feat: add MiniMax as a new LLM engine

README.md

@@ -28,13 +28,14 @@ Works on **anything text-based** — prompts, code, configs, copy, schemas — i
 
 ### Prerequisites
 
-You need one of these CLI tools installed:
+You need one of these LLM backends:
 
 - [**Claude Code**](https://docs.anthropic.com/en/docs/claude-code) — `claude` CLI
 - [**Codex**](https://github.com/openai/codex) — `codex` CLI
 - [**Ollama**](https://ollama.com/) — run local models (Qwen, Llama, Mistral, etc.)
+- [**MiniMax**](https://www.minimax.io/) — cloud API with 204K context (`MINIMAX_API_KEY` env var)
 
-No API keys needed. No pip install. Just Python 3.10+ and an LLM.
+No pip install. Just Python 3.10+ and an LLM.
 
 ### Run it
 
@@ -75,6 +76,24 @@ python3 autoprompt.py seed.txt criteria.md -e ollama -m qwen2.5-coder:14b
 
 Fully offline. No API keys. No tokens. Just your GPU.
 
+### ☁️ Run with MiniMax cloud API
+
+```bash
+export MINIMAX_API_KEY=your-api-key
+
+# use MiniMax-M2.5 (default, 204K context)
+python3 autoprompt.py examples/prompt-optimizer/seed.txt \
+  examples/prompt-optimizer/criteria.md \
+  -e minimax --target 9.0
+
+# use the high-speed variant for faster iteration
+python3 autoprompt.py examples/prompt-optimizer/seed.txt \
+  examples/prompt-optimizer/criteria.md \
+  -e minimax -m MiniMax-M2.5-highspeed
+```
+
+No CLI tool installation required — uses Python stdlib only.
+
 ---
 
 ## 🎯 How It Works
@@ -144,8 +163,8 @@ python3 autoprompt.py config.yaml criteria.md -e codex
 | `-g, --generations` | Max generations to run | `10` |
 | `-n, --population` | Mutations per generation | `3` |
 | `-b, --bench` | Benchmark command (`{file}` = candidate path) | None |
-| `-e, --engine` | LLM backend: `claude`, `codex`, or `ollama` | `claude` |
-| `-m, --model` | Ollama model name (ignored for claude/codex) | `qwen3.5:9b` |
+| `-e, --engine` | LLM backend: `claude`, `codex`, `ollama`, or `minimax` | `claude` |
+| `-m, --model` | Model name for ollama or minimax | `qwen3.5:9b` / `MiniMax-M2.5` |
 | `--target` | Stop when score reaches this value | None |
 | `--patience` | Stop after N gens with no improvement | None |
 | `--timeout` | Stop after N seconds total | None |

autoprompt.py

@@ -20,6 +20,8 @@
 import tempfile
 import time
 import argparse
+import urllib.request
+import urllib.error
 from pathlib import Path
 
 def read(path):
@@ -45,6 +47,33 @@ def llm(prompt, engine="claude", reasoning="medium", model=None):
         if result.returncode != 0:
             raise RuntimeError(f"codex error: {result.stderr[:200]}")
         return result.stdout.strip()
+    elif engine == "minimax":
+        api_key = os.environ.get("MINIMAX_API_KEY")
+        if not api_key:
+            raise RuntimeError("MINIMAX_API_KEY environment variable is required for minimax engine")
+        m = model or "MiniMax-M2.5"
+        body = json.dumps({
+            "model": m,
+            "messages": [{"role": "user", "content": prompt}],
+            "temperature": 0.7,
+        }).encode()
+        req = urllib.request.Request(
+            "https://api.minimax.io/v1/chat/completions",
+            data=body,
+            headers={
+                "Content-Type": "application/json",
+                "Authorization": f"Bearer {api_key}",
+            },
+        )
+        try:
+            with urllib.request.urlopen(req, timeout=300) as resp:
+                data = json.loads(resp.read().decode())
+            return data["choices"][0]["message"]["content"].strip()
+        except urllib.error.HTTPError as e:
+            err_body = e.read().decode()[:200] if e.fp else str(e)
+            raise RuntimeError(f"minimax API error ({e.code}): {err_body}")
+        except urllib.error.URLError as e:
+            raise RuntimeError(f"minimax connection error: {e.reason}")
     elif engine == "ollama":
         m = model or "qwen3.5:9b"
         result = subprocess.run(
@@ -60,7 +89,7 @@ def llm(prompt, engine="claude", reasoning="medium", model=None):
         out = re.sub(r'```(?:json)?\s*', '', out).strip()
         return out
     else:
-        raise ValueError(f"Unknown engine: {engine}. Use: claude, codex, ollama")
+        raise ValueError(f"Unknown engine: {engine}. Use: claude, codex, ollama, minimax")
 
 def extract_json(text):
     """Extract JSON from LLM response (handles markdown fences etc)."""
@@ -182,7 +211,12 @@ def evolve(seed_path, criteria_path, generations=10, population=3, bench_cmd=Non
     print(f"{'='*60}")
     print(f"  seed:        {seed_path}")
     print(f"  criteria:    {criteria_path}")
-    engine_str = engine if engine != "ollama" else f"ollama ({model or 'qwen3.5:9b'})"
+    if engine == "ollama":
+        engine_str = f"ollama ({model or 'qwen3.5:9b'})"
+    elif engine == "minimax":
+        engine_str = f"minimax ({model or 'MiniMax-M2.5'})"
+    else:
+        engine_str = engine
     print(f"  engine:      {engine_str}")
     print(f"  generations: {generations} max")
     print(f"  population:  {population}/gen")
@@ -312,17 +346,23 @@ def evolve(seed_path, criteria_path, generations=10, population=3, bench_cmd=Non
 
   # use a small local model for fast iteration
   python3 autoprompt.py prompt.txt criteria.md -e ollama -m qwen3.5:2b -g 5
+
+  # use MiniMax cloud API (set MINIMAX_API_KEY env var)
+  MINIMAX_API_KEY=your-key python3 autoprompt.py prompt.txt criteria.md -e minimax
+
+  # use MiniMax high-speed model
+  python3 autoprompt.py prompt.txt criteria.md -e minimax -m MiniMax-M2.5-highspeed
 """
     )
     p.add_argument("seed", help="path to the seed file to evolve")
     p.add_argument("criteria", help="path to criteria markdown with fitness definition")
     p.add_argument("-g", "--generations", type=int, default=10, help="max generations (default: 10)")
     p.add_argument("-n", "--population", type=int, default=3, help="mutations per generation (default: 3)")
     p.add_argument("-b", "--bench", type=str, default=None, help="benchmark command ({file} = temp file path)")
-    p.add_argument("-e", "--engine", type=str, default="claude", choices=["claude", "codex", "ollama"],
-                   help="LLM engine: claude, codex, or ollama (default: claude)")
+    p.add_argument("-e", "--engine", type=str, default="claude", choices=["claude", "codex", "ollama", "minimax"],
+                   help="LLM engine: claude, codex, ollama, or minimax (default: claude)")
     p.add_argument("-m", "--model", type=str, default=None,
-                   help="model name for ollama (default: qwen3.5:9b). ignored for claude/codex")
+                   help="model name for ollama or minimax (default: qwen3.5:9b / MiniMax-M2.5)")
     # stopping conditions
     p.add_argument("--target", type=float, default=None, help="stop when score reaches this value")
     p.add_argument("--patience", type=int, default=None, help="stop after N generations with no improvement")