Add MiniMax provider support

.env.example

@@ -11,7 +11,7 @@
 #
 # Supported values:
 #   nebius | openai | anthropic | groq | ollama | together | fireworks |
-#   mistralai | google
+#   mistralai | google | minimax
 #
 LLM_PROVIDER=nebius
 
@@ -23,6 +23,7 @@ LLM_PROVIDER=nebius
 #   ollama     → llama3.1:8b
 #   together   → meta-llama/Llama-3.3-70B-Instruct-Turbo
 #   google     → gemini-2.0-flash
+#   minimax    → MiniMax-M3
 LLM_MODEL=meta-llama/Llama-3.3-70B-Instruct
 
 # Default sampling temperature (architectures override per-node when needed).
@@ -39,6 +40,8 @@ TOGETHER_API_KEY=
 FIREWORKS_API_KEY=
 MISTRAL_API_KEY=
 GOOGLE_API_KEY=
+MINIMAX_API_KEY=
+MINIMAX_BASE_URL=https://api.minimax.io/v1
 
 # Ollama runs locally; no key needed. Override the host if not on default port.
 OLLAMA_BASE_URL=http://localhost:11434

README.md

@@ -42,7 +42,7 @@ leaderboard that ranks every architecture against every relevant task.
 <td align="center" width="140"><h2><kbd>&nbsp; 35 &nbsp;</kbd></h2><sub>ARCHITECTURES</sub></td>
 <td align="center" width="140"><h2><kbd>&nbsp; 283 &nbsp;</kbd></h2><sub>PASSING TESTS</sub></td>
 <td align="center" width="140"><h2><kbd>&nbsp; 17 &nbsp;</kbd></h2><sub>BENCHMARK TASKS</sub></td>
-<td align="center" width="140"><h2><kbd>&nbsp; 9 &nbsp;</kbd></h2><sub>LLM PROVIDERS</sub></td>
+<td align="center" width="140"><h2><kbd>&nbsp; 10 &nbsp;</kbd></h2><sub>LLM PROVIDERS</sub></td>
 <td align="center" width="140"><h2><kbd>&nbsp; 0 &nbsp;</kbd></h2><sub>MOCKED RUNS</sub></td>
 </tr>
 </table>
@@ -55,7 +55,7 @@ leaderboard that ranks every architecture against every relevant task.
 
 ## Overview
 
-A single Python library that packages every major agentic AI pattern from the literature as a runnable `Architecture` class with a uniform contract. Each pattern ships with a fully executed Jupyter notebook whose theory is written *against* the captured run — not synthetic examples. The library is multi-provider (Nebius, OpenAI, Anthropic, Groq, Ollama, Together, Fireworks, Mistral, Google) and built on top of LangGraph state machines.
+A single Python library that packages every major agentic AI pattern from the literature as a runnable `Architecture` class with a uniform contract. Each pattern ships with a fully executed Jupyter notebook whose theory is written *against* the captured run — not synthetic examples. The library is multi-provider (Nebius, OpenAI, Anthropic, Groq, Ollama, Together, Fireworks, Mistral, Google, MiniMax) and built on top of LangGraph state machines.
 
 The central technical discipline of the repository is the **deterministic-picker pattern** — every LLM-as-Scorer surface has the LLM commit to categorical features (booleans, enums) and lets Python compose the deciding signal. This is the universal escape from the LLM-as-Scorer flat-band pathology, applied in 13 of 35 architectures; 9 more are architecturally immune by design.
 
@@ -296,6 +296,7 @@ Patterns that appear across families.
 <tr><td>Fireworks</td><td><code>[fireworks]</code></td><td>Function-calling first-class</td></tr>
 <tr><td>Mistral</td><td><code>[mistral]</code></td><td>EU-hosted option</td></tr>
 <tr><td>Google</td><td><code>[google]</code></td><td>Gemini 2.x via Generative AI API</td></tr>
+<tr><td>MiniMax</td><td><code>[openai]</code></td><td>OpenAI-compatible MiniMax-M3 with a 1M-token context window</td></tr>
 </table>
 
 Switch via `LLM_PROVIDER` + the corresponding key in `.env`. No code changes.

docs/getting-started/providers.md

@@ -1,6 +1,6 @@
 # Switching LLM providers
 
-Every architecture in this library is provider-agnostic. The same `Reflection` class works against Nebius Llama, OpenAI GPT-4o, Anthropic Claude, Groq, Ollama, Together, Fireworks, Mistral, or Google Gemini — you change one config setting.
+Every architecture in this library is provider-agnostic. The same `Reflection` class works against Nebius Llama, OpenAI GPT-4o, Anthropic Claude, Groq, Ollama, Together, Fireworks, Mistral, Google Gemini, or MiniMax — you change one config setting.
 
