codeboarding 0.12.0__tar.gz → 0.12.3__tar.gz

{codeboarding-0.12.0/codeboarding.egg-info → codeboarding-0.12.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codeboarding
-Version: 0.12.0
+Version: 0.12.3
 Summary: Interactive Diagrams for Code
 Author: CodeBoarding Team
 License-Expression: MIT
@@ -40,6 +40,8 @@ Requires-Dist: markitdown>=0.1
 Requires-Dist: networkx>=3.4
 Requires-Dist: nodeenv>=1.10.0
 Requires-Dist: pathspec>=0.12
+Requires-Dist: posthog>=3.7
+Requires-Dist: pydantic>=2.0
 Requires-Dist: pyyaml>=6.0
 Requires-Dist: regex>=2024.11
 Requires-Dist: rich>=12.6
@@ -93,15 +95,17 @@ Dynamic: license-file
 The recommended way to install the CLI is with [pipx](https://pipx.pypa.io), which automatically creates an isolated environment:
 
 ```bash
-pipx install codeboarding --python python3.12
+pipx install codeboarding --python python3.12 --pip-args="--extra-index-url https://pip.codeboarding.org/simple/"
 ```
 
 Alternatively, install into an existing virtual environment with pip:
 
 ```bash
-pip install codeboarding
+pip install codeboarding --extra-index-url https://pip.codeboarding.org/simple/
 ```
 
+
+
 > Installing into the global Python environment with `pip` is not recommended — it can cause dependency conflicts and will fail if the system Python is not 3.12 or 3.13.
 
 Language server binaries are downloaded automatically on first use. To pre-install them explicitly (useful in CI or restricted environments):

{codeboarding-0.12.0 → codeboarding-0.12.3}/PYPI.md

@@ -23,15 +23,17 @@
 The recommended way to install the CLI is with [pipx](https://pipx.pypa.io), which automatically creates an isolated environment:
 
 ```bash
-pipx install codeboarding --python python3.12
+pipx install codeboarding --python python3.12 --pip-args="--extra-index-url https://pip.codeboarding.org/simple/"
 ```
 
 Alternatively, install into an existing virtual environment with pip:
 
 ```bash
-pip install codeboarding
+pip install codeboarding --extra-index-url https://pip.codeboarding.org/simple/
 ```
 
+
+
 > Installing into the global Python environment with `pip` is not recommended — it can cause dependency conflicts and will fail if the system Python is not 3.12 or 3.13.
 
 Language server binaries are downloaded automatically on first use. To pre-install them explicitly (useful in CI or restricted environments):

{codeboarding-0.12.0 → codeboarding-0.12.3}/README.md

@@ -4,9 +4,9 @@ See what your AI is building before it breaks.
 
 CodeBoarding gives developers and coding agents a visual map of a codebase. It combines static analysis with LLM reasoning to generate architecture diagrams, component-level documentation, and navigable outputs you can use in your IDE, CI, and docs.
 
-[Website](https://codeboarding.org) · [Open VSX extension](https://open-vsx.org/extension/CodeBoarding/codeboarding) · [Explore examples](https://codeboarding.org/diagrams) · [VS Code extension](https://marketplace.visualstudio.com/items?itemName=Codeboarding.codeboarding) · [GitHub Action](https://github.com/marketplace/actions/codeboarding-diagram-first-documentation) ·[Discord](https://discord.gg/T5zHTJYFuy)
+[Website](https://codeboarding.org) <img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=0855d476-b2d0-44cc-b93d-69b47504719c" width="0" height="0" /> · [Open VSX extension](https://open-vsx.org/extension/CodeBoarding/codeboarding) <img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=ce87464c-2792-46b0-9ea2-87eefe853d7e" width="0" height="0" /> · [Explore examples](https://codeboarding.org/diagrams) · [VS Code extension](https://marketplace.visualstudio.com/items?itemName=Codeboarding.codeboarding) <img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=8a3d26e0-6f6b-49c0-8482-114445de56a5" width="0" height="0" /> · [GitHub Action](https://github.com/marketplace/actions/codeboarding-action) ·[Discord](https://discord.gg/T5zHTJYFuy)
 
