feat: add MiniMax as first-class LLM provider

[octo-patch]

Summary

This PR adds first-class support for MiniMax as an LLM provider in plexe, enabling users to route any agent to MiniMax models via the minimax/ prefix with zero-config auto-routing.

What's Changed

• plexe/config.py: Added MINIMAX_API_BASE, MINIMAX_MODELS constants and auto-routing logic in get_routing_for_model() — when a model ID starts with minimax/, it auto-routes to https://api.minimax.io/v1 with the MINIMAX_API_KEY env var
• plexe/utils/litellm_wrapper.py: Rewrites minimax/ prefix → openai/ for LiteLLM compatibility, clamps temperature to MiniMax's [0, 1.0] range
• config.yaml.template: Added MiniMax configuration examples with available models
• README.md: Documented MiniMax as a supported provider with setup guide in the Multi-Provider LLM Support section

Usage

export MINIMAX_API_KEY=<your-key>

# config.yaml
hypothesiser_llm: "minimax/MiniMax-M2.7"
planner_llm: "minimax/MiniMax-M2.7"
model_definer_llm: "minimax/MiniMax-M2.5-highspeed"
litellm_drop_params: true

Available models: MiniMax-M2.7, MiniMax-M2.7-highspeed (1M context), MiniMax-M2.5, MiniMax-M2.5-highspeed (204K context).

Design Decisions

• Explicit routing config takes priority: If a user maps minimax/MiniMax-M2.7 to a custom provider in routing_config.models, that mapping is used instead of auto-routing
• No changes to agent code: All 14 agents work with MiniMax automatically through the existing get_routing_for_model() → PlexeLiteLLMModel pipeline
• Temperature clamping: MiniMax supports [0, 1.0] temperature range; values outside this range are clamped automatically

Test Plan

• 10 unit tests for model ID rewriting and temperature clamping (tests/unit/utils/test_litellm_wrapper.py)
• 10 unit tests for MiniMax auto-routing, config loading, and explicit mapping override (tests/unit/test_config.py)
• 3 integration tests for YAML config roundtrip, mixed provider config, and MiniMax API completion (tests/unit/test_minimax_integration.py)
• All 5 existing config tests continue to pass

7 files changed, 413 additions

[greptile-apps]

Greptile Summary

This PR adds MiniMax as a first-class LLM provider by extending get_routing_for_model in plexe/config.py with auto-routing for the minimax/ prefix, and adding model ID rewriting (minimax/ → openai/) plus temperature clamping in PlexeLiteLLMModel.__init__. The integration slots cleanly into the existing per-agent routing pipeline — no agent code changes are required.

• Routing logic is correct: explicit routing_config.models mappings take priority, MiniMax auto-routing is second, and the default fallback is preserved for all other providers.
• Temperature clamping at init time is a reasonable safeguard; callers that override temperature per-call (via generate/chat kwargs) would bypass it, but that code path is not used by any current agent.
• MINIMAX_MODELS is defined but never consumed by any production code — it should either be wired in (e.g. for model ID validation) or removed.
• Warning message in get_routing_for_model says "Using default routing" but for minimax/ models with a broken explicit mapping it actually falls through to MiniMax auto-routing, not the default config.
• Imports inside test functions (from plexe.utils.litellm_wrapper import LiteLLMModel) appear in 10 test methods in tests/unit/utils/test_litellm_wrapper.py, violating the project style guide ("NEVER inside functions").

Confidence Score: 4/5

• Safe to merge after fixing the misleading warning message and the style-guide import violations in tests.
• The core routing and rewriting logic is correct and well-tested. Three non-blocking issues remain: a misleading log message for an edge-case (bad provider mapping on a minimax model), unused MINIMAX_MODELS dead code, and style-guide import violations in test files. None of these affect runtime correctness for the primary user path.
• plexe/config.py (warning message) and tests/unit/utils/test_litellm_wrapper.py (imports inside functions).

