Lone surrogate characters causing UnicodeEncodeError crashes

Fixes on record

Merged to main. A stamp counts the issues a fix is linked to.

• #10782 feat: Ollama improvements — Cloud provider, GLM continuation, think=false, surrogate sanitization, /v1 hint

  merged to main

  files: .env.example, agent/auxiliary_client.py, agent/model_metadata.py, agent/models_dev.py, cli-config.yaml.example, hermes_cli/auth.py, hermes_cli/config.py, hermes_cli/main.py, hermes_cli/model_normalize.py, hermes_cli/models.py, hermes_cli/providers.py, run_agent.py, tests/hermes_cli/test_ollama_cloud_provider.py, tests/run_agent/test_run_agent.py

  Show the diff

  • .env.example

    @@ -24,6 +24,15 @@
     # Optional base URL override (default: Google's OpenAI-compatible endpoint)
     # GEMINI_BASE_URL=https://generativelanguage.googleapis.com/v1beta/openai
     
    +# =============================================================================
    +# LLM PROVIDER (Ollama Cloud)
    +# =============================================================================
    +# Cloud-hosted open models via Ollama's OpenAI-compatible endpoint.
    +# Get your key at: https://ollama.com/settings
    +# OLLAMA_API_KEY=[redacted:secret-assignment]
    +# Optional base URL override (default: https://ollama.com/v1)
    +# OLLAMA_BASE_URL=https://ollama.com/v1
    +
     # =============================================================================
     # LLM PROVIDER (z.ai / GLM)
     # =============================================================================

  • agent/auxiliary_client.py

    @@ -104,6 +104,7 @@ def _normalize_aux_provider(provider: Optional[str]) -> str:
         "opencode-zen": "gemini-3-flash",
         "opencode-go": "glm-5",
         "kilocode": "google/gemini-3-flash-preview",
    +    "ollama-cloud": "nemotron-3-nano:30b",
     }
     
     # Vision-specific model overrides for direct providers.

  • agent/model_metadata.py

    @@ -23,7 +23,7 @@
     # are preserved so the full model name reaches cache lookups and server queries.
     _PROVIDER_PREFIXES: frozenset[str] = frozenset({
         "openrouter", "nous", "openai-codex", "copilot", "copilot-acp",
    -    "gemini", "zai", "kimi-coding", "kimi-coding-cn", "minimax", "minimax-cn", "anthropic", "deepseek",
    +    "gemini", "ollama-cloud", "zai", "kimi-coding", "kimi-coding-cn", "minimax", "minimax-cn", "anthropic", "deepseek",
         "opencode-zen", "opencode-go", "ai-gateway", "kilocode", "alibaba",
         "qwen-oauth",
         "xiaomi",
    @@ -33,6 +33,7 @@
         "google", "google-gemini", "google-ai-studio",
         "glm", "z-ai", "z.ai", "zhipu", "github", "github-copilot",
         "github-models", "kimi", "moonshot", "kimi-cn", "moonshot-cn", "claude", "deep-seek",
    +    "ollama",
         "opencode", "zen", "go", "vercel", "kilo", "dashscope", "aliyun", "qwen",
         "mimo", "xiaomi-mimo",
         "arcee-ai", "arceeai",
    @@ -239,6 +240,7 @@ def _is_custom_endpoint(base_url: str) -> bool:
         "api.x.ai": "xai",
         "api.xiaomimimo.com": "xiaomi",
         "xiaomimimo.com": "xiaomi",
    +    "ollama.com": "ollama-cloud",
     }
     
     

  • agent/models_dev.py

    @@ -169,6 +169,7 @@ class ProviderInfo:
         "togetherai": "togetherai",
         "perplexity": "perplexity",
         "cohere": "cohere",
    +    "ollama-cloud": "ollama-cloud",
     }
     
     # Reverse mapping: models.dev → Hermes (built lazily)

  • cli-config.yaml.example

    @@ -26,6 +26,7 @@ model:
       #   "huggingface"  - Hugging Face Inference (requires: HF_TOKEN)
       #   "xiaomi"       - Xiaomi MiMo (requires: XIAOMI_API_KEY)
       #   "arcee"        - Arcee AI Trinity models (requires: ARCEEAI_API_KEY)
    +  #   "ollama-cloud" - Ollama Cloud (requires: OLLAMA_API_KEY — https://ollama.com/settings)
       #   "kilocode"     - KiloCode gateway (requires: KILOCODE_API_KEY)
       #   "ai-gateway"   - Vercel AI Gateway (requires: AI_GATEWAY_API_KEY)
       #
    @@ -37,12 +38,6 @@ model:
       #     base_url: "http://localhost:1234/v1"
       #   No API key needed — local servers typically ignore auth.
       #
    -  #   For Ollama Cloud (https://ollama.com/pricing):
    -  #     provider: "custom"
    -  #     base_url: "https://ollama.com/v1"
    -  #   Set OLLAMA_API_KEY in .env — automatically picked up when base_url
    -  #   points to ollama.com.
    -  #
       # Can also be overridden with --provider flag or HERMES_INFERENCE_PROVIDER env var.
       provider: "auto"
       
    @@ -337,6 +332,7 @@ compression:
     #   "openrouter" - Force OpenRouter (requires OPENROUTER_API_KEY)
     #   "nous"       - Force Nous Portal (requires: hermes login)
     #   "gemini"      - Force Google AI Studio direct (requires: GOOGLE_API_KEY or GEMINI_API_KEY)
    +#   "ollama-cloud" - Ollama Cloud (requires: OLLAMA_API_KEY)
     #   "codex"       - Force Codex OAuth (requires: hermes model → Codex).
     #                  Uses gpt-5.3-codex which supports vision.
     #   "main"       - Use your custom endpoint (OPENAI_BASE_URL + OPENAI_API_KEY).

