aider

Aider v0.71.0: DeepSeek Prompts & Streaming Fixes

Aider v0.71.0 brings DeepSeek optimizations, faster streaming responses, automatic model detection, and expanded context windows. Aider wrote 60% of the code.

TL;DR

Better DeepSeek handling when switching between `/ask` and `/code` modes
Streaming responses are faster and smoother for long replies
Automatic streaming detection — disables for models that don't support it
Repomap and chat history limits increased for longer context windows

New

DeepSeek mode prompts — Optimized prompts help DeepSeek work better when alternating between `/ask` and `/code` commands.
Bare command mode switching — `/ask`, `/code`, and `/architect` commands now switch chat mode directly without extra arguments.
Custom voice settings — Added support for custom voice format and input device configuration.
Token count feedback — See token counts when adding command output to chat.
Adaptive markdown rendering — Performance improved with delay based on actual render time.

Improved

Streaming performance — Pretty LLM responses stream smoother and faster for long replies.
Automatic streaming detection — Streaming now disables for models that don't support it, allowing seamless switching between `/model o1` and streaming models.
Pretty output with code fences — Fancy output stays enabled even when editing files with triple-backtick fences.
Repomap size — Increased default size for better context awareness.
Chat history limit — Raised max tokens from 4k to 8k for longer conversations.
Audio file handling — Automatic format conversion for large audio files with improved error handling.
Git error handling — Better handling of repo index errors and added AssertionError/AttributeError support.
Unicode console output — Improved handling with ASCII fallback for dumb terminals.
Terminal detection — Fancy input and file watching disabled on dumb terminals.