Important Files Changed

Filename	Overview
plexe/config.py	Adds MiniMax routing constants and auto-routing logic in get_routing_for_model; one misleading warning message and one unused MINIMAX_MODELS constant found.
plexe/utils/litellm_wrapper.py	Adds minimax/ → openai/ model ID rewriting and temperature clamping in PlexeLiteLLMModel.__init__; logic is correct and integrates cleanly with the existing routing pipeline.
tests/unit/utils/test_litellm_wrapper.py	Good coverage of model ID rewriting and temperature clamping; violates the project style guide by importing LiteLLMModel inside 10 test method bodies instead of at module top level.
tests/unit/test_config.py	Thorough tests for auto-routing, explicit mapping override, no-key edge case, and YAML/env loading; no issues found.
tests/unit/test_minimax_integration.py	Integration tests for YAML roundtrip, mixed-provider config, and live API completion (correctly guarded by pytest.skip when MINIMAX_API_KEY is absent); no issues found.
README.md	Adds MiniMax setup guide and API-key export snippet; documentation is accurate and well-placed.
config.yaml.template	Adds commented MiniMax config example with all four model variants and a note about litellm_drop_params; no issues found.

Prompt To Fix All With AI

This is a comment left during a code review.
Path: tests/unit/utils/test_litellm_wrapper.py
Line: 22-25

Comment:
**Imports inside test functions violate style guide**

The project's style guide (`CLAUDE.md`) states: "Imports: ALWAYS at top level in order: stdlib, third-party, local; **NEVER inside functions**." The `from plexe.utils.litellm_wrapper import LiteLLMModel` import appears inside 10 separate test methods instead of at the module top level.

Move the import to the top of the file:

```suggestion
from unittest.mock import patch, MagicMock

import pytest

from plexe.utils.litellm_wrapper import LiteLLMModel, PlexeLiteLLMModel
```

Then remove the repeated `from plexe.utils.litellm_wrapper import LiteLLMModel` lines inside each test method body.

**Context Used:** CLAUDE.md ([source](https://app.greptile.com/review/custom-context?memory=72154d9b-42a4-4734-9db8-1821c15d8b84))

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: plexe/config.py
Line: 178-184

Comment:
**`MINIMAX_MODELS` is dead code — never consumed by production code**

`MINIMAX_MODELS` is defined and exported but is only referenced in the test `test_minimax_models_constant`, which just asserts the constant has certain keys. No production code path (routing, validation, context-window capping, etc.) reads from it. Either wire it into `get_routing_for_model` / `PlexeLiteLLMModel` to serve a real purpose (e.g. rejecting unknown model IDs or surfacing context-window metadata), or remove it to keep the codebase lean per the project's "prefer deleting code over adding code" principle.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: plexe/config.py
Line: 551-554

Comment:
**Misleading warning message when an explicit mapping references a non-existent provider**

When `model_id` is a `minimax/` model AND its explicit provider mapping resolves to a non-existent provider, the warning says **"Using default routing"** but execution then falls through to the MiniMax auto-routing branch — not the `config.default` path. The log message will mislead anyone debugging the routing decision.

```suggestion
            logging.getLogger(__name__).warning(
                f"Model '{model_id}' references non-existent provider '{provider_name}'. "
                "Falling through to next routing step."
            )
```

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "feat: add MiniMax as first-class LLM pro..." | Re-trigger Greptile}

[greptile-apps]

Imports inside test functions violate style guide

The project's style guide (CLAUDE.md) states: "Imports: ALWAYS at top level in order: stdlib, third-party, local; NEVER inside functions." The from plexe.utils.litellm_wrapper import LiteLLMModel import appears inside 10 separate test methods instead of at the module top level.

Move the import to the top of the file:

Suggested change

	model = PlexeLiteLLMModel(model_id="minimax/MiniMax-M2.7", temperature=0.2)
	# The super().__init__ was patched, so check the call args
	from plexe.utils.litellm_wrapper import LiteLLMModel
	from unittest.mock import patch, MagicMock
	import pytest
	from plexe.utils.litellm_wrapper import LiteLLMModel, PlexeLiteLLMModel

Then remove the repeated from plexe.utils.litellm_wrapper import LiteLLMModel lines inside each test method body.

