Squeezr — AI Context Compression

Run squeezr setup once after installation. It configures everything automatically — env vars, shell wrapper, auto-start, and TLS certificates. You don't need to edit shell profiles or set environment variables manually.

squeezr setup
squeezr start

That's all that is required for Claude Code, Aider, Gemini CLI, and Ollama. Codex requires one additional per-session step described below.

What setup does

Environment variables

Sets the following vars in your user environment (Windows registry / macOS/Linux ~/.bashrc or ~/.zshrc):

ANTHROPIC_BASE_URL=http://localhost:8080
GEMINI_API_BASE_URL=http://localhost:8080

Your existing API keys are not touched. Squeezr forwards them to the upstream API automatically.

Shell wrapper

Because child processes cannot modify the parent shell's environment, setup installs a persistent wrapper function so env vars refresh in the current terminal without restarting it:

Windows: function added to PowerShell $PROFILE
Linux / macOS / WSL: function added to ~/.bashrc or ~/.zshrc

After setup runs, open a new terminal once (or source your profile) and the wrapper will be active in all future sessions.

Auto-start

Registers Squeezr to start automatically after a reboot:

Windows: Task Scheduler or NSSM
Linux: systemd user service
macOS: launchd agent

TLS certificates (for Codex MITM)

Windows: imports the MITM CA into the Windows Certificate Store at user level (no admin required)
macOS/Linux/WSL: generates a CA bundle at ~/.squeezr/mitm-ca/bundle.crt and sets NODE_EXTRA_CA_CERTS

Tool-specific notes

Claude Code

Works automatically after setup. Claude Code reads ANTHROPIC_BASE_URL and routes all API calls through the proxy. No further configuration needed.

Aider

Set ANTHROPIC_BASE_URL (already done by setup) for Anthropic models, or configure openai_base_url in your .aider.conf.yml for OpenAI models:

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

Gemini CLI

Works automatically after setup. Gemini CLI reads GEMINI_API_BASE_URL.

Ollama

Squeezr detects Ollama automatically when it sees a dummy API key (e.g. ollama) or a local upstream URL. No extra configuration needed if you are using Ollama with a tool that already targets the proxy.

To use a local model for compression itself, configure in squeezr.toml:

[local]
enabled = true
upstream_url = "http://localhost:11434"
compression_model = "qwen2.5-coder:1.5b"

Codex

Codex uses WebSocket over TLS to chatgpt.com and cannot be proxied via a simple base URL override. Squeezr runs a TLS-terminating MITM proxy on port 8081.

Set HTTPS_PROXY only in the terminal session where you run Codex— do not set it globally as it will break other tools:

# Run this in the terminal where you launch Codex — not globally
HTTPS_PROXY=http://localhost:8081 codex

See the Codex guide for the full technical breakdown.

Changing ports

To change the HTTP proxy port (default 8080) or the MITM proxy port (default 8081):

squeezr ports

Or edit squeezr.toml directly:

[proxy]
port = 9090
mitm_port = 9091

Verify the connection

squeezr status

Setup