Detailed Personal AI Stack: Model Selection and Orchestration Workflow
EXM7777 · x · 2026-08-31
Shared a comprehensive personal AI workflow configuration. For model selection: GPT-5.6 Sol Low for daily driving, Fable High for planning, GPT-5.6 Sol High for coding, Opus 5 High for research, and Gemini 3.7 Flash/Kimi K3 for vision and design. Audio and video generation are handled by ElevenLabs, Seedance, and Minimax.
Infrastructure-wise, Claude and ChatGPT run on official subscriptions, while others connect via API. The toolchain covers research, SEO, email, and analytics. Orchestration uses Hermes or Orca to spawn worktrees and sessions for Claude Code or Codex. The planning loop involves Fable drafting an initial plan which is then stress-tested by GPT-5.6 Sol workers. Knowledge resides in Obsidian, with cross-platform memory unified via mem0. The entire system runs on a Hetzner AX42-1 server.
More from coding & agent
- Context and memory enable agents to act like employees — shensi · 2026-08-31
- Reference Documentation for ChatGPT Work Tool and Skill — ijidak · 2026-08-31
- Setup OpenClaw 2.0 with Gemini 3.7 Flash in Under 60 Seconds — _philschmid · 2026-08-31
- Bun WebView API Test: Agent-Built Service Uses 100MB RAM — JeremyCMorgan · 2026-08-31
- Chained multi-service agent workflows break existing sandboxes built for single calls — Common_Dream9420 · 2026-08-31
- Developers Leverage Agents Around the Clock Amid Wave of Open Weight Model Releases — antirez · 2026-08-31