  • hermes_cli/auth.py

    @@ -70,6 +70,7 @@
     DEFAULT_QWEN_BASE_URL = "https://portal.qwen.ai/v1"
     DEFAULT_GITHUB_MODELS_BASE_URL = "https://api.githubcopilot.com"
     DEFAULT_COPILOT_ACP_BASE_URL = "acp://copilot"
    +DEFAULT_OLLAMA_CLOUD_BASE_URL = "https://ollama.com/v1"
     CODEX_OAUTH_CLIENT_ID = "app_EMoamEEZ73f0CkXaXp7hrann"
     CODEX_OAUTH_TOKEN_URL = "[redacted:secret-assignment]"
     CODEX_ACCESS_TOKEN_REFRESH_SKEW_SECONDS = 120
    @@ -274,6 +275,14 @@ class ProviderConfig:
             api_key_env_vars=("XIAOMI_API_KEY",),
             base_url_env_var="XIAOMI_BASE_URL",
         ),
    +    "ollama-cloud": ProviderConfig(
    +        id="ollama-cloud",
    +        name="Ollama Cloud",
    +        auth_type="api_key",
    +        inference_base_url=DEFAULT_OLLAMA_CLOUD_BASE_URL,
    +        api_key_env_vars=("OLLAMA_API_KEY",),
    +        base_url_env_var="OLLAMA_BASE_URL",
    +    ),
         "bedrock": ProviderConfig(
             id="bedrock",
             name="AWS Bedrock",
    @@ -937,7 +946,8 @@ def resolve_provider(
             "kilo": "kilocode", "kilo-code": "kilocode", "kilo-gateway": "kilocode",
             # Local server aliases — route through the generic custom provider
             "lmstudio": "custom", "lm-studio": "custom", "lm_studio": "custom",
    -        "ollama": "custom", "vllm": "custom", "llamacpp": "custom",
    +        "ollama": "custom", "ollama_cloud": "ollama-cloud",
    +        "vllm": "custom", "llamacpp": "custom",
             "llama.cpp": "custom", "llama-cpp": "custom",
         }
         normalized = _PROVIDER_ALIASES.get(normalized, normalized)

  • hermes_cli/config.py

    @@ -1024,6 +1024,22 @@ def _ensure_hermes_home_managed(home: Path):
             "category": "provider",
             "advanced": True,
         },
    +    "OLLAMA_API_KEY": {
    +        "description": "Ollama Cloud API key (ollama.com — cloud-hosted open models)",
    +        "prompt": "Ollama Cloud API key",
    +        "url": "https://ollama.com/settings",
    +        "password": True,
    +        "category": "provider",
    +        "advanced": True,
    +    },
    +    "OLLAMA_BASE_URL": {
    +        "description": "Ollama Cloud base URL override (default: https://ollama.com/v1)",
    +        "prompt": "Ollama base URL (leave empty for default)",
    +        "url": None,
    +        "password": False,
    +        "category": "provider",
    +        "advanced": True,
    +    },
         "XIAOMI_API_KEY": {
             "description": "Xiaomi MiMo API key for MiMo models (mimo-v2-pro, mimo-v2-omni, mimo-v2-flash)",
             "prompt": "Xiaomi MiMo API Key",

  • hermes_cli/main.py

    @@ -1141,7 +1141,7 @@ def _named_custom_provider_map(cfg) -> dict[str, dict[str, str]]:
             _model_flow_kimi(config, current_model)
         elif selected_provider == "bedrock":
             _model_flow_bedrock(config, current_model)
    -    elif selected_provider in ("gemini", "deepseek", "xai", "zai", "kimi-coding-cn", "minimax", "minimax-cn", "kilocode", "opencode-zen", "opencode-go", "ai-gateway", "alibaba", "huggingface", "xiaomi", "arcee"):
    +    elif selected_provider in ("gemini", "deepseek", "xai", "zai", "kimi-coding-cn", "minimax", "minimax-cn", "kilocode", "opencode-zen", "opencode-go", "ai-gateway", "alibaba", "huggingface", "xiaomi", "arcee", "ollama-cloud"):
             _model_flow_api_key_provider(config, selected_provider, current_model)
     
         # ── Post-switch cleanup: clear stale OPENAI_BASE_URL ──────────────
    @@ -1568,6 +1568,27 @@ def _model_flow_custom(config):
     
         effective_key = api_key or current_key
     
    +    # Hint: most local model servers (Ollama, vLLM, llama.cpp) require /v1
    +    # in the base URL for OpenAI-compatible chat completions.  Prompt the
    +    # user if the URL looks like a local server without /v1.
    +    _url_lower = effective_url.rstrip("/").lower()
    +    _looks_local = any(h in _url_lower for h in ("localhost", "127.0.0.1", "0.0.0.0", ":11434", ":8080", ":5000"))
    +    if _looks_local and not _url_lower.endswith("/v1"):
    +        print()
    +        print(f"  Hint: Did you mean to add /v1 at the end?")
    +        print(f"  Most local model servers (Ollama, vLLM, llama.cpp) require it.")
    +        print(f"  e.g. {effective_url.rstrip('/')}/v1")
    +        try:
    +            _add_v1 = input("  Add /v1? [Y/n]: ").strip().lower()
    +        except (KeyboardInterrupt, EOFError):
    +            _add_v1 = "n"
    +        if _add_v1 in ("", "y", "yes"):
    +            effective_url = effective_url.rstrip("/") + "/v1"
    +            if base_url:
    +                base_url = effective_url
    +            print(f"  Updated URL: {effective_url}")
    +        print()
    +
         from hermes_cli.models import probe_api_models
     
         probe = probe_api_models(effective_key, effective_url)
    @@ -2734,34 +2755,43 @@ def _model_flow_api_key_provider(config, provider_id, current_model=""):
         #   1. models.dev registry (cached, filtered for agentic/tool-capable models)
         #   2. Curated static fallback list (offline insurance)
         #   3. Live /models endpoint probe (small providers without models.dev data)
    -    curated = _PROVIDER_MODELS.get(provider_id, [])
    +    #
    +    # Ollama Cloud: dedicated merged discovery (live API + models.dev + disk cache)
    +    if provider_id == "ollama-cloud":
    +        from hermes_cli.models import fetch_ollama_cloud_models
    +        api_key_for_probe = [redacted:secret-assignment] or (get_env_value(key_env) if key_env else "")
    +        model_list = fetch_ollama_cloud_models(api_key=[redacted:secret-assignment] base_url=effective_base)
    +        if model_list:
    +            print(f"  Found {len(model_list)} model(s) from Ollama Cloud")
    +    else:
    +        curated = _PROVIDER_MODELS.get(provider_id, [])
     
    -    # Try models.dev first — returns tool-capable models, filtered for noise
    -    mdev_models: list = []
    -    try:
    -        from agent.models_dev import list_agentic_models
    -        mdev_models = list_agentic_models(provider_id)
    -    except Exception:
    -        pass
    +        # Try models.dev first — returns tool-capable models, filtered for noise
    +        mdev_models: list = []
    +        try:
    +            from agent.models_dev import list_agentic_models
    +            mdev_models = list_agentic_models(provider_id)
    +        except Exception:
    +            pass
     
