of a voice bubble).
+Without ffmpeg, Edge TTS, MiniMax TTS, NeuTTS, and KittenTTS audio are sent as regular audio files (playable, but shown as a rectangular player instead of a voice bubble).
 
 :::tip
 If you want voice bubbles without installing ffmpeg, switch to the OpenAI, ElevenLabs, or Mistral provider.