修复: 关闭reasoning防止gemma-4思考token泄露

付费通道: gemma-4主模型 + deepseek-v4-pro备选

Upload config.yaml with huggingface_hub

Upload config.yaml with huggingface_hub

Upload config.yaml with huggingface_hub

tune: optimize memory/compress/terminal params + add quality red lines to SOUL.md

feat: enable dual-layer memory — MEMORY.md (5K chars) + Holographic SQLite

fix: switch to free models (ling-2.6 + qwen3-coder)

swap fallback to Llama 4 Scout (109B MoE, 327K ctx, /bin/bash.08/M)

upgrade primary model to Qwen3.5-35B-A3B (MoE, 256K ctx)

fix: use dict model format with context_length override (≥64K)

add context_length override for 32K/128K models

swap models: Qwen2.5-7B primary, Llama 3.1 8B fallback (OpenRouter)

fix: WeChat session persistence - auto-reconnect after gateway restart

fix: bind api_server to 127.0.0.1 + fix watchdog process detection

feat: integrate hermes-web-ui BFF server

v4.1: 工具补全 + 压缩优化 + 健康检查

v4.0: 全面升级 SOUL.md + 模型互换 (Qwen3 Coder主/Gemma 4备)

fix: enable Pollinations image gen plugin + add SOUL.md instructions

feat: add Pollinations.ai free image generation plugin

refactor: remove MemPalace to free ~1-2GB RAM

fix: terminal backend cloud -> local (unknown type error)

feat: enable Holographic memory + MemPalace MCP

swap: Gemma 4 31B as primary, Qwen3 Coder as fallback

fix: remove :free suffix to use paid models (avoid 20rpm rate limit)

feat: Qwen3 Coder 480B as primary + Gemma 4 31B as fallback

switch model to Qwen3 Coder (free, 480B MoE)

fix: revert MCP and mempalace to restore basic functionality

feat: integrate MemPalace memory palace + MCP + SOUL.md instructions

chore: terminal=cloud, timezone display=北京时间

Hermes Bot: initial deployment with Feishu gateway