    -    if mdev_models:
    -        model_list = mdev_models
    -        print(f"  Found {len(model_list)} model(s) from models.dev registry")
    -    elif curated and len(curated) >= 8:
    -        # Curated list is substantial — use it directly, skip live probe
    -        model_list = curated
    -        print(f"  Showing {len(model_list)} curated models — use \"Enter custom model name\" for others.")
    -    else:
    -        api_key_for_probe = [redacted:secret-assignment] or (get_env_value(key_env) if key_env else "")
    -        live_models = fetch_api_models(api_key_for_probe, effective_base)
    -        if live_models and len(live_models) >= len(curated):
    -            model_list = live_models
    -            print(f"  Found {len(model_list)} model(s) from {pconfig.name} API")
    -        else:
    +        if mdev_models:
    +            model_list = mdev_models
    +            print(f"  Found {len(model_list)} model(s) from models.dev registry")
    +        elif curated and len(curated) >= 8:
    +            # Curated list is substantial — use it directly, skip live probe
                 model_list = curated
    -            if model_list:
    -                print(f"  Showing {len(model_list)} curated models — use \"Enter custom model name\" for others.")
    -        # else: no defaults either, will fall through to raw input
    +            print(f"  Showing {len(model_list)} curated models — use \"Enter custom model name\" for others.")
    +        else:
    +            api_key_for_probe = [redacted:secret-assignment] or (get_env_value(key_env) if key_env else "")
    +            live_models = fetch_api_models(api_key_for_probe, effective_base)
    +            if live_models and len(live_models) >= len(curated):
    +                model_list = live_models
    +                print(f"  Found {len(model_list)} model(s) from {pconfig.name} API")
    +            else:
    +                model_list = curated
    +                if model_list:
    +                    print(f"  Showing {len(model_list)} curated models — use \"Enter custom model name\" for others.")
    +            # else: no defaults either, will fall through to raw input
     
         if provider_id in {"opencode-zen", "opencode-go"}:
             model_list = [normalize_opencode_model_id(provider_id, mid) for mid in model_list]
    @@ -4860,7 +4890,7 @@ def main():
         )
         chat_parser.add_argument(
             "--provider",
    -        choices=["auto", "openrouter", "nous", "openai-codex", "copilot-acp", "copilot", "anthropic", "gemini", "huggingface", "zai", "kimi-coding", "kimi-coding-cn", "minimax", "minimax-cn", "kilocode", "xiaomi", "arcee"],
    +        choices=["auto", "openrouter", "nous", "openai-codex", "copilot-acp", "copilot", "anthropic", "gemini", "ollama-cloud", "huggingface", "zai", "kimi-coding", "kimi-coding-cn", "minimax", "minimax-cn", "kilocode", "xiaomi", "arcee"],
             default=None,
             help="Inference provider (default: auto)"
         )

  • hermes_cli/model_normalize.py

    @@ -96,6 +96,7 @@
         "qwen-oauth",
         "xiaomi",
         "arcee",
    +    "ollama-cloud",
         "custom",
     })
     

  • hermes_cli/models.py

    @@ -11,7 +11,9 @@
     import os
     import urllib.request
     import urllib.error
    +import time
     from difflib import get_close_matches
    +from pathlib import Path
     from typing import Any, NamedTuple, Optional
     
     COPILOT_BASE_URL = "https://api.githubcopilot.com"
    @@ -547,6 +549,7 @@ class ProviderEntry(NamedTuple):
         ProviderEntry("minimax",        "MiniMax",                  "MiniMax (global direct API)"),
         ProviderEntry("minimax-cn",     "MiniMax (China)",          "MiniMax China (domestic direct API)"),
         ProviderEntry("alibaba",        "Alibaba Cloud (DashScope)","Alibaba Cloud / DashScope Coding (Qwen + multi-provider)"),
    +    ProviderEntry("ollama-cloud",   "Ollama Cloud",             "Ollama Cloud (cloud-hosted open models — ollama.com)"),
         ProviderEntry("arcee",          "Arcee AI",                 "Arcee AI (Trinity models — direct API)"),
         ProviderEntry("kilocode",       "Kilo Code",                "Kilo Code (Kilo Gateway API)"),
         ProviderEntry("opencode-zen",   "OpenCode Zen",             "OpenCode Zen (35+ curated models, pay-as-you-go)"),
    @@ -559,6 +562,7 @@ class ProviderEntry(NamedTuple):
     _PROVIDER_LABELS = {p.slug: p.label for p in CANONICAL_PROVIDERS}
     _PROVIDER_LABELS["custom"] = "Custom endpoint"  # special case: not a named provider
     
    +
     _PROVIDER_ALIASES = {
         "glm": "zai",
         "z-ai": "zai",
    @@ -611,6 +615,8 @@ class ProviderEntry(NamedTuple):
         "grok": "xai",
         "x-ai": "xai",
         "x.ai": "xai",
    +    "ollama": "custom",  # bare "ollama" = local; use "ollama-cloud" for cloud
    +    "ollama_cloud": "ollama-cloud",
     }
     
     
    @@ -1786,6 +1792,125 @@ def fetch_api_models(
         return probe_api_models(api_key, base_url, timeout=timeout).get("models")
     