-[![CodeBoarding demo](https://gist.githubusercontent.com/ivanmilevtues/1c4f921066613516cfd7b938014a6877/raw/611aec7711556807860ff2e1679a5dc4c0c23fed/CodeBoarding_extension_demo.gif)](https://open-vsx.org/extension/CodeBoarding/codeboarding)
+[![CodeBoarding demo](https://gist.githubusercontent.com/ivanmilevtues/1c4f921066613516cfd7b938014a6877/raw/611aec7711556807860ff2e1679a5dc4c0c23fed/CodeBoarding_extension_demo.gif)](https://open-vsx.org/extension/CodeBoarding/codeboarding) <img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=ce87464c-2792-46b0-9ea2-87eefe853d7e" width="0" height="0" />
 
 Install the extension from Open VSX.
 
@@ -86,7 +86,7 @@ codeboarding full --local /path/to/repo
 Or, if you prefer pip, install into a virtual environment (not the global Python):
 
 ```bash
-pip install codeboarding
+pip install codeboarding --extra-index-url https://pip.codeboarding.org/simple/
 codeboarding-setup
 codeboarding full --local /path/to/repo
 ```
@@ -108,6 +108,8 @@ On first run, CodeBoarding creates `~/.codeboarding/config.toml`. Set one provid
 # aws_bearer_token_bedrock  = "..."
 # ollama_base_url           = "http://localhost:11434"
 # openrouter_api_key        = "sk-..."
+# litellm_base_url          = "http://localhost:4000"  # LiteLLM proxy server URL (required)
+# litellm_api_key           = "sk-..."           # LiteLLM proxy server key (optional)
 
 [llm]
 # agent_model   = "gemini-3-flash"
@@ -138,19 +140,32 @@ python main.py full https://github.com/pytorch/pytorch
 ## Where to use it
 
 - [CLI](https://github.com/CodeBoarding/CodeBoarding) for local analysis, automation, and CI workflows.
-- [VS Code extension](https://marketplace.visualstudio.com/items?itemName=Codeboarding.codeboarding) for in-editor visual architecture.
-- [GitHub Action](https://github.com/marketplace/actions/codeboarding-diagram-first-documentation) to keep diagrams updated in CI.
+- [VS Code extension](https://marketplace.visualstudio.com/items?itemName=Codeboarding.codeboarding) <img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=8a3d26e0-6f6b-49c0-8482-114445de56a5" width="0" height="0" /> for in-editor visual architecture.
+- [GitHub Action](https://github.com/marketplace/actions/codeboarding-action) to keep diagrams updated in CI.
 
 ## Supported stack
 
 - Languages: Python, TypeScript, JavaScript, Java, Go, PHP, Rust, C#.
-- LLM providers: OpenAI, Anthropic, Google, Vercel AI Gateway, AWS Bedrock, Ollama, OpenRouter, and more.
+- LLM providers: OpenAI, Anthropic, Google, Vercel AI Gateway, AWS Bedrock, Ollama, OpenRouter, LiteLLM proxy, and more.
 
 ## Examples
 
 - Visualized 800+ open-source repositories.
 - Browse generated examples in [GeneratedOnBoardings](https://github.com/CodeBoarding/GeneratedOnBoardings).
-- Try the hosted explorer at [codeboarding.org/diagrams](https://codeboarding.org/diagrams).
+- Try the hosted explorer at [codeboarding.org/diagrams](https://codeboarding.org/diagrams) <img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=0855d476-b2d0-44cc-b93d-69b47504719c" width="0" height="0" />.
+
+## Telemetry
+
+CodeBoarding collects anonymous, aggregate usage telemetry (which command ran,
+success/failure, duration, and token cost) to help us improve the tool. It is on
+by default and never collects source code, file names, repository names, paths,
+prompts, model outputs, API keys, or any personal information. Opt out anytime:
+
+```bash
+export CODEBOARDING_TELEMETRY=false   # or: export DO_NOT_TRACK=1
+```
+
+See [TELEMETRY.md](TELEMETRY.md) for the full list of events and properties.
 
