pi-openai-tweaks

A simple Pi extension that gives you:

• clear OpenAI/Codex usage in your status line (/usage)
• one-command Fast mode control (/fast)
• optional preventive compaction for long agent loops (/autocompact)

Install

From npm:

pi install npm:@m4riok/pi-openai-tweaks

What this extension provides

1) Usage status

Shows 7d usage with reset countdowns in the footer/status line. The 5h window is hidden by default.

You can control:

• format (default or compact)
• show/hide 5h usage
• show/hide weekly usage
• percentage mode (remaining or used)
• refresh mode (turn updates or background polling)

Use:

/usage

This opens a settings-style menu where changes are saved automatically.

Quick commands also work:

/usage refresh
/usage format default
/usage format compact
/usage 5h on
/usage 5h off
/usage weekly on
/usage weekly off
/usage percent remaining
/usage percent used
/usage update turn
/usage update poll

2) Continuous-loop auto-compaction

Use /autocompact to configure preventive context compaction during long-running agent loops.

Defaults:

• disabled
• trigger at 90% of Pi's active-model context-window limit

The trigger can be set from 50% through 99% in 1% steps. Percentages use Pi's contextWindow metadata for the active model, so the token threshold updates automatically after a model switch. When enabled, the extension checks usage after each model-response/tool batch. At the threshold it interrupts the active loop, compacts context, and starts a continuation turn when unfinished tool-driven work remains.

Settings persist under piOpenaiTweaks.autoCompact in ~/.pi/agent/settings.json (or the agent directory selected by PI_CODING_AGENT_DIR). Usage preferences share the piOpenaiTweaks section.

Settings updates use Pi's shared lock and atomic file replacement, preserving unrelated settings and changes from other sessions. Invalid or unreadable settings are reported instead of reset. Failed saves leave the last successfully applied configuration active.

/autocompact
/autocompact on
/autocompact off
/autocompact status
/autocompact threshold
/autocompact threshold 90

3) Fast mode

Use /fast to control OpenAI Codex Fast mode.

When active and eligible, requests include:

• service_tier: "priority"

Supported commands:

/fast
/fast on
/fast off
/fast auto
/fast status

Notes:

• Fast mode is blocked on unsupported model/provider/auth combinations.
• Session starts in auto mode, which follows openai-fast.json defaults if present.

Fast config file locations:

• Global: ~/.pi/agent/extensions/openai-fast.json
• Project: .pi/openai-fast.json (overrides global)

Config contents:

{
  "enabled": false
}

• enabled: default Fast-mode state used when /fast auto is active.