     
    +# ---------------------------------------------------------------------------
    +# Ollama Cloud — merged model discovery with disk cache
    +# ---------------------------------------------------------------------------
    +
    +
    +
    +_OLLAMA_CLOUD_CACHE_TTL = 3600  # 1 hour
    +
    +
    +def _ollama_cloud_cache_path() -> Path:
    +    """Return the path for the Ollama Cloud model cache."""
    +    from hermes_constants import get_hermes_home
    +    return get_hermes_home() / "ollama_cloud_models_cache.json"
    +
    +
    +def _load_ollama_cloud_cache(*, ignore_ttl: bool = False) -> Optional[dict]:
    +    """Load cached Ollama Cloud models from disk.
    +
    +    Args:
    +        ignore_ttl: If True, return data even if the TTL has expired (stale fallback).
    +    """
    +    try:
    +        cache_path = _ollama_cloud_cache_path()
    +        if not cache_path.exists():
    +            return None
    +        with open(cache_path, encoding="utf-8") as f:
    +            data = json.load(f)
    +        if not isinstance(data, dict):
    +            return None
    +        models = data.get("models")
    +        if not (isinstance(models, list) and models):
    +            return None
    +        if not ignore_ttl:
    +            cached_at = data.get("cached_at", 0)
    +            if (time.time() - cached_at) > _OLLAMA_CLOUD_CACHE_TTL:
    +                return None  # stale
    +        return data
    +    except Exception:
    +        pass
    +    return None
    +
    +
    +def _save_ollama_cloud_cache(models: list[str]) -> None:
    +    """Persist the merged Ollama Cloud model list to disk."""
    +    try:
    +        from utils import atomic_json_write
    +        cache_path = _ollama_cloud_cache_path()
    +        cache_path.parent.mkdir(parents=True, exist_ok=True)
    +        atomic_json_write(cache_path, {"models": models, "cached_at": time.time()}, indent=None)
    +    except Exception:
    +        pass
    +
    +
    +def fetch_ollama_cloud_models(
    +    api_key: [redacted:secret-assignment] = None,
    +    base_url: Optional[str] = None,
    +    *,
    +    force_refresh: bool = False,
    +) -> list[str]:
    +    """Fetch Ollama Cloud models by merging live API + models.dev, with disk cache.
    +
    +    Resolution order:
    +      1. Disk cache (if fresh, < 1 hour, and not force_refresh)
    +      2. Live ``/v1/models`` endpoint (primary — freshest source)
    +      3. models.dev registry (secondary — fills gaps for unlisted models)
    +      4. Merge: live models first, then models.dev additions (deduped)
    +
    +    Returns a list of model IDs (never None — empty list on total failure).
    +    """
    +    # 1. Check disk cache
    +    if not force_refresh:
    +        cached = _load_ollama_cloud_cache()
    +        if cached is not None:
    +            return cached["models"]
    +
    +    # 2. Live API probe
    +    if not api_key:
    +        api_key = [redacted:secret-assignment]"OLLAMA_API_KEY", "")
    +    if not base_url:
    +        base_url = os.getenv("OLLAMA_BASE_URL", "") or "https://ollama.com/v1"
    +
    +    live_models: list[str] = []
    +    if api_key:
    +        result = fetch_api_models(api_key, base_url, timeout=8.0)
    +        if result:
    +            live_models = result
    +
    +    # 3. models.dev registry
    +    mdev_models: list[str] = []
    +    try:
    +        from agent.models_dev import list_agentic_models
    +        mdev_models = list_agentic_models("ollama-cloud")
    +    except Exception:
    +        pass
    +
    +    # 4. Merge: live first, then models.dev additions (deduped, order-preserving)
    +    if live_models or mdev_models:
    +        seen: set[str] = set()
    +        merged: list[str] = []
    +        for m in live_models:
    +            if m and m not in seen:
    +                seen.add(m)
    +                merged.append(m)
    +        for m in mdev_models:
    +            if m and m not in seen:
    +                seen.add(m)
    +                merged.append(m)
    +        if merged:
    +            _save_ollama_cloud_cache(merged)
    +            return merged
    +
    +    # Total failure — return stale cache if available (ignore TTL)
    +    stale = _load_ollama_cloud_cache(ignore_ttl=True)
    +    if stale is not None:
    +        return stale["models"]
    +
    +    return []
    +
    +
     def validate_requested_model(
         model_name: str,
         provider: Optional[str],

  • hermes_cli/providers.py

    @@ -141,6 +141,10 @@ class HermesOverlay:
             base_url_override="https://api.arcee.ai/api/v1",
             base_url_env_var="ARCEE_BASE_URL",
         ),
    +    "ollama-cloud": HermesOverlay(
    +        transport="openai_chat",
    +        base_url_env_var="OLLAMA_BASE_URL",
    +    ),
     }
     
     
    @@ -250,7 +254,7 @@ class ProviderDef:
         "lmstudio": "lmstudio",
         "lm-studio": "lmstudio",
         "lm_studio": "lmstudio",
    -    "ollama": "ollama-cloud",
    +    "ollama": "custom",  # bare "ollama" = local; use "ollama-cloud" for cloud
         "vllm": "local",
         "llamacpp": "local",
         "llama.cpp": "local",
    @@ -269,6 +273,7 @@ class ProviderDef:
         "xiaomi": "Xiaomi MiMo",
         "local": "Local endpoint",
         "bedrock": "AWS Bedrock",
    +    "ollama-cloud": "Ollama Cloud",
     }
     
     

  • run_agent.py

    @@ -2103,6 +2103,59 @@ def _strip_think_blocks(self, content: str) -> str:
             content = re.sub(r'</?(?:think|thinking|reasoning|thought|REASONING_SCRATCHPAD)>\s*', '', content, flags=re.IGNORECASE)
             return content
     
    +    @staticmethod
    +    def _has_natural_response_ending(content: str) -> bool:
    +        """Heuristic: does visible assistant text look intentionally finished?"""
    +        if not content:
    +            return False
    +        stripped = content.rstrip()
    +        if not stripped:
    +            return False
    +        if stripped.endswith("```"):
    +            return True
    +        return stripped[-1] in '.!?:)"\']}。！？：）】」』》'
    +
    +    def _is_ollama_glm_backend(self) -> bool:
    +        """Detect the narrow backend family affected by Ollama/GLM stop misreports."""
    +        model_lower = (self.model or "").lower()
    +        provider_lower = (self.provider or "").lower()
    +        if "glm" not in model_lower and provider_lower != "zai":
    +            return False
    +        if "ollama" in self._base_url_lower or ":11434" in self._base_url_lower:
    +            return True
    +        return bool(self.base_url and is_local_endpoint(self.base_url))
    +
    +    def _should_treat_stop_as_truncated(
    +        self,
    +        finish_reason: str,
    +        assistant_message,
    +        messages: Optional[list] = None,
    +    ) -> bool:
    +        """Detect conservative stop->length misreports for Ollama-hosted GLM models."""
    +        if finish_reason != "stop" or self.api_mode != "chat_completions":
    +            return False
    +        if not self._is_ollama_glm_backend():
    +            return False
    +        if not any(
    +            isinstance(msg, dict) and msg.get("role") == "tool"
    +            for msg in (messages or [])
    +        ):
    +            return False
    +        if assistant_message is None or getattr(assistant_message, "tool_calls", None):
    +            return False
    +
    +        content = getattr(assistant_message, "content", None)
    +        if not isinstance(content, str):
    +            return False
    +
    +        visible_text = self._strip_think_blocks(content).strip()
    +        if not visible_text:
    +            return False
    +        if len(visible_text) < 20 or not re.search(r"\s", visible_text):
    +            return False
    +
    +        return not self._has_natural_response_ending(visible_text)
    +
         def _looks_like_codex_intermediate_ack(
             self,
             user_message: str,
    @@ -6635,6 +6688,18 @@ def _build_api_kwargs(self, api_messages: list) -> dict:
                 options["num_ctx"] = self._ollama_num_ctx
                 extra_body["options"] = options
     
    +        # Ollama / custom provider: pass think=false when reasoning is disabled.
    +        # Ollama does not recognise the OpenRouter-style `reasoning` extra_body
    +        # field, so we use its native `think` parameter instead.
    +        # This prevents thinking-capable models (Qwen3, etc.) from generating
    +        # <think> blocks and producing empty-response errors when the user has
    +        # set reasoning_effort: none.
    +        if self.provider == "custom" and self.reasoning_config and isinstance(self.reasoning_config, dict):
    +            _effort = (self.reasoning_config.get("effort") or "").strip().lower()
    +            _enabled = self.reasoning_config.get("enabled", True)
    +            if _effort == "none" or _enabled is False:
    +                extra_body["think"] = False
    +
             if self._is_qwen_portal():
                 extra_body["vl_high_resolution_images"] = True
     