 ## Contributing
 
@@ -159,3 +174,5 @@ If you want to improve CodeBoarding, open an [issue](https://github.com/CodeBoar
 ## Vision
 
 CodeBoarding is building an open standard for code understanding: a visual, accurate, high-level representation of a codebase that both humans and agents can use.
+
+<img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=1942e7e7-0762-4cdd-9f08-024acc098071" />

{codeboarding-0.12.0 → codeboarding-0.12.3}/agents/cluster_methods_mixin.py

@@ -16,7 +16,8 @@ from agents.agent_responses import (
     MethodEntry,
 )
 from agents.cluster_budget import ClusterPromptBudget
-from agents.llm_config import get_current_agent_context_window
+from agents.llm_config import get_current_agent_context_window, get_current_agent_model_ref
+from agents.model_capabilities import ContextWindow
 from constants import MIN_CLUSTERS_THRESHOLD
 from static_analyzer.analysis_result import StaticAnalysisResults
 from static_analyzer.cfg_skip_planner import ContextBudgetExceededError, plan_skip_set
@@ -46,6 +47,20 @@ class _RenderedClusterString:
     cluster_ids: set[int]
 
 
+def _describe_window(ctx: ContextWindow) -> str:
+    suffix = "; fallback default, model window unresolved" if ctx.is_fallback else ""
+    return f"{ctx.input_tokens} input tokens for {get_current_agent_model_ref()}{suffix}"
+
+
+def _window_telemetry(ctx: ContextWindow, char_budget: int) -> dict:
+    return {
+        "char_budget": char_budget,
+        "window_input_tokens": ctx.input_tokens,
+        "window_is_fallback": ctx.is_fallback,
+        "agent_model": get_current_agent_model_ref(),
+    }
+
+
 class ClusterMethodsMixin:
     """
     Mixin providing shared cluster-related functionality for agents.
@@ -156,10 +171,10 @@ class ClusterMethodsMixin:
             ctx = get_current_agent_context_window()
             msg = (
                 f"Prompt overhead ({prompt_overhead_chars} chars) consumes the entire agent input "
-                f"window ({ctx.input_tokens} tokens); no room for cluster renderings."
+                f"window ({_describe_window(ctx)}); no room for cluster renderings."
             )
             logger.error("[CFG skip planner] %s", msg)
-            raise ContextBudgetExceededError(msg)
+            raise ContextBudgetExceededError(msg, telemetry_properties=_window_telemetry(ctx, char_budget))
 
         langs_with_clusters = [l for l in programming_langs if cluster_results.get(l)]
         if not langs_with_clusters:
@@ -240,14 +255,21 @@ class ClusterMethodsMixin:
         rendered: _RenderedClusterString,
         skip_sets: dict[str, set[str]],
     ) -> NoReturn:
+        ctx = get_current_agent_context_window()
         per_lang_sizes = {lang: len(text) for lang, text in rendered.by_language.items()}
         skipped_counts = {lang: len(skip) for lang, skip in skip_sets.items() if skip}
         msg = (
-            f"Cluster render {len(rendered.text)} chars exceeds budget {char_budget}. "
+            f"Cluster render {len(rendered.text)} chars exceeds budget {char_budget} "
+            f"(agent window: {_describe_window(ctx)}). "
             f"Per-language sizes: {per_lang_sizes}; skipped nodes: {skipped_counts}."
         )
         logger.error("[CFG skip planner] %s", msg)
-        raise ContextBudgetExceededError(msg)
+        telemetry = _window_telemetry(ctx, char_budget) | {
+            "render_chars": len(rendered.text),
+            "per_language_chars": per_lang_sizes,
+            "skipped_node_counts": skipped_counts,
+        }
+        raise ContextBudgetExceededError(msg, telemetry_properties=telemetry)
 
     @staticmethod
     def _cluster_prompt_budget(prompt_overhead_chars: int) -> int:

{codeboarding-0.12.0 → codeboarding-0.12.3}/agents/llm_config.py

@@ -23,6 +23,25 @@ MONITORING_CALLBACK = MonitoringCallback(stats_container=RunStats())
 
 logger = logging.getLogger(__name__)
 
