@cpzombie/pi-settings

pi coding agent extensions for personal setup convenience.

Extensions

raw-context-counter

Replaces the built-in TUI footer with a version where the context counter shows raw token counts (e.g. 25k/200k (auto)) instead of a percentage (e.g. 12.5%/200k (auto)) — the absolute count is more useful than a percentage, and rounding follows the built-in footer's token formatting (24,600 tokens renders as 25k).

Everything else mirrors the built-in footer: pwd + git branch + session name line, token stats (cache/cost omitted — local-only), color thresholds (>90% red, >70% yellow), model name + thinking level on the right, and the extension status line. It is a line-for-line mirror of the built-in footer internals, so a pi release that changes the footer may need a sync pass.

setup-sync

Seeds the personal config bundled under setup/ into the agent dir (~/.pi/agent) on session start:

• setup/models.json → custom providers (local llama-cpp server on localhost:8080 with Qwen3.8-27B-Q8)
• setup/settings.json → compaction defaults (enabled, keepRecentTokens: 60000)

Merge rules (never clobber): bundled providers are added only if that provider isn't already defined in models.json; for settings.json, only missing leaf keys are filled in. If your files already exist with your own values, they win. New values take effect on the next pi start.

LLM server address. The bundled provider's baseUrl is machine-specific (default http://localhost:8080/v1). When the provider is first seeded on a machine:

1. PI_LLM_BASE_URL=http://gpu-box:8080/v1 env var wins (non-interactive runs, print mode, CI)
2. otherwise the TUI asks once at startup — Enter the address, or Esc to keep the default
3. a non-interactive first run with no env var gets the bundled default (http://localhost:8080/v1)
4. afterwards, use the /llm-setup command to set a new address, force-update the provider from the bundled config, or remove it entirely

If the provider is already defined in your models.json, nothing is asked and nothing is changed.

Edit the JSON files under setup/ in this repo and bump the version to change what gets seeded — or just edit your local files directly to override. Because of the never-clobber merge rules, updated bundled values do not propagate to machines that already have the provider seeded: use /llm-setup → “Force-update from bundled config” to replace the provider entry with the bundled definition (your baseUrl is preserved).

Note: the bundled model config points at a local llama-cpp server; on a machine without one the model will appear in /model but not connect.

Install

pi install npm:@cpzombie/pi-settings
# or pinned to a version:
pi install npm:@cpzombie/[email protected]

Uninstall (or pi remove then /reload in an interactive session):

pi remove npm:@cpzombie/pi-settings

Note: removing the package does not undo the settings that were already seeded into ~/.pi/agent/ — delete the provider from models.json and the compaction block from settings.json manually if you want that.

Notes

• Extensions are TypeScript, loaded by pi via jiti — no build step.
• @earendil-works/pi-coding-agent, -pi-ai, and -pi-tui are peer dependencies provided by your pi installation; they are not bundled here.