promptgarten 🌱
🌍 ES

Command reference · Aider

Architect Mode

Two-model mode: one model plans the change, a second editor model turns it into concrete file edits.

In architect mode, Aider first sends the request to the main model (the architect), which proposes a solution; a separate editor model (--editor-model, otherwise chosen automatically based on the main model) then translates that proposal into concrete file edits. By default the edits are applied automatically (--auto-accept-architect, default True). Especially useful with models that are strong at reasoning but weaker at precise file editing (per Aider's docs, e.g. OpenAI's o1 models, paired with an editor model like GPT-4o or Sonnet). Activate via /chat-mode architect, /architect (no prompt for a permanent switch, with a prompt for a single message), or at startup with --architect or --chat-mode architect.

✅ WHEN TO USE IT?

Pair a strong reasoning model as architect with a separate, edit-strong model

aider --architect --model o1 --editor-model anthropic/claude-sonnet-4-5 for a complex architectural decision followed by precise implementation

Get better results on a hard change even with just one model

/architect for a complex change where two separate LLM requests (planning, then editing) yield better results than a single step in code mode

⛔ WHEN NOT TO?

Simple, unambiguous changes where planning and editing clearly coincide

Architect mode needs two LLM requests instead of one, which costs time and money — unnecessary for trivial changes.

Better: Stay in default code mode

Proposed changes should be reviewed manually before landing in files

With the default --auto-accept-architect, the edits produced by the editor model are applied automatically.

Better: Set --no-auto-accept-architect to confirm every change manually

SOURCES