fix: preserve Ollama model:tag colons in context length detection (#2149)

The colon-split logic in get_model_context_length() and
_query_local_context_length() assumed any colon meant provider:model
format (e.g. "local:my-model"). But Ollama uses model:tag format
(e.g. "qwen3.5:27b"), so the split turned "qwen3.5:27b" into just
"27b" — which matches nothing, causing a fallback to the 2M token
probe tier.

Now only recognised provider prefixes (local, openrouter, anthropic,
etc.) are stripped. Ollama model:tag names pass through intact.

Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>

agent/model_metadata.py

@@ -19,6 +19,34 @@ from hermes_constants import OPENROUTER_MODELS_URL
 
 logger = logging.getLogger(__name__)
 
+# Provider names that can appear as a "provider:" prefix before a model ID.
+# Only these are stripped — Ollama-style "model:tag" colons (e.g. "qwen3.5:27b")
+# 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",
+    "zai", "kimi-coding", "minimax", "minimax-cn", "anthropic", "deepseek",
+    "opencode-zen", "opencode-go", "ai-gateway", "kilocode", "alibaba",
+    "custom", "local",
+    # Common aliases
+    "glm", "z-ai", "z.ai", "zhipu", "github", "github-copilot",
+    "github-models", "kimi", "moonshot", "claude", "deep-seek",
+    "opencode", "zen", "go", "vercel", "kilo", "dashscope", "aliyun", "qwen",
+})
+
+
+def _strip_provider_prefix(model: str) -> str:
+    """Strip a recognised provider prefix from a model string.
+
+    ``"local:my-model"`` → ``"my-model"``
+    ``"qwen3.5:27b"``   → ``"qwen3.5:27b"``  (unchanged — not a provider prefix)
+    """
+    if ":" not in model or model.startswith("http"):
+        return model
+    prefix = model.split(":", 1)[0].strip().lower()
+    if prefix in _PROVIDER_PREFIXES:
+        return model.split(":", 1)[1]
+    return model
+
 _model_metadata_cache: Dict[str, Dict[str, Any]] = {}
 _model_metadata_cache_time: float = 0
 _MODEL_CACHE_TTL = 3600
@@ -579,10 +607,9 @@ def _query_local_context_length(model: str, base_url: str) -> Optional[int]:
     """Query a local server for the model's context length."""
     import httpx
 
-    # Strip provider prefix (e.g., "local:model-name" → "model-name").
+    # Strip recognised provider prefix (e.g., "local:model-name" → "model-name").
-    # LM Studio and Ollama don't use provider prefixes in their model IDs.
+    # Ollama "model:tag" colons (e.g. "qwen3.5:27b") are intentionally preserved.
-    if ":" in model and not model.startswith("http"):
+    model = _strip_provider_prefix(model)
-        model = model.split(":", 1)[1]
 
     # Strip /v1 suffix to get the server root
     server_url = base_url.rstrip("/")
@@ -689,9 +716,8 @@ def get_model_context_length(
 
     # Normalise provider-prefixed model names (e.g. "local:model-name" →
     # "model-name") so cache lookups and server queries use the bare ID that
-    # local servers actually know about.
+    # local servers actually know about.  Ollama "model:tag" colons are preserved.
-    if ":" in model and not model.startswith("http"):
+    model = _strip_provider_prefix(model)
-        model = model.split(":", 1)[1]
 
     # 1. Check persistent cache (model+provider)
     if base_url:

tests/agent/test_model_metadata.py

@@ -22,6 +22,7 @@ from unittest.mock import patch, MagicMock
 from agent.model_metadata import (
     CONTEXT_PROBE_TIERS,
     DEFAULT_CONTEXT_LENGTHS,
+    _strip_provider_prefix,
     estimate_tokens_rough,
     estimate_messages_tokens_rough,
     get_model_context_length,
@@ -292,6 +293,49 @@ class TestGetModelContextLength:
         assert result == 200000
 
 
+# =========================================================================
+# _strip_provider_prefix — Ollama model:tag vs provider:model
+# =========================================================================
+
+class TestStripProviderPrefix:
+    def test_known_provider_prefix_is_stripped(self):
+        assert _strip_provider_prefix("local:my-model") == "my-model"
+        assert _strip_provider_prefix("openrouter:anthropic/claude-sonnet-4") == "anthropic/claude-sonnet-4"
+        assert _strip_provider_prefix("anthropic:claude-sonnet-4") == "claude-sonnet-4"
+
+    def test_ollama_model_tag_preserved(self):
+        """Ollama model:tag format must NOT be stripped."""
+        assert _strip_provider_prefix("qwen3.5:27b") == "qwen3.5:27b"
+        assert _strip_provider_prefix("llama3.3:70b") == "llama3.3:70b"
+        assert _strip_provider_prefix("gemma2:9b") == "gemma2:9b"
+        assert _strip_provider_prefix("codellama:13b-instruct-q4_0") == "codellama:13b-instruct-q4_0"
+
+    def test_http_urls_preserved(self):
+        assert _strip_provider_prefix("http://example.com") == "http://example.com"
+        assert _strip_provider_prefix("https://example.com") == "https://example.com"
+
+    def test_no_colon_returns_unchanged(self):
+        assert _strip_provider_prefix("gpt-4o") == "gpt-4o"
+        assert _strip_provider_prefix("anthropic/claude-sonnet-4") == "anthropic/claude-sonnet-4"
+
+    @patch("agent.model_metadata.fetch_model_metadata")
+    def test_ollama_model_tag_not_mangled_in_context_lookup(self, mock_fetch):
+        """Ensure 'qwen3.5:27b' is NOT reduced to '27b' during context length lookup.
+
+        We mock a custom endpoint that knows 'qwen3.5:27b' — the full name
+        must reach the endpoint metadata lookup intact.
+        """
+        mock_fetch.return_value = {}
+        with patch("agent.model_metadata.fetch_endpoint_model_metadata") as mock_ep, \
+             patch("agent.model_metadata._is_custom_endpoint", return_value=True):
+            mock_ep.return_value = {"qwen3.5:27b": {"context_length": 32768}}
+            result = get_model_context_length(
+                "qwen3.5:27b",
+                base_url="http://localhost:11434/v1",
+            )
+        assert result == 32768
+
+
 # =========================================================================
 # fetch_model_metadata — caching, TTL, slugs, failures
 # =========================================================================