    @@ -6757,9 +6822,16 @@ def _build_assistant_message(self, assistant_message, finish_reason: str) -> dic
                     except Exception:
                         pass
     
    +        # Sanitize surrogates from API response — some models (e.g. Kimi/GLM via Ollama)
    +        # can return invalid surrogate code points that crash json.dumps() on persist.
    +        _raw_content = assistant_message.content or ""
    +        _san_content = _sanitize_surrogates(_raw_content)
    +        if reasoning_text:
    +            reasoning_text = _sanitize_surrogates(reasoning_text)
    +
             msg = {
                 "role": "assistant",
    -            "content": assistant_message.content or "",
    +            "content": _san_content,
                 "reasoning": reasoning_text,
                 "finish_reason": finish_reason,
             }
    @@ -8640,6 +8712,12 @@ def run_conversation(
                         new_tcs.append(tc)
                     am["tool_calls"] = new_tcs
     
    +            # Proactively strip any surrogate characters before the API call.
    +            # Models served via Ollama (Kimi K2.5, GLM-5, Qwen) can return
    +            # lone surrogates (U+D800-U+DFFF) that crash json.dumps() inside
    +            # the OpenAI SDK. Sanitizing here prevents the 3-retry cycle.
    +            _sanitize_messages_surrogates(api_messages)
    +
                 # Calculate approximate request size for logging
                 total_chars = sum(len(str(msg)) for msg in api_messages)
                 approx_tokens = [redacted:secret-assignment]
    @@ -9038,6 +9116,17 @@ def _stop_spinner():
                             finish_reason = stop_reason_map.get(response.stop_reason, "stop")
                         else:
                             finish_reason = response.choices[0].finish_reason
    +                        assistant_message = response.choices[0].message
    +                        if self._should_treat_stop_as_truncated(
    +                            finish_reason,
    +                            assistant_message,
    +                            messages,
    +                        ):
    +                            self._vprint(
    +                                f"{self.log_prefix}⚠️  Treating suspicious Ollama/GLM stop response as truncated",
    +                                force=True,
    +                            )
    +                            finish_reason = "length"
     
                         if finish_reason == "length":
                             self._vprint(f"{self.log_prefix}⚠️  Response truncated (finish_reason='length') - model hit max output tokens", force=True)