+_OPENROUTER_FALLBACK_CONTEXT_WINDOW = ContextWindow(1_048_576, 65_536, is_fallback=True)
+
+# Model families that reject sampling params (temperature/top_p/top_k) with HTTP 400.
+# Why: Anthropic removed them starting with Opus 4.7; sending temperature (even 0) 400s.
+# Substrings match bare IDs and Bedrock-prefixed forms (e.g. us.anthropic.claude-opus-4-8-v1:0).
+_SAMPLING_PARAM_FREE_MODEL_SUBSTRINGS = (
+    "claude-opus-4-7",
+    "claude-opus-4-8",
+    "claude-fable-5",
+    "claude-mythos-5",
+)
+
+
+def _model_accepts_temperature(model_name: str) -> bool:
+    """False for models that reject sampling params; temperature must be omitted for those."""
+    lowered = model_name.lower()
+    return not any(s in lowered for s in _SAMPLING_PARAM_FREE_MODEL_SUBSTRINGS)
+
+
 # ---------------------------------------------------------------------------
 # Module-level model overrides – set once by the orchestrator (main.py) and
 # consumed by initialize_llms() without needing to thread the values through
@@ -67,6 +86,9 @@ class LLMConfig:
     Configuration for LLM providers.
 
     Attributes:
+        selection_envs: Env vars that select this provider — any one being set selects it.
+        api_key_env: Env var holding the provider's secret, or None when the
+                     underlying SDK reads its credentials from the environment itself.
         agent_model: The "agent" model used for complex reasoning and agentic tasks.
         parsing_model: The "parsing" model used for fast, cost-effective extraction and parsing tasks.
         agent_temperature: Temperature for the agent model. Defaults to 0 for deterministic behavior
@@ -77,23 +99,37 @@ class LLMConfig:
     """
 
     chat_class: Type[BaseChatModel]
-    api_key_env: str
+    selection_envs: list[str]
     agent_model: str
     parsing_model: str
     llm_type: LLMType
     agent_temperature: float = LLMDefaults.DEFAULT_AGENT_TEMPERATURE
     parsing_temperature: float = LLMDefaults.DEFAULT_PARSING_TEMPERATURE
     extra_args: dict[str, Any] = field(default_factory=dict)
-    alt_env_vars: list[str] = field(default_factory=list)
+    api_key_env: str | None = None
+    keyless_capable: bool = False
+    """Whether this provider can run without a real API key.
+
+    True for self-hosted / OpenAI-compatible endpoints that accept
+    unauthenticated requests. When such a provider is the sole selected one
+    and no real key is set, key validation warns instead of failing, and the
+    client uses a placeholder.
+    """
 
     def get_api_key(self) -> str | None:
-        return os.getenv(self.api_key_env)
+        return os.getenv(self.api_key_env) if self.api_key_env else None
+
+    def has_real_api_key(self) -> bool:
+        """True if the provider's API-key env var holds a value.
+
+        Distinct from ``is_selected_by_env()``: a keyless-capable provider can
+        be selected via a base-URL var while having no real key here.
+        """
+        return bool(self.get_api_key())
 
-    def is_active(self) -> bool:
-        """Check if any of the environment variables (primary or alternate) are set."""
-        if os.getenv(self.api_key_env):
-            return True
-        return any(os.getenv(var) for var in self.alt_env_vars)
+    def is_selected_by_env(self) -> bool:
+        """True when any of this provider's selection env vars is set."""
+        return any(os.getenv(var) for var in self.selection_envs)
 
