You’re working with Aider and get:

Warning: messages exceed model context window (32768 tokens)

The conversation plus file contents exceeded your model’s context limit. Here are the fixes, from quickest to most thorough.

Fix 1: Drop files from context

Aider loads every added file into context. Remove files you’re not actively editing:

/drop src/large-file.ts
/drop src/utils/helpers.ts

# See what's loaded
/tokens

Only keep files Aider needs to edit. Reference files can be mentioned in your prompt instead of added to context.

Fix 2: Use a model with larger context

Model	Context	Command
Qwen3 8B	32K	aider --model ollama/qwen3:8b
Claude Sonnet	200K	aider --model claude-sonnet-4
GPT-5.4	1M	aider --model gpt-5.4
GPT-5.4 Mini	400K	aider --model gpt-5.4-mini
DeepSeek Chat	128K	aider --model deepseek/deepseek-chat

Switching from a 32K model to Claude Sonnet (200K) gives you 6x more context.

Fix 3: Reduce map tokens

Aider creates a “repo map” of your codebase. Reduce its size:

# Default map tokens varies by model
aider --map-tokens 1024  # Reduce from default

# Or disable the map entirely
aider --map-tokens 0

The repo map helps Aider understand your codebase structure, but it consumes context. For focused edits on specific files, you can reduce or disable it.

Fix 4: Start a fresh session

Long sessions accumulate context from previous messages:

# Clear and start fresh
/clear

# Or exit and restart
exit
aider src/specific-file.ts  # Only add the file you need

Fix 5: Use /undo and retry concisely

If a failed attempt bloated the context:

/undo  # Remove the last change and its context
# Then rephrase your request more concisely

Prevention

• Add only the files you’re editing, not the whole project
• Use specific, concise prompts (not “refactor everything”)
• Start fresh sessions for new tasks
• Use models with larger context for big refactors
• Check /tokens regularly

Related: Aider Complete Guide · Aider Git Error Fix · Ollama Out of Memory Fix · Long-Running AI Agents · Best AI Coding Tools