  • tests/hermes_cli/test_ollama_cloud_provider.py

    @@ -0,0 +1,351 @@
    +"""Tests for Ollama Cloud provider integration."""
    +
    +import os
    +import pytest
    +from unittest.mock import patch, MagicMock
    +
    +from hermes_cli.auth import PROVIDER_REGISTRY, resolve_provider, resolve_api_key_provider_credentials
    +from hermes_cli.models import _PROVIDER_MODELS, _PROVIDER_LABELS, _PROVIDER_ALIASES, normalize_provider
    +from hermes_cli.model_normalize import normalize_model_for_provider
    +from agent.model_metadata import _URL_TO_PROVIDER, _PROVIDER_PREFIXES
    +from agent.models_dev import PROVIDER_TO_MODELS_DEV, list_agentic_models
    +
    +
    +# ── Provider Registry ──
    +
    +class TestOllamaCloudProviderRegistry:
    +    def test_ollama_cloud_in_registry(self):
    +        assert "ollama-cloud" in PROVIDER_REGISTRY
    +
    +    def test_ollama_cloud_config(self):
    +        pconfig = PROVIDER_REGISTRY["ollama-cloud"]
    +        assert pconfig.id == "ollama-cloud"
    +        assert pconfig.name == "Ollama Cloud"
    +        assert pconfig.auth_type == "api_key"
    +        assert pconfig.inference_base_url == "https://ollama.com/v1"
    +
    +    def test_ollama_cloud_env_vars(self):
    +        pconfig = PROVIDER_REGISTRY["ollama-cloud"]
    +        assert pconfig.api_key_env_vars == ("OLLAMA_API_KEY",)
    +        assert pconfig.base_url_env_var == "OLLAMA_BASE_URL"
    +
    +    def test_ollama_cloud_base_url(self):
    +        assert "ollama.com" in PROVIDER_REGISTRY["ollama-cloud"].inference_base_url
    +
    +
    +# ── Provider Aliases ──
    +
    +PROVIDER_ENV_VARS = (
    +    "OPENROUTER_API_KEY", "OPENAI_API_KEY", "ANTHROPIC_API_KEY",
    +    "GOOGLE_API_KEY", "GEMINI_API_KEY", "OLLAMA_API_KEY",
    +    "GLM_API_KEY", "ZAI_API_KEY", "KIMI_API_KEY",
    +    "MINIMAX_API_KEY", "DEEPSEEK_API_KEY",
    +)
    +
    [redacted:email](autouse=True)
    +def _clean_provider_env(monkeypatch):
    +    for var in PROVIDER_ENV_VARS:
    +        monkeypatch.delenv(var, raising=False)
    +
    +
    +class TestOllamaCloudAliases:
    +    def test_explicit_ollama_cloud(self):
    +        assert resolve_provider("ollama-cloud") == "ollama-cloud"
    +
    +    def test_alias_ollama_underscore(self):
    +        """ollama_cloud (underscore) is the unambiguous cloud alias."""
    +        assert resolve_provider("ollama_cloud") == "ollama-cloud"
    +
    +    def test_bare_ollama_stays_local(self):
    +        """Bare 'ollama' alias routes to 'custom' (local) — not cloud."""
    +        assert resolve_provider("ollama") == "custom"
    +
    +    def test_models_py_aliases(self):
    +        assert _PROVIDER_ALIASES.get("ollama_cloud") == "ollama-cloud"
    +        # bare "ollama" stays local
    +        assert _PROVIDER_ALIASES.get("ollama") == "custom"
    +
    +    def test_normalize_provider(self):
    +        assert normalize_provider("ollama-cloud") == "ollama-cloud"
    +
    +
    +# ── Auto-detection ──
    +
    +class TestOllamaCloudAutoDetection:
    +    def test_auto_detects_ollama_api_key(self, monkeypatch):
    +        monkeypatch.setenv("OLLAMA_API_KEY", "test-ollama-key")
    +        assert resolve_provider("auto") == "ollama-cloud"
    +
    +
    +# ── Credential Resolution ──
    +
    +class TestOllamaCloudCredentials:
    +    def test_resolve_with_ollama_api_key(self, monkeypatch):
    +        monkeypatch.setenv("OLLAMA_API_KEY", "ollama-secret")
    +        creds = resolve_api_key_provider_credentials("ollama-cloud")
    +        assert creds["provider"] == "ollama-cloud"
    +        assert creds["api_key"] == "ollama-secret"
    +        assert creds["base_url"] == "https://ollama.com/v1"
    +
    +    def test_resolve_with_custom_base_url(self, monkeypatch):
    +        monkeypatch.setenv("OLLAMA_API_KEY", "key")
    +        monkeypatch.setenv("OLLAMA_BASE_URL", "https://custom.ollama/v1")
    +        creds = resolve_api_key_provider_credentials("ollama-cloud")
    +        assert creds["base_url"] == "https://custom.ollama/v1"
    +
    +    def test_runtime_ollama_cloud(self, monkeypatch):
    +        monkeypatch.setenv("OLLAMA_API_KEY", "ollama-key")
    +        from hermes_cli.runtime_provider import resolve_runtime_provider
    +        result = resolve_runtime_provider(requested="ollama-cloud")
    +        assert result["provider"] == "ollama-cloud"
    +        assert result["api_mode"] == "chat_completions"
    +        assert result["api_key"] == "ollama-key"
    +        assert result["base_url"] == "https://ollama.com/v1"
    +
    +
    +# ── Model Catalog (dynamic — no static list) ──
    +
    +class TestOllamaCloudModelCatalog:
    +    def test_no_static_model_list(self):
    +        """Ollama Cloud models are fetched dynamically — no static list to maintain."""
    +        assert "ollama-cloud" not in _PROVIDER_MODELS
    +
    +    def test_provider_label(self):
    +        assert "ollama-cloud" in _PROVIDER_LABELS
    +        assert _PROVIDER_LABELS["ollama-cloud"] == "Ollama Cloud"
    +
    +
    +# ── Merged Model Discovery ──
    +
    +class TestOllamaCloudMergedDiscovery:
    +    def test_merges_live_and_models_dev(self, tmp_path, monkeypatch):
    +        """Live API models appear first, models.dev additions fill gaps."""
    +        from hermes_cli.models import fetch_ollama_cloud_models
    +
    +        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    +        monkeypatch.setenv("OLLAMA_API_KEY", "test-key")
    +
    +        mock_mdev = {
    +            "ollama-cloud": {
    +                "models": {
    +                    "glm-5": {"tool_call": True},
    +                    "kimi-k2.5": {"tool_call": True},
    +                    "nemotron-3-super": {"tool_call": True},
    +                }
    +            }
    +        }
    +        with patch("hermes_cli.models.fetch_api_models", return_value=["qwen3.5:397b", "glm-5"]), \
    +             patch("agent.models_dev.fetch_models_dev", return_value=mock_mdev):
    +            result = fetch_ollama_cloud_models(force_refresh=True)
    +
    +        # Live models first, then models.dev additions (deduped)
    +        assert result[0] == "qwen3.5:397b"  # from live API
    +        assert result[1] == "glm-5"          # from live API (also in models.dev)
    +        assert "kimi-k2.5" in result         # from models.dev only
    +        assert "nemotron-3-super" in result  # from models.dev only
    +        assert result.count("glm-5") == 1    # no duplicates
    +
    +    def test_falls_back_to_models_dev_without_api_key(self, tmp_path, monkeypatch):
    +        """Without API key, only models.dev results are returned."""
    +        from hermes_cli.models import fetch_ollama_cloud_models
    +
    +        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    +        monkeypatch.delenv("OLLAMA_API_KEY", raising=False)
    +
    +        mock_mdev = {
    +            "ollama-cloud": {
    +                "models": {
    +                    "glm-5": {"tool_call": True},
    +                }
    +            }
    +        }
    +        with patch("agent.models_dev.fetch_models_dev", return_value=mock_mdev):
    +            result = fetch_ollama_cloud_models(force_refresh=True)
    +
    +        assert result == ["glm-5"]
    +
    +    def test_uses_disk_cache(self, tmp_path, monkeypatch):
    +        """Second call returns cached results without hitting APIs."""
    +        from hermes_cli.models import fetch_ollama_cloud_models
    +
    +        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    +        monkeypatch.setenv("OLLAMA_API_KEY", "test-key")
    +
    +        with patch("hermes_cli.models.fetch_api_models", return_value=["model-a"]) as mock_api, \
