[REFACTOR][llm_backend][Remove litellm and make RouteHub the default and only model-call backend]

.env.example

@@ -4,7 +4,6 @@ SWARMS_VERBOSE_GLOBAL="False"
 SWARMS_API_KEY=""
 # Telemetry is ON by default. Set to false/0/no/off to opt out.
 SWARMS_TELEMETRY_ON="true"
-# Library for model calls: "routehub" (installed by swarms[fast]) or "litellm". Empty uses RouteHub when installed.
 SWARMS_LLM_BACKEND=""
 # Write log files to WORKSPACE_DIR/logs. False by default (console only).
 SWARMS_LOG_FILES="False"

CLAUDE.md

@@ -16,7 +16,7 @@ Set your LLM API key as an environment variable before running:
 export OPENAI_API_KEY="sk-..."        # OpenAI / GPT models
 export ANTHROPIC_API_KEY="sk-ant-..." # Claude models
 export GROQ_API_KEY="..."             # Groq
-# Any provider supported by LiteLLM works
+# Any provider supported by RouteHub works
 ```
 
 All imports come from the top-level `swarms` package:
@@ -103,7 +103,7 @@ print(result)
 | `agent_name` | str | `"swarm-worker-01"` | Unique name — used for memory file paths |
 | `agent_description` | str | generic | Shown to orchestrators for routing |
 | `system_prompt` | str | built-in | The agent's persona / instructions |
-| `model_name` | str | `"gpt-5.4"` | Any LiteLLM model string |
+| `model_name` | str | `"gpt-5.4"` | Any RouteHub model string |
 | `max_loops` | int \| `"auto"` | `1` | Loops before returning; `"auto"` = autonomous until done |
 | `tools` | list[Callable] | `None` | Python functions the agent can call |
 | `streaming_on` | bool | `False` | Stream tokens to stdout |
@@ -140,7 +140,7 @@ result = agent.run("Research the top 5 vector databases and compare them.")
 
 ### Model names
 
-Use any LiteLLM-compatible string:
+Use any RouteHub-compatible string:
 
 ```python
 # OpenAI

SKILL.md

@@ -22,7 +22,7 @@ Swarms is a multi-agent orchestration framework. Everything is built from one pr
 pip install -U swarms
 ```
 
-Set the key for whichever provider you use — any [LiteLLM](https://docs.litellm.ai/docs/providers) model string works:
+Set the key for whichever provider you use — any [RouteHub](https://github.com/The-Swarm-Corporation/RouteHub) model string works:
 
 ```bash
 export OPENAI_API_KEY="sk-..."
@@ -56,7 +56,7 @@ result = agent.run("Summarize the state of the semiconductor market.")
 | `agent_name` | `str` | `"swarm-worker-01"` | Unique identity; keys memory + routing |
 | `agent_description` | `str` | generic | How orchestrators decide to route to it |
 | `system_prompt` | `str` | built-in | Persona and instructions |
-| `model_name` | `str` | `"gpt-5.4"` | Any LiteLLM model string |
+| `model_name` | `str` | `"gpt-5.4"` | Any RouteHub model string |
 | `max_loops` | `int \| "auto"` | `1` | Iterations, or autonomous mode |
 | `tools` | `list[Callable]` | `None` | Python functions the agent may call |
 | `temperature` | `float` | `0.5` | Sampling temperature |

pyproject.toml

@@ -66,21 +66,20 @@ python-dotenv = "*"
 PyYAML = "*"
 networkx = "*"
 rich = "*"
-litellm = "1.76.1"
+tqdm = "*"
 httpx = "*"
 requests = "*"
 mcp = "*"
 schedule = "*"
 opentelemetry-sdk = "*"
 opentelemetry-exporter-otlp-proto-http = "*"
 zstandard = "*"
-routehub = { version = ">=0.2.0", extras = ["fast"], optional = true }
+routehub = { version = ">=0.2.0", extras = ["fast"] }
 rustworkx = { version = "*", optional = true }
 
 [tool.poetry.extras]
-# RouteHub replaces litellm for model calls when installed.
 # rustworkx backs GraphWorkflow(backend="rustworkx") in place of networkx.
-fast = ["routehub", "rustworkx"]
+fast = ["rustworkx"]
 
 [tool.poetry.scripts]
 swarms = "swarms.cli.main:main"

requirements.txt

@@ -3,13 +3,14 @@ loguru
 pydantic==2.12.5
 tenacity
 rich
+tqdm
 psutil
 python-dotenv
 PyYAML
 networkx
 httpx
 requests
-litellm==1.76.1
+routehub[fast]>=0.2.0
 mcp>=1.28.1,<3.0.0
 schedule
 opentelemetry-sdk

swarms/agents/context_compressor.py

@@ -149,7 +149,7 @@ def _summarize(self, agent: Any, history: str) -> str:
             ],
             temperature=self.summarizer_temperature,
             max_tokens=self.summarizer_max_tokens,
-            # RouteHub reads this per call; litellm otherwise inherits it from the last agent built.
+            # RouteHub reads this per call.
             drop_params=True,
         )
         call_usage = usage_from_response(response)

