[ayaangazali]

What is wrong

llm_backend picks the model-call library at import time (RouteHub when installed, litellm otherwise), and litellm is the required dependency while RouteHub is optional behind swarms[fast]. Two backends means every model-call behaviour has to hold on both.

What this changes

RouteHub becomes the only model-call backend, installed by default:

• Dependencies: litellm removed from pyproject.toml and requirements.txt; routehub[fast]>=0.2.0 becomes a required dependency and drops out of the fast extra. tqdm is now declared: agent_loader imports it and it used to arrive only through litellm, so a fresh install without it failed on import swarms.
• swarms/utils/llm_backend.py: the litellm branch, the BACKENDS tuple, the install hints and select_backend are gone; backend_module() always imports RouteHub. A leftover SWARMS_LLM_BACKEND=litellm warns once instead of switching; other values are ignored. Lazy import and the __getattr__ passthrough are unchanged.
• swarms/utils/litellm_wrapper.py: the branch that wrote module-level settings onto litellm and the RouteHub-only guard on the vision-catalog check collapse to their RouteHub paths.
• Text that named the old default: CLI strings and docstrings (cli/main.py, cli/models.py, cli/tips.py, cli/utils.py, tree_of_thoughts.py), the auto-builder prompt's model-string guidance, README.md, CLAUDE.md, SKILL.md and .env.example (SWARMS_LLM_BACKEND entry removed). cli/models.py keeps reading the litellm_provider key, because RouteHub returns the same model-info shape.
• Tests: tests/utils/test_llm_backend.py drops the backend-selection suite for RouteHub-only coverage, including the leftover-litellm warning.

Verified

RouteHub 0.2.0 exposes every name the package reads (completion, acompletion, embedding, model_list, get_model_info, get_max_tokens, encode, the supports_* checks, provider exceptions). Full offline suite, forced-RouteHub on master vs this branch: identical results (249 failed / 2135 passed both sides: the same pre-existing live-LLM-without-key failures the unit suite carries today).

Deliberately not in this PR

The litellm_wrapper module and LiteLLM class rename to a backend-neutral name (with a one-release deprecated alias) is pure churn across ~55 files with no behaviour change, so it gets its own PR, as does the sweep of the 60 example files. git grep -i litellm still names the wrapper, which the acceptance in #2781 anticipates staging.

Closes #2781

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[ayaangazali]

No change for this one. loguru is a declared runtime dependency (pyproject.toml and requirements.txt), and 45 modules under swarms/ on master already import it the same way. Pyre's environment does not install the package's dependencies, so it cannot resolve the import anywhere; it reports it here only because this PR added the line (the same alert was answered on #2451).

The test job does install dependencies and imports this module: all 9 tests in tests/utils/test_llm_backend.py pass, and the job's 246 failures and 7 errors are the same set as master's latest run on 6967b7d, all from tests that need a live OPENAI_API_KEY.

[kyegomez]

i assigned this to myself, if you didn't see