    +             patch("agent.models_dev.fetch_models_dev", return_value={}):
    +            first = fetch_ollama_cloud_models(force_refresh=True)
    +            assert first == ["model-a"]
    +            assert mock_api.call_count == 1
    +
    +            # Second call — should use disk cache, not call API
    +            second = fetch_ollama_cloud_models()
    +            assert second == ["model-a"]
    +            assert mock_api.call_count == 1  # no extra API call
    +
    +    def test_force_refresh_bypasses_cache(self, tmp_path, monkeypatch):
    +        """force_refresh=True always hits the API even with fresh cache."""
    +        from hermes_cli.models import fetch_ollama_cloud_models
    +
    +        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    +        monkeypatch.setenv("OLLAMA_API_KEY", "test-key")
    +
    +        with patch("hermes_cli.models.fetch_api_models", return_value=["model-a"]) as mock_api, \
    +             patch("agent.models_dev.fetch_models_dev", return_value={}):
    +            fetch_ollama_cloud_models(force_refresh=True)
    +            fetch_ollama_cloud_models(force_refresh=True)
    +            assert mock_api.call_count == 2
    +
    +    def test_stale_cache_used_on_total_failure(self, tmp_path, monkeypatch):
    +        """If both API and models.dev fail, stale cache is returned."""
    +        from hermes_cli.models import fetch_ollama_cloud_models, _save_ollama_cloud_cache
    +
    +        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    +        monkeypatch.setenv("OLLAMA_API_KEY", "test-key")
    +
    +        # Pre-populate a stale cache
    +        _save_ollama_cloud_cache(["stale-model"])
    +
    +        # Make the cache appear stale by backdating it
    +        import json
    +        cache_path = tmp_path / "ollama_cloud_models_cache.json"
    +        with open(cache_path) as f:
    +            data = json.load(f)
    +        data["cached_at"] = 0  # epoch = very stale
    +        with open(cache_path, "w") as f:
    +            json.dump(data, f)
    +
    +        with patch("hermes_cli.models.fetch_api_models", return_value=None), \
    +             patch("agent.models_dev.fetch_models_dev", return_value={}):
    +            result = fetch_ollama_cloud_models(force_refresh=True)
    +
    +        assert result == ["stale-model"]
    +
    +    def test_empty_on_total_failure_no_cache(self, tmp_path, monkeypatch):
    +        """Returns empty list when everything fails and no cache exists."""
    +        from hermes_cli.models import fetch_ollama_cloud_models
    +
    +        monkeypatch.setenv("HERMES_HOME", str(tmp_path))
    +        monkeypatch.delenv("OLLAMA_API_KEY", raising=False)
    +
    +        with patch("agent.models_dev.fetch_models_dev", return_value={}):
    +            result = fetch_ollama_cloud_models(force_refresh=True)
    +
    +        assert result == []
    +
    +
    +# ── Model Normalization ──
    +
    +class TestOllamaCloudModelNormalization:
    +    def test_passthrough_bare_name(self):
    +        """Ollama Cloud is a passthrough provider — model names used as-is."""
    +        assert normalize_model_for_provider("qwen3.5:397b", "ollama-cloud") == "qwen3.5:397b"
    +
    +    def test_passthrough_with_tag(self):
    +        assert normalize_model_for_provider("cogito-2.1:671b", "ollama-cloud") == "cogito-2.1:671b"
    +
    +    def test_passthrough_no_tag(self):
    +        assert normalize_model_for_provider("glm-5", "ollama-cloud") == "glm-5"
    +
    +
    +# ── URL-to-Provider Mapping ──
    +
    +class TestOllamaCloudUrlMapping:
    +    def test_url_to_provider(self):
    +        assert _URL_TO_PROVIDER.get("ollama.com") == "ollama-cloud"
    +
    +    def test_provider_prefix_canonical(self):
    +        assert "ollama-cloud" in _PROVIDER_PREFIXES
    +
    +    def test_provider_prefix_alias(self):
    +        assert "ollama" in _PROVIDER_PREFIXES
    +
    +
    +# ── models.dev Integration ──
    +
    +class TestOllamaCloudModelsDev:
    +    def test_ollama_cloud_mapped(self):
    +        assert PROVIDER_TO_MODELS_DEV.get("ollama-cloud") == "ollama-cloud"
    +
    +    def test_list_agentic_models_with_mock_data(self):
    +        """list_agentic_models filters correctly from mock models.dev data."""
    +        mock_data = {
    +            "ollama-cloud": {
    +                "models": {
    +                    "qwen3.5:397b": {"tool_call": True},
    +                    "glm-5": {"tool_call": True},
    +                    "nemotron-3-nano:30b": {"tool_call": True},
    +                    "some-embedding:latest": {"tool_call": False},
    +                }
    +            }
    +        }
    +        with patch("agent.models_dev.fetch_models_dev", return_value=mock_data):
    +            result = list_agentic_models("ollama-cloud")
    +        assert "qwen3.5:397b" in result
    +        assert "glm-5" in result
    +        assert "nemotron-3-nano:30b" in result
    +        assert "some-embedding:latest" not in result  # no tool_call
    +
    +
    +# ── Agent Init (no SyntaxError) ──
    +
    +class TestOllamaCloudAgentInit:
    +    def test_agent_imports_without_error(self):
    +        """Verify run_agent.py has no SyntaxError."""
    +        import importlib
    +        import run_agent
    +        importlib.reload(run_agent)
    +
    +    def test_ollama_cloud_agent_uses_chat_completions(self, monkeypatch):
    +        """Ollama Cloud falls through to chat_completions — no special elif needed."""
    +        monkeypatch.setenv("OLLAMA_API_KEY", "test-key")
    +        with patch("run_agent.OpenAI") as mock_openai:
    +            mock_openai.return_value = MagicMock()
    +            from run_agent import AIAgent
    +            agent = AIAgent(
    +                model="qwen3.5:397b",
    +                provider="ollama-cloud",
    +                api_key="[redacted:secret-assignment]",
    +                base_url="https://ollama.com/v1",
    +            )
    +            assert agent.api_mode == "chat_completions"
    +            assert agent.provider == "ollama-cloud"
    +
    +
    +# ── providers.py New System ──
    +
    +class TestOllamaCloudProvidersNew:
    +    def test_overlay_exists(self):
    +        from hermes_cli.providers import HERMES_OVERLAYS
    +        assert "ollama-cloud" in HERMES_OVERLAYS
    +        overlay = HERMES_OVERLAYS["ollama-cloud"]
    +        assert overlay.transport == "openai_chat"
    +        assert overlay.base_url_env_var == "OLLAMA_BASE_URL"
    +
    +    def test_alias_resolves(self):
    +        from hermes_cli.providers import normalize_provider as np
    +        assert np("ollama") == "custom"  # bare "ollama" = local
    +        assert np("ollama-cloud") == "ollama-cloud"
    +
    +    def test_label_override(self):
    +        from hermes_cli.providers import _LABEL_OVERRIDES
    +        assert _LABEL_OVERRIDES.get("ollama-cloud") == "Ollama Cloud"
    +
    +    def test_get_label(self):
    +        from hermes_cli.providers import get_label
    +        assert get_label("ollama-cloud") == "Ollama Cloud"
    +
    +    def test_get_provider(self):
    +        from hermes_cli.providers import get_provider
    +        pdef = get_provider("ollama-cloud")
    +        assert pdef is not None
    +        assert pdef.id == "ollama-cloud"
    +        assert pdef.transport == "openai_chat"
    +
    +
    +# ── Auxiliary Model ──
    +
    +class TestOllamaCloudAuxiliary:
    +    def test_aux_model_defined(self):
    +        from agent.auxiliary_client import _API_KEY_PROVIDER_AUX_MODELS
    +        assert "ollama-cloud" in _API_KEY_PROVIDER_AUX_MODELS
    +        assert _API_KEY_PROVIDER_AUX_MODELS["ollama-cloud"] == "nemotron-3-nano:30b"

