pi-deepseek-peak

Pi extension that shows DeepSeek peak hours in the TUI status bar.

DeepSeek API pricing varies by time of day. Peak hours are more expensive. This extension adds a persistent status indicator so you know when you are in a peak window.

Features

• Status bar indicator: "DS Peak" or "DS Normal"
• Follows DeepSeek's billing windows exactly: peak Mon-Fri 01:00-04:00 and 06:00-10:00 UTC; weekends (Sat/Sun Beijing time) are off-peak all day
• Countdown to the next price change next to the dot (on by default, toggle via /dsp-countdown): "DS Peak for 1h30m", "DS Normal for 27h"
• Auto-refresh at a configurable interval (30s / 1m / 5m, default 5m) via /dsp-refresh
• Correct DeepSeek V4 session pricing: each assistant reply's stored cost is patched at write time with the rate in effect when it completed (peak or off-peak), and /dsp-cost recomputes the whole session (peak/off-peak split, per model: flash, flash-vision-exp and pro)
• Uses theme colors (warning for peak, success for normal)
• Thats it

Install

pi install npm:pi-deepseek-peak

Restart pi or run /reload to activate.

Usage

Once installed, the status bar shows the current DeepSeek pricing period.

Real session cost

Every DeepSeek V4 assistant reply is priced at the rate in effect when the request completed (peak/off-peak), and the stored cost is corrected at write time — so pi's footer and /session cost totals already reflect the real price. To see the full breakdown:

/dsp-cost

It recomputes the whole session from the stored messages (model and timestamp per message), so it is correct even for sessions resumed from disk. Output:

• Current-period rates for both models (DS Peak: flash in $0.44/M (hit $0.014) out $1.32/M · pro in $1.32/M (hit $0.044) out $3.96/M)
• Total $X (peak $A / off-peak $B)
• Per-model token and cost lines
• A note when any message had a model outside the rate table (e.g. legacy deepseek-chat): those keep pi's bundled cost

DeepSeek's official rates (https://api-docs.deepseek.com/quick_start/pricing/): peak $0.44 input / $0.014 cache hit / $1.32 output per 1M tokens for flash and flash-vision-exp, $1.32 / $0.044 / $3.96 for pro; off-peak is exactly half.

Note: /session and the footer display the stored costs — correct for messages created after this extension was activated, but messages from before (or from a resumed session) keep pi's old flat-rate cost there. /dsp-cost is the only view that recomputes everything.

Countdown to the next price change

Show how long until the green/red dot flips (e.g. red peak until 04:00 UTC). The menu offers on/off:

/dsp-countdown    # pick on or off from a menu

enable: "🔴 DS Peak for 1h30m" disable: "🔴 DS Peak"

The remaining time is colored with the current state (red while peak, green while off-peak) and refreshes with the same status check. The setting is saved to ~/.pi/agent/deepseek-peak.json.

Auto-refresh interval

How often the status indicator refreshes. No value to type — a menu offers the three choices:

/dsp-refresh    # pick 30s, 1m or 5m from a menu

The current interval is shown in the picker title; pick another anytime to switch. The choice is saved to ~/.pi/agent/deepseek-peak.json (refresh, in seconds, default 300) and applies immediately.

The config file ~/.pi/agent/deepseek-peak.json stores { "countdown": true, "refresh": 300 } and persists across restarts.

How it works

On session start, the extension computes the current DeepSeek pricing period and displays either "DS Peak" or "DS Normal" in the pi status bar. A timer refreshes the indicator at the configured interval (default every 5 minutes, adjustable via /dsp-refresh).

Billing rule (https://api-docs.deepseek.com/quick_start/pricing/): peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday. All other hours are off-peak, and weekends (Saturdays and Sundays, Beijing time) are off-peak all day. Since the windows are UTC-based, the indicator is the same for every timezone.

When the countdown option is on, the time until the next status change is shown next to the dot, e.g. "🟢 DS Normal for 27h" on a weekend day (next peak starts Monday 01:00 UTC).

The same peak/off-peak rule drives pricing: the extension patches each DeepSeek V4 assistant message's stored cost at message_end with the rate in effect at completion, and /dsp-cost recomputes the full session history.

Testing an unpublished branch

To test a local branch (e.g. feat/[branch_name]) before publishing:

# from the repo on your host machine
git checkout feat/[branch_name]

With pi installed locally

pi install /absolute/path/to/pi-deepseek-peak

This replaces the published npm version (no need to uninstall it first). Then /reload (or restart pi). To switch back to the released version:

pi install npm:pi-deepseek-peak

and /reload again.

When pi runs inside a Docker container

Copy the branch into the running container, then install from there:

docker cp . <container-name>:/tmp/pi-deepseek-peak

Run the following inside the container's pi:

pi install /tmp/pi-deepseek-peak

If a previously installed copy of this extension is still active (e.g. an old npm/git source), remove it :

pi list          # see the installed sources
pi remove git:github.com/psychobarge/pi-deepseek-peak

Then /reload to pick up the change. To get back to the released version and set it as the only source:

pi install npm:pi-deepseek-peak

and /reload. Iterating often? Mount the repo as a Docker volume (-v ~/Workspace/Perso/pi-deepseek-peak:/app/pi-deepseek-peak ...) and reinstall after each git checkout on the host.

Development

Run the self-check for the peak/countdown/pricing logic:

npm test

Requires Node ≥ 22.18 (built-in TypeScript type stripping; no build step).

Changelog

See CHANGELOG.md for version history.

License

MIT