swarms/agents/tree_of_thoughts.py

@@ -545,7 +545,7 @@ def __init__(
             name: Name used in logs, in the conversation, and as the
                 prefix of every internal agent's name.
             description: Short description of the agent, for orchestrators.
-            model_name: Any LiteLLM model string. The model must support
+            model_name: Any RouteHub model string. The model must support
                 function calling.
             system_prompt: System prompt for every model call. Replace it to
                 give the search a domain persona; the step, evaluation and

swarms/cli/main.py

@@ -2091,7 +2091,7 @@ def handle_tips(args: argparse.Namespace) -> None:
 
 def handle_models(args: argparse.Namespace) -> None:
     """
-    Discover and inspect LLM models available via LiteLLM.
+    Discover and inspect LLM models available via RouteHub.
 
     Flags:
         --search PATTERN   Fuzzy-match against the model list.

swarms/cli/models.py

@@ -9,7 +9,7 @@
 - ``swarms models --info <name>``    show context window, capabilities, pricing
 
 All metadata comes from the LLM backend's ``model_list`` and
-``get_model_info``: litellm's bundled table, or OpenRouter's live list when
+``get_model_info``: RouteHub's bundled table, or OpenRouter's live list when
 RouteHub is installed. Either way it stays in sync as providers ship models.
 """
 
@@ -25,7 +25,7 @@
 
 
 def _provider_of(name: str) -> str:
-    """Extract a provider key from a LiteLLM model string."""
+    """Extract a provider key from a RouteHub model string."""
     if "/" in name:
         return name.split("/", 1)[0]
     # Heuristics for the common no-prefix providers

swarms/cli/tips.py

@@ -113,7 +113,7 @@
 ]
 
 MODEL_FLAG_TIPS: List[str] = [
-    "Any LiteLLM-supported model works in [bold]--model-name[/bold] — OpenAI, Anthropic, Groq, Gemini, Mistral, Cohere…",
+    "Any RouteHub-supported model works in [bold]--model-name[/bold] — OpenAI, Anthropic, Groq, Gemini, Mistral, Cohere…",
     "[bold]--model-name claude-opus-4-7[/bold] picks the latest frontier Anthropic model",
     "[bold]--model-name groq/llama-3.3-70b-versatile[/bold] is the fastest open-weight option",
     "[bold]--model-name gemini/gemini-2.5-pro[/bold] gives you million-token context natively",

swarms/cli/utils.py

@@ -307,7 +307,7 @@ def check_dependencies() -> Tuple[bool, str, str]:
     Returns:
         Tuple containing (success, status_icon, message)
     """
-    required_deps = ["torch", "transformers", "litellm", "rich"]
+    required_deps = ["torch", "transformers", "routehub", "rich"]
     missing_deps = []
 
     for dep in required_deps:

swarms/prompts/auto_agent_builder_prompt.py

@@ -97,7 +97,7 @@
 - **Structured extraction, classification, reformatting, routine summarization** — a smaller, cheaper model: `gpt-5.4-mini`, `claude-haiku-4-5-20251001`.
 - **High-throughput or latency-sensitive work** — a fast hosted model: `groq/llama-3.3-70b-versatile`.
 
-Use valid LiteLLM model strings. Provider-prefixed names route correctly (`groq/`, `together_ai/`, `openrouter/`, `gemini/`); bare OpenAI names do not need a prefix.
+Use valid RouteHub model strings. Provider-prefixed names route correctly (`groq/`, `together_ai/`, `openrouter/`, `gemini/`); bare OpenAI names do not need a prefix.
 
 Deliberately mixing tiers is a sign of good design. A roster where every agent runs the most expensive model usually means the work was never analyzed. So does one where every agent runs the cheapest.
 
@@ -129,7 +129,7 @@
 - The roster covers the task end to end with no gap.
 - Every name is unique and role-descriptive.
 - Every `system_prompt` is task-specific, several paragraphs, and would stand alone.
-- Every `model_name` is a valid LiteLLM string, chosen for that agent's actual workload.
+- Every `model_name` is a valid RouteHub string, chosen for that agent's actual workload.
 - The roster is within the requested maximum.
 
 Then call `build_agents` once with the full roster.

swarms/utils/litellm_wrapper.py

@@ -404,12 +404,7 @@ def __init__(
             "num_retries": retries,
             "drop_params": drop_params,
         }
-        if llm_backend.BACKEND == "litellm":
-            litellm = llm_backend.backend_module()
-            for name, value in backend_options.items():
-                setattr(litellm, name, value)
-        else:
-            self._backend_options = backend_options
+        self._backend_options = backend_options
 
         # Add system prompt if present (Anthropic rejects empty system blocks)
         if isinstance(self.system_prompt, str):
@@ -1213,7 +1208,6 @@ def check_if_model_supports_vision(self, img: str = None):
             # RouteHub's catalog comes from the network, so a model it can't find may still take images.
             if (
                 out is False
-                and llm_backend.BACKEND == "routehub"
                 and self.model_name not in llm_backend.model_list
             ):
                 logger.warning(

swarms/utils/llm_backend.py

@@ -1,84 +1,55 @@
 """The library swarms makes model calls through.
 