  • tests/run_agent/test_run_agent.py

    @@ -928,6 +928,7 @@ def test_max_tokens_injected(self, agent):
             kwargs = agent._build_api_kwargs(messages)
             assert kwargs["max_tokens"] == 4096
     
    +
         def test_qwen_portal_formats_messages_and_metadata(self, agent):
             agent.base_url = "https://portal.qwen.ai/v1"
             agent._base_url_lower = agent.base_url.lower()
    @@ -984,6 +985,46 @@ def test_qwen_portal_default_max_tokens(self, agent):
             kwargs = agent._build_api_kwargs(messages)
             assert kwargs["max_tokens"] == 65536
     
    +    def test_ollama_think_false_on_effort_none(self, agent):
    +        """Custom (Ollama) provider with effort=none should inject think=false."""
    +        agent.provider = "custom"
    +        agent.base_url = "http://localhost:11434/v1"
    +        agent._base_url_lower = agent.base_url.lower()
    +        agent.reasoning_config = {"effort": "none"}
    +        messages = [{"role": "user", "content": "hi"}]
    +        kwargs = agent._build_api_kwargs(messages)
    +        assert kwargs.get("extra_body", {}).get("think") is False
    +
    +    def test_ollama_think_false_on_enabled_false(self, agent):
    +        """Custom (Ollama) provider with enabled=false should inject think=false."""
    +        agent.provider = "custom"
    +        agent.base_url = "http://localhost:11434/v1"
    +        agent._base_url_lower = agent.base_url.lower()
    +        agent.reasoning_config = {"enabled": False}
    +        messages = [{"role": "user", "content": "hi"}]
    +        kwargs = agent._build_api_kwargs(messages)
    +        assert kwargs.get("extra_body", {}).get("think") is False
    +
    +    def test_ollama_no_think_param_when_reasoning_enabled(self, agent):
    +        """Custom provider with reasoning enabled should NOT inject think=false."""
    +        agent.provider = "custom"
    +        agent.base_url = "http://localhost:11434/v1"
    +        agent._base_url_lower = agent.base_url.lower()
    +        agent.reasoning_config = {"enabled": True, "effort": "medium"}
    +        messages = [{"role": "user", "content": "hi"}]
    +        kwargs = agent._build_api_kwargs(messages)
    +        assert kwargs.get("extra_body", {}).get("think") is None
    +
    +    def test_non_custom_provider_unaffected(self, agent):
    +        """OpenRouter provider with effort=none should NOT inject think=false."""
    +        agent.provider = "openrouter"
    +        agent.model = "qwen/qwen3.5-plus-02-15"
    +        agent.reasoning_config = {"effort": "none"}
    +        messages = [{"role": "user", "content": "hi"}]
    +        kwargs = agent._build_api_kwargs(messages)
    +        assert kwargs.get("extra_body", {}).get("think") is None
    +
    +
     
     class TestBuildAssistantMessage:
         def test_basic_message(self, agent):
    @@ -2202,6 +2243,114 @@ def test_length_finish_reason_requests_continuation(self, agent):
             assert second_call_messages[-1]["role"] == "user"
             assert "truncated by the output length limit" in second_call_messages[-1]["content"]
     
    +    def test_ollama_glm_stop_after_tools_without_terminal_boundary_requests_continuation(self, agent):
    +        """Ollama-hosted GLM responses can misreport truncated output as stop."""
    +        self._setup_agent(agent)
    +        agent.base_url = "http://localhost:11434/v1"
    +        agent._base_url_lower = agent.base_url.lower()
    +        agent.model = "glm-5.1:cloud"
    +
    +        tool_turn = _mock_response(
    +            content="",
    +            finish_reason="tool_calls",
    +            tool_calls=[_mock_tool_call(name="web_search", arguments="{}", call_id="c1")],
    +        )
    +        misreported_stop = _mock_response(
    +            content="Based on the search results, the best next",
    +            finish_reason="stop",
    +        )
    +        continued = _mock_response(
    +            content=" step is to update the config.",
    +            finish_reason="stop",
    +        )
    +        agent.client.chat.completions.create.side_effect = [
    +            tool_turn,
    +            misreported_stop,
    +            continued,
    +        ]
    +
    +        with (
    +            patch("run_agent.handle_function_call", return_value="search result"),
    +            patch.object(agent, "_persist_session"),
    +            patch.object(agent, "_save_trajectory"),
    +            patch.object(agent, "_cleanup_task_resources"),
    +        ):
    +            result = agent.run_conversation("hello")
    +
    +        assert result["completed"] is True
    +        assert result["api_calls"] == 3
    +        assert (
    +            result["final_response"]
    +            == "Based on the search results, the best next step is to update the config."
    +        )
    +
    +        third_call_messages = agent.client.chat.completions.create.call_args_list[2].kwargs["messages"]
    +        assert third_call_messages[-1]["role"] == "user"
    +        assert "truncated by the output length limit" in third_call_messages[-1]["content"]
    +
    +    def test_ollama_glm_stop_with_terminal_boundary_does_not_continue(self, agent):
    +        """Complete Ollama/GLM responses should not be reclassified as truncated."""
    +        self._setup_agent(agent)
    +        agent.base_url = "http://localhost:11434/v1"
    +        agent._base_url_lower = agent.base_url.lower()
    +        agent.model = "glm-5.1:cloud"
    +
    +        tool_turn = _mock_response(
    +            content="",
    +            finish_reason="tool_calls",
    +            tool_calls=[_mock_tool_call(name="web_search", arguments="{}", call_id="c1")],
    +        )
    +        complete_stop = _mock_response(
    +            content="Based on the search results, the best next step is to update the config.",
    +            finish_reason="stop",
    +        )
    +        agent.client.chat.completions.create.side_effect = [tool_turn, complete_stop]
    +
    +        with (
    +            patch("run_agent.handle_function_call", return_value="search result"),
    +            patch.object(agent, "_persist_session"),
    +            patch.object(agent, "_save_trajectory"),
    +            patch.object(agent, "_cleanup_task_resources"),
    +        ):
    +            result = agent.run_conversation("hello")
    +
    +        assert result["completed"] is True
    +        assert result["api_calls"] == 2
    +        assert (
    +            result["final_response"]
    +            == "Based on the search results, the best next step is to update the config."
    +        )
    +
    +    def test_non_ollama_stop_without_terminal_boundary_does_not_continue(self, agent):
    +        """The stop->length workaround should stay scoped to Ollama/GLM backends."""
    +        self._setup_agent(agent)
    +        agent.base_url = "https://api.openai.com/v1"
    +        agent._base_url_lower = agent.base_url.lower()
    +        agent.model = "gpt-4o-mini"
    +
    +        tool_turn = _mock_response(
    +            content="",
    +            finish_reason="tool_calls",
    +            tool_calls=[_mock_tool_call(name="web_search", arguments="{}", call_id="c1")],
    +        )
    +        normal_stop = _mock_response(
    +            content="Based on the search results, the best next",
    +            finish_reason="stop",
    +        )
    +        agent.client.chat.completions.create.side_effect = [tool_turn, normal_stop]
    +
    +        with (
    +            patch("run_agent.handle_function_call", return_value="search result"),
    +            patch.object(agent, "_persist_session"),
    +            patch.object(agent, "_save_trajectory"),
    +            patch.object(agent, "_cleanup_task_resources"),
    +        ):
    +            result = agent.run_conversation("hello")
    +
    +        assert result["completed"] is True
    +        assert result["api_calls"] == 2
    +        assert result["final_response"] == "Based on the search results, the best next"
    +
         def test_length_thinking_exhausted_skips_continuation(self, agent):
             """When finish_reason='length' but content is only thinking, skip retries."""
             self._setup_agent(agent)