Context Used: CLAUDE.md (source)

Prompt To Fix With AI

This is a comment left during a code review.
Path: tests/unit/utils/test_litellm_wrapper.py
Line: 22-25

Comment:
**Imports inside test functions violate style guide**

The project's style guide (`CLAUDE.md`) states: "Imports: ALWAYS at top level in order: stdlib, third-party, local; **NEVER inside functions**." The `from plexe.utils.litellm_wrapper import LiteLLMModel` import appears inside 10 separate test methods instead of at the module top level.

Move the import to the top of the file:

```suggestion
from unittest.mock import patch, MagicMock

import pytest

from plexe.utils.litellm_wrapper import LiteLLMModel, PlexeLiteLLMModel
```

Then remove the repeated `from plexe.utils.litellm_wrapper import LiteLLMModel` lines inside each test method body.

**Context Used:** CLAUDE.md ([source](https://app.greptile.com/review/custom-context?memory=72154d9b-42a4-4734-9db8-1821c15d8b84))

How can I resolve this? If you propose a fix, please make it concise.

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}

[greptile-apps]

MINIMAX_MODELS is dead code — never consumed by production code

MINIMAX_MODELS is defined and exported but is only referenced in the test test_minimax_models_constant, which just asserts the constant has certain keys. No production code path (routing, validation, context-window capping, etc.) reads from it. Either wire it into get_routing_for_model / PlexeLiteLLMModel to serve a real purpose (e.g. rejecting unknown model IDs or surfacing context-window metadata), or remove it to keep the codebase lean per the project's "prefer deleting code over adding code" principle.

Prompt To Fix With AI

This is a comment left during a code review.
Path: plexe/config.py
Line: 178-184

Comment:
**`MINIMAX_MODELS` is dead code — never consumed by production code**

`MINIMAX_MODELS` is defined and exported but is only referenced in the test `test_minimax_models_constant`, which just asserts the constant has certain keys. No production code path (routing, validation, context-window capping, etc.) reads from it. Either wire it into `get_routing_for_model` / `PlexeLiteLLMModel` to serve a real purpose (e.g. rejecting unknown model IDs or surfacing context-window metadata), or remove it to keep the codebase lean per the project's "prefer deleting code over adding code" principle.

How can I resolve this? If you propose a fix, please make it concise.

[greptile-apps]

Misleading warning message when an explicit mapping references a non-existent provider

When model_id is a minimax/ model AND its explicit provider mapping resolves to a non-existent provider, the warning says "Using default routing" but execution then falls through to the MiniMax auto-routing branch — not the config.default path. The log message will mislead anyone debugging the routing decision.

Suggested change

	if provider_name not in config.providers:
	# This should have been caught by validation, but handle gracefully
	logging.getLogger(__name__).warning(
	f"Model '{model_id}' references non-existent provider '{provider_name}'. Using default routing."
	)
	logging.getLogger(__name__).warning(
	f"Model '{model_id}' references non-existent provider '{provider_name}'. "
	"Falling through to next routing step."
	)

Prompt To Fix With AI

This is a comment left during a code review.
Path: plexe/config.py
Line: 551-554

Comment:
**Misleading warning message when an explicit mapping references a non-existent provider**

When `model_id` is a `minimax/` model AND its explicit provider mapping resolves to a non-existent provider, the warning says **"Using default routing"** but execution then falls through to the MiniMax auto-routing branch — not the `config.default` path. The log message will mislead anyone debugging the routing decision.

```suggestion
            logging.getLogger(__name__).warning(
                f"Model '{model_id}' references non-existent provider '{provider_name}'. "
                "Falling through to next routing step."
            )
```

How can I resolve this? If you propose a fix, please make it concise.