-RouteHub is used when it is installed, which pip install "swarms[fast]"
-does, and litellm otherwise. Set SWARMS_LLM_BACKEND to "litellm" or
-"routehub" to choose one explicitly.
-
-Neither library is imported until a name is first used, so importing swarms
+RouteHub is the only model-call backend and a required dependency. The
+library is not imported until a name is first used, so importing swarms
 stays fast. completion, acompletion and embedding are defined here; every
-other name, such as get_model_info or AuthenticationError, is read from the
-chosen library on first access.
+other name, such as get_model_info or AuthenticationError, is read from
+RouteHub on first access.
+
+SWARMS_LLM_BACKEND is no longer read. A leftover "litellm" value draws a
+warning so old setups learn the switch is gone; every other value is
+ignored.
 """
 
 import importlib
-import importlib.util
 import os
 from types import ModuleType
 from typing import Any
 
-BACKENDS = ("routehub", "litellm")
+from loguru import logger
 
-# The libraries refresh these themselves, so they are read on every access.
+# The library refreshes these itself, so they are read on every access.
 _UNCACHED = frozenset({"model_list", "model_cost"})
 
-_INSTALL_HINTS = {
-    "routehub": 'pip install "swarms[fast]"',
-    "litellm": "pip install litellm",
-}
-
-
-def select_backend() -> str:
-    """Name the library to make model calls through.
-
-    Returns:
-        str: "routehub" or "litellm".
-
-    Raises:
-        ValueError: If SWARMS_LLM_BACKEND names another library.
-        ImportError: If SWARMS_LLM_BACKEND names a library that is not
-            installed.
-    """
-    requested = os.getenv("SWARMS_LLM_BACKEND", "").strip().lower()
-    if not requested:
-        if importlib.util.find_spec("routehub") is not None:
-            return "routehub"
-        return "litellm"
-    if requested not in BACKENDS:
-        raise ValueError(
-            f"SWARMS_LLM_BACKEND must be one of {', '.join(BACKENDS)}, "
-            f"got {requested!r}."
-        )
-    if importlib.util.find_spec(requested) is None:
-        raise ImportError(
-            f"SWARMS_LLM_BACKEND is {requested!r}, but {requested} is not "
-            f"installed. Install it with: {_INSTALL_HINTS[requested]}"
-        )
-    return requested
-
-
-BACKEND = select_backend()
-
 _module = None
 
 
 def backend_module() -> ModuleType:
-    """Import the chosen library on first use.
+    """Import RouteHub on first use.
 
     Returns:
-        ModuleType: The routehub or litellm module.
+        ModuleType: The routehub module.
     """
     global _module
     if _module is None:
-        _module = importlib.import_module(BACKEND)
+        if (
+            os.getenv("SWARMS_LLM_BACKEND", "").strip().lower()
+            == "litellm"
+        ):
+            logger.warning(
+                "SWARMS_LLM_BACKEND=litellm is no longer supported; "
+                "swarms makes model calls through RouteHub."
+            )
+        _module = importlib.import_module("routehub")
     return _module
 
 
 def completion(*args: Any, **kwargs: Any) -> Any:
     """Make a chat completion call.
 
     Args:
-        *args (Any): Positional arguments for the library's completion.
-        **kwargs (Any): Keyword arguments for the library's completion.
+        *args (Any): Positional arguments for RouteHub's completion.
+        **kwargs (Any): Keyword arguments for RouteHub's completion.
 
     Returns:
         Any: The response, or a stream of chunks when stream is on.
@@ -90,8 +61,8 @@
     """Make an async chat completion call.
 
     Args:
-        *args (Any): Positional arguments for the library's acompletion.
-        **kwargs (Any): Keyword arguments for the library's acompletion.
+        *args (Any): Positional arguments for RouteHub's acompletion.
+        **kwargs (Any): Keyword arguments for RouteHub's acompletion.
 
     Returns:
         Any: The response, or an async stream of chunks when stream is on.
@@ -103,8 +74,8 @@
     """Make an embedding call.
 
     Args:
-        *args (Any): Positional arguments for the library's embedding.
-        **kwargs (Any): Keyword arguments for the library's embedding.
+        *args (Any): Positional arguments for RouteHub's embedding.
+        **kwargs (Any): Keyword arguments for RouteHub's embedding.
 
     Returns:
         Any: The embedding response.
@@ -113,16 +84,16 @@
 
 
 def __getattr__(name: str) -> Any:
-    """Read a name from the chosen library on first access.
+    """Read a name from RouteHub on first access.
 
     Args:
-        name (str): A name the library exports, such as get_model_info.
+        name (str): A name RouteHub exports, such as get_model_info.
 
     Returns:
         Any: The library's function, class or value.
 
     Raises:
-        AttributeError: If the library has no such name.
+        AttributeError: If RouteHub has no such name.
     """
     # Dunder lookups come from tooling such as import machinery, not callers.
     if name.startswith("__"):