Isolated CODEX_HOME profiles for Codex CLI, one per Ollama model. Each profile keeps its own config, AGENTS.md, history, and skills so identities/memory stay separate.
profiles/oss20b→gpt-oss:20b-cloud(main lighter operator)profiles/oss120b→gpt-oss:120b-cloud(main larger operator)profiles/deepseek→deepseek-v3.1:671b-cloud(reasoning-heavy)profiles/qwen-coder→qwen3-coder:480b-cloud(coding specialist)profiles/kimi→kimi-k2-thinking:cloud(security/review)
Pick the profile you want and set CODEX_HOME before running Codex CLI:
CODEX_HOME="$PWD/profiles/oss20b" codex chat
CODEX_HOME="$PWD/profiles/oss120b" codex exec "build X"
CODEX_HOME="$PWD/profiles/deepseek" codex exec "plan a complex refactor"
CODEX_HOME="$PWD/profiles/qwen-coder" codex exec "implement feature Y"
CODEX_HOME="$PWD/profiles/kimi" codex exec "review for vulns"- Model per profile (see above)
- Provider: Ollama at
http://localhost:11434/v1,wire_api="responses" approval_policy=on-failure,sandbox_mode=danger-full-access(tune to your risk tolerance)- Trusted project:
/Users/cole2/Desktop/MAIN WORKSPACE(adjust if different) - Empty MCP list (add per-profile as needed)
Add a new profile by copying a folder under profiles/ with your model and settings. Each profile has its own AGENTS.md for instructions.