     def get_resolved_extra_args(self) -> dict[str, Any]:
         resolved = {}
@@ -108,11 +144,12 @@ class LLMConfig:
 LLM_PROVIDERS = {
     "openai": LLMConfig(
         chat_class=ChatOpenAI,
+        selection_envs=["OPENAI_API_KEY", "OPENAI_BASE_URL"],
         api_key_env="OPENAI_API_KEY",
         agent_model="gpt-4o",
         parsing_model="gpt-4o-mini",
         llm_type=LLMType.GPT4,
-        alt_env_vars=["OPENAI_BASE_URL"],
+        keyless_capable=True,
         extra_args={
             "base_url": lambda: os.getenv("OPENAI_BASE_URL"),
             "max_tokens": None,
@@ -122,11 +159,11 @@ LLM_PROVIDERS = {
     ),
     "vercel": LLMConfig(
         chat_class=ChatOpenAI,
+        selection_envs=["VERCEL_API_KEY", "VERCEL_BASE_URL"],
         api_key_env="VERCEL_API_KEY",
         agent_model="google/gemini-3-flash",
         parsing_model="openai/gpt-5-mini",
         llm_type=LLMType.GEMINI_FLASH,
-        alt_env_vars=["VERCEL_BASE_URL"],
         extra_args={
             "base_url": lambda: os.getenv("VERCEL_BASE_URL", f"https://ai-gateway.vercel.sh/v1"),
             "max_tokens": None,
@@ -136,6 +173,7 @@ LLM_PROVIDERS = {
     ),
     "anthropic": LLMConfig(
         chat_class=ChatAnthropic,
+        selection_envs=["ANTHROPIC_API_KEY"],
         api_key_env="ANTHROPIC_API_KEY",
         agent_model="claude-sonnet-4-6",
         parsing_model="claude-haiku-4-5",
@@ -148,6 +186,7 @@ LLM_PROVIDERS = {
     ),
     "google": LLMConfig(
         chat_class=ChatGoogleGenerativeAI,
+        selection_envs=["GOOGLE_API_KEY"],
         api_key_env="GOOGLE_API_KEY",
         agent_model="gemini-3-flash-preview",
         parsing_model="gemini-3.1-flash-lite",
@@ -160,7 +199,8 @@ LLM_PROVIDERS = {
     ),
     "aws": LLMConfig(
         chat_class=ChatBedrockConverse,
-        api_key_env="AWS_BEARER_TOKEN_BEDROCK",  # Used for existence check
+        # No api_key_env: botocore reads AWS_BEARER_TOKEN_BEDROCK from the environment itself.
+        selection_envs=["AWS_BEARER_TOKEN_BEDROCK"],
         agent_model="anthropic.claude-sonnet-4-6",
         parsing_model="claude-haiku-4-5",
         llm_type=LLMType.CLAUDE_SONNET,
@@ -172,6 +212,7 @@ LLM_PROVIDERS = {
     ),
     "cerebras": LLMConfig(
         chat_class=ChatCerebras,
+        selection_envs=["CEREBRAS_API_KEY"],
         api_key_env="CEREBRAS_API_KEY",
         agent_model="zai-glm-4.7",
         parsing_model="gpt-oss-120b",
@@ -184,7 +225,11 @@ LLM_PROVIDERS = {
     ),
     "ollama": LLMConfig(
         chat_class=ChatOllama,
-        api_key_env="OLLAMA_BASE_URL",  # Used for existence check
+        # OLLAMA_HOST is Ollama's canonical host var; the client falls back to it
+        # when no base_url is passed, and sends OLLAMA_API_KEY (Ollama cloud) itself.
+        selection_envs=["OLLAMA_BASE_URL", "OLLAMA_HOST"],
+        api_key_env="OLLAMA_API_KEY",
+        keyless_capable=True,
         agent_model="qwen3:30b",
         parsing_model="qwen2.5:7b",
         llm_type=LLMType.GEMINI_FLASH,
@@ -196,11 +241,11 @@ LLM_PROVIDERS = {
     ),
     "deepseek": LLMConfig(
         chat_class=ChatOpenAI,
+        selection_envs=["DEEPSEEK_API_KEY", "DEEPSEEK_BASE_URL"],
         api_key_env="DEEPSEEK_API_KEY",
         agent_model="deepseek-chat",
         parsing_model="deepseek-chat",
         llm_type=LLMType.DEEPSEEK,
-        alt_env_vars=["DEEPSEEK_BASE_URL"],
         extra_args={
             "base_url": lambda: os.getenv("DEEPSEEK_BASE_URL", "https://api.deepseek.com/v1"),
             "max_tokens": None,
@@ -210,11 +255,11 @@ LLM_PROVIDERS = {
     ),
     "glm": LLMConfig(
         chat_class=ChatOpenAI,
+        selection_envs=["GLM_API_KEY", "GLM_BASE_URL"],
         api_key_env="GLM_API_KEY",
         agent_model="glm-4.7-flash",
         parsing_model="glm-4.7-flash",
         llm_type=LLMType.GLM,
-        alt_env_vars=["GLM_BASE_URL"],
         extra_args={
             "base_url": lambda: os.getenv("GLM_BASE_URL", "https://open.bigmodel.cn/api/paas/v4"),
             "max_tokens": None,
@@ -224,11 +269,11 @@ LLM_PROVIDERS = {
     ),
     "kimi": LLMConfig(
         chat_class=ChatOpenAI,
+        selection_envs=["KIMI_API_KEY", "KIMI_BASE_URL"],
         api_key_env="KIMI_API_KEY",
         agent_model="kimi-k2.5",
         parsing_model="kimi-k2.5",
         llm_type=LLMType.KIMI,
-        alt_env_vars=["KIMI_BASE_URL"],
         extra_args={
             "base_url": lambda: os.getenv("KIMI_BASE_URL", "https://api.moonshot.cn/v1"),
             "max_tokens": None,
@@ -238,6 +283,7 @@ LLM_PROVIDERS = {
     ),
     "openrouter": LLMConfig(
         chat_class=ChatOpenAI,
+        selection_envs=["OPENROUTER_API_KEY"],
         api_key_env="OPENROUTER_API_KEY",
         agent_model="google/gemini-3-flash-preview",
         parsing_model="google/gemini-3-flash-preview",
@@ -249,9 +295,45 @@ LLM_PROVIDERS = {
             "max_retries": 0,
         },
     ),
+    "litellm": LLMConfig(
+        chat_class=ChatOpenAI,
+        # Base URL only: a key alone must not select litellm, since there is no
+        # universal proxy endpoint to default to.
+        selection_envs=["LITELLM_BASE_URL"],
+        api_key_env="LITELLM_API_KEY",
+        agent_model="gpt-4o",
+        parsing_model="gpt-4o-mini",
+        llm_type=LLMType.GPT4,
+        keyless_capable=True,
+        extra_args={
+            "base_url": lambda: os.getenv("LITELLM_BASE_URL"),
+            "max_tokens": None,
+            "timeout": None,
+            "max_retries": 0,
+        },
+    ),
 }
 
