snapshot: NODE1 production state 2026-02-09

Complete snapshot of /opt/microdao-daarion/ from NODE1 (144.76.224.179).
This represents the actual running production code that has diverged
significantly from the previous main branch.

Key changes from old main:
- Gateway (http_api.py): expanded from ~40KB to 164KB with full agent support
- Router: new /v1/agents/{id}/infer endpoint with vision + DeepSeek routing
- Behavior Policy: SOWA v2.2 (3-level: FULL/ACK/SILENT)
- Agent Registry: config/agent_registry.yml as single source of truth
- 13 agents configured (was 3)
- Memory service integration
- CrewAI teams and roles

Excluded from snapshot: venv/, .env, data/, backups, .tgz archives

Co-authored-by: Cursor <cursoragent@cursor.com>

services/image-gen-service/app/main.py

@@ -4,7 +4,7 @@ import os
 from typing import Optional
 
 import torch
-from diffusers import Flux2KleinPipeline
+from diffusers import DiffusionPipeline
 from fastapi import FastAPI, HTTPException
 from pydantic import BaseModel, Field
 
@@ -16,8 +16,8 @@ class GenerateRequest(BaseModel):
     negative_prompt: Optional[str] = None
     width: int = Field(1024, ge=256, le=2048)
     height: int = Field(1024, ge=256, le=2048)
-    num_inference_steps: int = Field(50, ge=1, le=100)
-    guidance_scale: float = Field(4.0, ge=0.0, le=20.0)
+    num_inference_steps: int = Field(25, ge=1, le=100)
+    guidance_scale: float = Field(3.5, ge=0.0, le=20.0)
     seed: Optional[int] = Field(None, ge=0)
 
 
@@ -32,7 +32,7 @@ def _resolve_dtype() -> torch.dtype:
     return torch.float32
 
 
-PIPELINE: Optional[Flux2KleinPipeline] = None
+PIPELINE: Optional[DiffusionPipeline] = None
 LOAD_ERROR: Optional[str] = None
 
 
@@ -40,21 +40,13 @@ def _load_pipeline() -> None:
     global PIPELINE, LOAD_ERROR
     try:
         dtype = _resolve_dtype()
-        # Use bfloat16 for FLUX.2 Klein as recommended
-        if dtype == torch.float16 and DEVICE.startswith("cuda"):
-            dtype = torch.bfloat16
-        
-        pipe = Flux2KleinPipeline.from_pretrained(
+        pipe = DiffusionPipeline.from_pretrained(
             MODEL_ID,
             torch_dtype=dtype,
+            variant="fp16" if dtype == torch.float16 else None,
+            use_safetensors=True
         )
-        
-        # Enable CPU offload to reduce VRAM usage
-        if DEVICE.startswith("cuda"):
-            pipe.enable_model_cpu_offload()
-        else:
-            pipe.to(DEVICE)
-        
+        pipe.to(DEVICE)
         PIPELINE = pipe
         LOAD_ERROR = None
     except Exception as exc:  # pragma: no cover - surface error via health/info
@@ -100,15 +92,15 @@ def generate(payload: GenerateRequest) -> dict:
 
     generator = None
     if payload.seed is not None:
-        generator = torch.Generator(device="cuda" if DEVICE.startswith("cuda") else "cpu")
+        generator = torch.Generator(device=DEVICE)
         generator.manual_seed(payload.seed)
 
     with torch.inference_mode():
         result = PIPELINE(
             prompt=payload.prompt,
-            negative_prompt=payload.negative_prompt if payload.negative_prompt else None,
-            height=payload.height,
+            negative_prompt=payload.negative_prompt,
             width=payload.width,
+            height=payload.height,
             num_inference_steps=payload.num_inference_steps,
             guidance_scale=payload.guidance_scale,
             generator=generator,