All docs

Aider

Aider is an AI pair programming tool that works in the terminal. Squeezr supports both Aider with Anthropic models and Aider with OpenAI models.

Setup

Run squeezr setup first. It automatically sets ANTHROPIC_BASE_URL=http://localhost:8080.

Aider with Anthropic models (Claude)

ANTHROPIC_BASE_URL is already set by squeezr setup. Just run Aider as normal:

aider --model claude-sonnet-4-5

Aider with OpenAI models

Set openai_base_url in your .aider.conf.yml:

# .aider.conf.yml
openai_base_url: http://localhost:8080

Or via environment variable:

OPENAI_BASE_URL=http://localhost:8080 aider --model gpt-4o

What gets compressed

Aider sends file contents, git diffs, and command outputs as part of its prompts. Squeezr compresses:

  • Repository map — Aider's repo map can be very large. Squeezr deduplicates it across turns and summarizes the structure for large repos.
  • File contents — Files added to the chat are compressed with cross-turn dedup (files read again in later turns are replaced with references).
  • Git diffs — Diff output is compressed using the git-specific pattern matcher (1-line context, cap on log output).
  • Command output — Test results, lint output, and build logs are pattern-matched and compressed.

Project-level configuration

Create a .squeezr.toml alongside your .aider.conf.yml to tune compression for that project:

# .squeezr.toml (in project root)
[compression]
threshold = 600
compress_conversation = true

Tips

  • Aider sessions can be very long. Squeezr's cross-turn dedup is especially valuable here — the same files are often re-sent many times.
  • Use squeezr status to verify the proxy is running before starting an Aider session.
  • If you use /add to add many large files, Squeezr will compress them more aggressively once the context window starts filling up.