 
+def _all_selection_envs() -> list[str]:
+    return sorted({var for config in LLM_PROVIDERS.values() for var in config.selection_envs})
+
+
+def _unselected_key_hints() -> list[str]:
+    """Messages for providers whose API key is set but which nothing selects."""
+    return [
+        f"{config.api_key_env} is set, but the '{name}' provider is selected by "
+        f"{' or '.join(config.selection_envs)}."
+        for name, config in LLM_PROVIDERS.items()
+        if config.has_real_api_key() and not config.is_selected_by_env()
+    ]
+
+
+def selected_providers() -> list[str]:
+    """Names of providers the environment currently selects."""
+    return [name for name, config in LLM_PROVIDERS.items() if config.is_selected_by_env()]
+
+
 def _initialize_llm(
     model_override: str | None,
     model_attr: str,
@@ -259,16 +341,10 @@ def _initialize_llm(
     log_prefix: str,
     init_factory: bool = False,
 ) -> tuple[BaseChatModel, str]:
-    resolved = _resolve_active_provider(model_override, model_attr)
+    resolved = _resolve_selected_provider(model_override, model_attr)
     if resolved is None:
-        required_vars = []
-        for config in LLM_PROVIDERS.values():
-            required_vars.append(config.api_key_env)
-            required_vars.extend(config.alt_env_vars)
-
-        raise ValueError(
-            f"No valid LLM configuration found. Please set one of: {', '.join(sorted(set(required_vars)))}"
-        )
+        message = f"No valid LLM configuration found. Please set one of: {', '.join(_all_selection_envs())}."
+        raise ValueError(" ".join([message, *_unselected_key_hints()]))
 
     name, config, model_name = resolved
 
@@ -282,12 +358,13 @@ def _initialize_llm(
 
     logger.info(f"Using {name.title()} {log_prefix}LLM with model: {model_name}")
 
-    kwargs = {
-        "model": model_name,
-        "temperature": getattr(config, temperature_attr),
-    }
+    kwargs: dict[str, Any] = {"model": model_name}
+    if _model_accepts_temperature(model_name):
+        kwargs["temperature"] = getattr(config, temperature_attr)
     kwargs.update(config.get_resolved_extra_args())
 
+    # ChatBedrockConverse and ChatOllama take no api_key kwarg; their SDKs read
+    # AWS_BEARER_TOKEN_BEDROCK / OLLAMA_API_KEY from the environment directly.
     if name not in ["aws", "ollama"]:
         api_key = config.get_api_key()
         kwargs["api_key"] = api_key or "no-key-required"
@@ -296,13 +373,13 @@ def _initialize_llm(
     return model, model_name
 