 ## The `get_llm()` factory
 
@@ -13,6 +13,7 @@ llm = get_llm()
 # Override per-call
 llm = get_llm(provider="openai", model="gpt-4o-mini", temperature=0.4)
 llm = get_llm(provider="anthropic", model="claude-haiku-4-5-20251001")
+llm = get_llm(provider="minimax", model="MiniMax-M3")
 llm = get_llm(provider="ollama", model="llama3.3:70b")
 ```
 
@@ -36,6 +37,7 @@ arch = Reflection(llm=llm)
 | **Fireworks** | `[fireworks]` | `FIREWORKS_API_KEY` | Function-calling first-class. |
 | **Mistral** | `[mistral]` | `MISTRAL_API_KEY` | EU-hosted option. |
 | **Google** | `[google]` | `GOOGLE_API_KEY` | Gemini-2.x via Generative AI API. |
+| **MiniMax** | `[openai]` | `MINIMAX_API_KEY` | OpenAI-compatible API with MiniMax-M3 and a 1M-token context window. |
 
 ## Environment configuration
 
@@ -53,7 +55,7 @@ LANGSMITH_API_KEY=ls-...           # optional, for tracing
 LANGSMITH_PROJECT=all-agentic-architectures
 ```
 
-To switch globally, change `LLM_PROVIDER` + the corresponding key. No code changes needed.
+To switch globally, change `LLM_PROVIDER` + the corresponding key. No code changes needed. For MiniMax's China endpoint, also set `MINIMAX_BASE_URL=https://api.minimaxi.com/v1`.
 
 ## When you'd override per-architecture
 

pyproject.toml

@@ -15,7 +15,7 @@ requires-python = ">=3.10"
 keywords = [
   "agents", "agentic", "llm", "langchain", "langgraph", "langsmith",
   "rag", "reflexion", "react", "tree-of-thoughts", "lats", "graphrag",
-  "multi-agent", "ai", "openai", "anthropic", "nebius",
+  "multi-agent", "ai", "openai", "anthropic", "nebius", "minimax",
 ]
 authors = [{ name = "Fareed Khan", email = "fareedhassankhan12@gmail.com" }]
 classifiers = [

src/agentic_architectures/config.py

@@ -36,6 +36,7 @@ def _locate_env_file() -> str:
     "fireworks",
     "mistralai",
     "google",
+    "minimax",
     "huggingface",
 ]
 
@@ -60,6 +61,8 @@ class Settings(BaseSettings):
     fireworks_api_key: SecretStr | None = None
     mistral_api_key: SecretStr | None = None
     google_api_key: SecretStr | None = None
+    minimax_api_key: SecretStr | None = None
+    minimax_base_url: str = "https://api.minimax.io/v1"
     huggingfacehub_api_token: SecretStr | None = None
     ollama_base_url: str = "http://localhost:11434"
 
@@ -110,6 +113,7 @@ def api_key_for(self, provider: Provider) -> SecretStr | None:
             "fireworks": self.fireworks_api_key,
             "mistralai": self.mistral_api_key,
             "google": self.google_api_key,
+            "minimax": self.minimax_api_key,
             "huggingface": self.huggingfacehub_api_token,
             "ollama": None,  # local, no key
         }.get(provider)

src/agentic_architectures/llm/factory.py

@@ -31,6 +31,7 @@
     "fireworks": {"tools": True, "structured_output": True},
     "mistralai": {"tools": True, "structured_output": True},
     "google": {"tools": True, "structured_output": True},
+    "minimax": {"tools": True, "structured_output": True},
     "ollama": {"tools": True, "structured_output": False},
     "huggingface": {"tools": False, "structured_output": False},
 }
@@ -75,6 +76,19 @@ def get_llm(
             **kwargs,
         )
 
+    if provider == "minimax":
+        try:
+            from langchain_openai import ChatOpenAI
+        except ImportError as e:
+            raise ImportError("MiniMax provider requires `pip install agentic-architectures[openai]`") from e
+        return ChatOpenAI(
+            model=model,
+            temperature=temperature,
+            api_key=api_key,
+            base_url=settings.minimax_base_url,
+            **kwargs,
+        )
+
     # All other providers go through langchain's init_chat_model.
     try:
         from langchain.chat_models import init_chat_model

src/agentic_architectures/llm/providers.py

@@ -90,4 +90,12 @@ class ProviderInfo:
         docs_url="https://ai.google.dev/gemini-api/docs",
         recommended_model="gemini-2.0-flash",
     ),
+    "minimax": ProviderInfo(
+        id="minimax",
+        name="MiniMax",
+        signup_url="https://platform.minimax.io",
+        docs_url="https://platform.minimax.io/docs/api-reference/api-overview",
+        recommended_model="MiniMax-M3",
+        notes="OpenAI-compatible API with a 1M-token context window.",
+    ),
 }