 
-def _resolve_active_provider(
+def _resolve_selected_provider(
     model_override: str | None,
     model_attr: str,
 ) -> tuple[str, LLMConfig, str] | None:
-    """Return the active provider, config, and resolved model name."""
+    """Return the selected provider, config, and resolved model name."""
     for name, config in LLM_PROVIDERS.items():
-        if not config.is_active():
+        if not config.is_selected_by_env():
             continue
         return name, config, model_override or getattr(config, model_attr)
     return None
@@ -313,13 +390,36 @@ class LLMConfigError(ValueError):
 
 
 def validate_api_key_provided() -> None:
-    """Raise LLMConfigError if zero or more than one LLM provider key is configured."""
-    active = [name for name, config in LLM_PROVIDERS.items() if config.is_active()]
-    if not active:
-        required = sorted({config.api_key_env for config in LLM_PROVIDERS.values()})
-        raise LLMConfigError(f"No LLM provider API key found. Set one of: {', '.join(required)}")
-    if len(active) > 1:
-        raise LLMConfigError(f"Multiple LLM provider keys detected ({', '.join(active)}); please set only one.")
+    """Raise LLMConfigError unless exactly one LLM provider is selected.
+
+    A provider is selected when any of its ``selection_envs`` is set. Keyless-
+    capable providers (self-hosted / OpenAI-compatible endpoints, e.g. an
+    ``OPENAI_BASE_URL`` with no key) are therefore valid: they are selected by
+    their base URL, and the client falls back to a placeholder key downstream.
+    In that case we log a warning rather than fail. Ambiguity detection (more
+    than one selected provider) is preserved so a stray second key is still
+    surfaced, and a key set for an unselected provider (e.g. LITELLM_API_KEY
+    without LITELLM_BASE_URL) is reported rather than silently ignored.
+    """
+    hints = _unselected_key_hints()
+    selected = selected_providers()
+    if not selected:
+        message = f"No LLM provider selected. Set one of: {', '.join(_all_selection_envs())}."
+        raise LLMConfigError(" ".join([message, *hints]))
+    if len(selected) > 1:
+        raise LLMConfigError(f"Multiple LLM providers selected ({', '.join(selected)}); please set only one.")
+    for hint in hints:
+        logger.warning(hint)
+
+    (name,) = selected
+    config = LLM_PROVIDERS[name]
+    if config.keyless_capable and not config.has_real_api_key():
+        logger.warning(
+            "Provider '%s' is selected via a base URL with no %s set; "
+            "treating as a keyless local endpoint (a placeholder key is used).",
+            name,
+            config.api_key_env,
+        )
 
 
 def initialize_agent_llm(model_override: str | None = None) -> BaseChatModel:
@@ -329,17 +429,29 @@ def initialize_agent_llm(model_override: str | None = None) -> BaseChatModel:
 
 
 def get_current_agent_context_window() -> ContextWindow:
-    """Context window for the currently active agent provider/model.
+    """Context window for the currently selected agent provider/model.
 
-    Resolves the first active provider (same rule as ``_initialize_llm``) on
+    Resolves the first selected provider (same rule as ``_initialize_llm``) on
     every call. ``get_context_window`` handles its own caching, so this is
     cheap enough to call without a module-level cache.
     """
-    resolved = _resolve_active_provider(_agent_model_override or os.getenv("AGENT_MODEL"), "agent_model")
+    resolved = _resolve_selected_provider(_agent_model_override or os.getenv("AGENT_MODEL"), "agent_model")
     if resolved is not None:
         name, _config, model_name = resolved
-        return get_context_window(name, model_name)
-    return ContextWindow(ModelCapabilities.FALLBACK_INPUT, ModelCapabilities.FALLBACK_OUTPUT)
+        ctx = get_context_window(name, model_name)
+        if name == "openrouter" and ctx.is_fallback:
+            return _OPENROUTER_FALLBACK_CONTEXT_WINDOW
+        return ctx
+    return ContextWindow(ModelCapabilities.FALLBACK_INPUT, ModelCapabilities.FALLBACK_OUTPUT, is_fallback=True)
+
+
+def get_current_agent_model_ref() -> str:
+    """``provider/model`` for the currently active agent LLM, or ``"unknown"``."""
+    resolved = _resolve_selected_provider(_agent_model_override or os.getenv("AGENT_MODEL"), "agent_model")
+    if resolved is None:
+        return "unknown"
+    name, _config, model_name = resolved
+    return f"{name}/{model_name}"
 
 
 def initialize_parsing_llm(model_override: str | None = None) -> BaseChatModel:

{codeboarding-0.12.0 → codeboarding-0.12.3}/agents/model_capabilities.py

@@ -24,6 +24,7 @@ _OLLAMA_CACHE: dict[tuple[str, str], tuple[int, int]] = {}
 class ContextWindow:
     input_tokens: int
     output_tokens: int
+    is_fallback: bool = False
 
 
 def get_context_window(provider: str, model_name: str) -> ContextWindow:
@@ -40,7 +41,7 @@ def get_context_window(provider: str, model_name: str) -> ContextWindow:
         if hit is not None:
             return ContextWindow(*hit)
     logger.warning(f"No context window for {provider}/{model_name}; using fallback {ModelCapabilities.FALLBACK_INPUT}")
-    return ContextWindow(ModelCapabilities.FALLBACK_INPUT, ModelCapabilities.FALLBACK_OUTPUT)
+    return ContextWindow(ModelCapabilities.FALLBACK_INPUT, ModelCapabilities.FALLBACK_OUTPUT, is_fallback=True)
 
 
 def _resolve_env(provider: str, model_name: str) -> tuple[int, int] | None:
@@ -81,9 +82,12 @@ def _user_context_window_override() -> int | None:
 def _resolve_ollama(provider: str, model_name: str) -> tuple[int, int] | None:
     if provider != "ollama":
         return None
-    base = os.getenv("OLLAMA_BASE_URL")
+    base = os.getenv("OLLAMA_BASE_URL") or os.getenv("OLLAMA_HOST")
     if not base:
         return None
+    if "://" not in base:
+        # OLLAMA_HOST conventionally allows bare host:port.
+        base = f"http://{base}"
     return _ollama_show(model_name, base.rstrip("/"))
 
 

{codeboarding-0.12.0 → codeboarding-0.12.3}/agents/tools/read_file_structure.py

@@ -36,7 +36,7 @@ class FileStructureTool(BaseRepoTool):
         # Ensure they are sorted by depth for is_subsequence logic
         return sorted(dirs, key=lambda x: len(x.parts))
 
-    def _run(self, dir: str) -> str:
+    def _run(self, dir: str = ".") -> str:
         """
         Run the tool with the given input.
         """

{codeboarding-0.12.0 → codeboarding-0.12.3/codeboarding.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codeboarding
-Version: 0.12.0
+Version: 0.12.3
 Summary: Interactive Diagrams for Code
 Author: CodeBoarding Team
 License-Expression: MIT
@@ -40,6 +40,8 @@ Requires-Dist: markitdown>=0.1
 Requires-Dist: networkx>=3.4
 Requires-Dist: nodeenv>=1.10.0
 Requires-Dist: pathspec>=0.12
+Requires-Dist: posthog>=3.7
+Requires-Dist: pydantic>=2.0
 Requires-Dist: pyyaml>=6.0
 Requires-Dist: regex>=2024.11
 Requires-Dist: rich>=12.6
@@ -93,15 +95,17 @@ Dynamic: license-file
 The recommended way to install the CLI is with [pipx](https://pipx.pypa.io), which automatically creates an isolated environment:
 
 ```bash
-pipx install codeboarding --python python3.12
+pipx install codeboarding --python python3.12 --pip-args="--extra-index-url https://pip.codeboarding.org/simple/"
 ```
 
 Alternatively, install into an existing virtual environment with pip:
 
 ```bash
-pip install codeboarding
+pip install codeboarding --extra-index-url https://pip.codeboarding.org/simple/
 ```
 
+
+
 > Installing into the global Python environment with `pip` is not recommended — it can cause dependency conflicts and will fail if the system Python is not 3.12 or 3.13.
 
 Language server binaries are downloaded automatically on first use. To pre-install them explicitly (useful in CI or restricted environments):