Terminal views¶
Views that draw in the terminal. They read the same data as the commands and follow the same rules: dollars first, ✓ Saved and ≈ Estimated kept apart, no raw token counts. Every example below is the shipped output on synthetic data.
Display settings¶
The six commands print plain text. The views use colour on a terminal; --theme picks light or dark and NO_COLOR turns it off.
- Width follows the terminal, 40 to 140 columns; lines wrap or drop their least important column, never overflow.
NO_COLORturns colour off. Colour is also off when output is not a terminal, so a command run in chat stays plain. Every view reads the same without colour.--ascii(orCG_ASCII=1, or a non-UTF-8 locale) swaps every glyph for ASCII.--theme light|dark(orCG_THEME) picks the heat ramp; other colours are the terminal's own 16.- A proxy that does not answer within 4 seconds is not waited for: the view shows its last picture and how old it is.
/context-guru:watch¶
Question: Which cache is about to cost me, and what did I save today?
$43.95 saved today. 2 of 3 caches are cold: their next message costs about
$0.65 extra.
Caches time left before the next message costs extra
● session-A/main opus-5 █████░░░░░░░ 2:14 ctx 34%
○ session-A/sub-1 sonnet-5 ░░░░░░░░░░░░ COLD ctx 9%
○ session-B/main opus-5 ░░░░░░░░░░░░ COLD ctx 58%
Saved today $43.95 █▅█▁▆▁▄▂▁▆█▅▄▁▁▅
Last 20 requests oldest to newest
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■↻✖
■ hit ▨ partial ✖ miss ↻ ping
Do: the next message costs extra; /compact first to cut it.
updated 22s ago q quit r refresh
One row per live thread, the main conversation and each subagent. The bar drains to expiry; COLD means the next message pays to re-send the conversation. ↻ marks a keep-alive ping. The last-20 strip shows ■ hit, ▨ partial, ✖ miss. The proxy does not report when the next ping fires, so none is shown.
Live view: run python3 <plugin>/scripts/tv.py watch in a second terminal. Keys: q quit, r refresh. It redraws every second and fetches in the background, so a slow proxy never freezes it. In a chat a skill prints one frame and that command.
/context-guru:timeline¶
Question: What happened in this session, and which moments were expensive?
Cache misses cost about $2.22 of $12.58 billed in this session (3 of 126
messages).
$0.76 $1.20
▮▮▮▮▮▮▮▪▮▮▮▮▪▮▮▮▪▮▮▮▮───✖▮▪▮▪▮▮▮▮▮▮▪▮▮▮▮▮▪▮──↻───✖▮▮▮▮▪▮▮▮▮▪▼▮▮▮▪▮▪▮▪─────↻──✖
12:31 13:15 13:59 14:43 15:27
▪ message ▮ busy ✖ cache miss, extra cost above it ▼ compacted
↻ keep-alive ping ─ pause over 5 min
Most expensive moments
14:22 $1.20 after a pause of 8m
13:27 $0.76 after a pause of 9m
15:27 $0.25 after a pause of 8m
Do: run /compact before a break longer than 5 minutes, so the extra cost is
smaller.
Extra = cost above a normal turn of the same model (estimate).
Misses are read from the token columns. The dollar figure above each ✖ is its extra cost above a normal turn of the same model (an estimate). Pauses over 5 minutes are drawn as a line.
/context-guru:chart [14|30]¶
Question: How much did I save each day, from which lever, and what is still avoidable?
![/context-guru:chart [14|30]](chart-dark.svg#only-dark)
![/context-guru:chart [14|30]](chart-light.svg#only-light)
Saved $43.95 in 14 days (4.9% of the bill). Best day Tue $4.83.
✓ Saved per day measured, net of Context Guru's own spend
$4.83 ██ ▃▃ ▇▇ ▄▄
▃▃ ▂▂ ██ ██ ██ ██
██ ██ ██ ██ ██ ██ ▅▅
$2.42 ██ ██ ██ ██ ██ ▇▇ ██ ██ ██
██ ██ ██ ██ ██ ██ ██ ▁▁ ██ ██
██ ██ ██ ██ ██ ██ ▆▆ ██ ██ ██ ██
F S S M T W T F S S M T W T
Sep 25 Oct 08
By lever measured, same window; adds up to the total
Compaction $33.90 ██████████████
Keep-alive pings $6.30 ███░░░░░░░░░░░
Tool filtering $7.85 ███░░░░░░░░░░░
Context Guru's own model spend -$4.10
Total $43.95
≈ Estimated spend you could still avoid; rows overlap, do not add
You paused longer than the cache lasts, 214 times $108 ▒▒▒▒▒▒▒▒▒▒▒▒▒▒
Settings or tools changed mid-session $23.31 ▒▒▒░░░░░░░░░░░
Do: run /compact before a long break.
Days are UTC days, as in the dashboard. Rows under By lever add up to the total. ≈ Estimated is drawn hatched and is not part of the saved total.
/context-guru:heatmap [days]¶
Question: When do my pauses cost me?
![/context-guru:heatmap [days]](heatmap-dark.svg#only-dark)
![/context-guru:heatmap [days]](heatmap-light.svg#only-light)
Cache misses cost you about $21.89 in the last 28 days. Costliest hour: Thu
13:00 ($2.00).
0 3 6 9 12 15 18 21
Mon ░░ ░░ ▓▓ ░░ $2.73
Tue ░░ ░░ ▓▓░░ ░░ $2.73
Wed ░░ ░░ ▓▓░░ ░░ $2.73
Thu ░░ ░░░░████░░░░ $5.54
Fri ░░ ░░ ▓▓ ░░ $2.73
Sat ░░ ░░ ▓▓ ░░ $2.73
Sun ░░ ░░ ▓▓ ░░ $2.73
Scale ░░▒▒▓▓██ $0 to $2.00 per hour; blank = no miss
Do: before Thu 13:00, run /compact if you will step away.
Hours are local time (UTC+00:00). 88 misses; extra cost is an estimate.
Weekday by hour in local time. Covers up to the newest 6,000 requests and says so when it is partial. The first request of a thread is not counted: it always writes.
/context-guru:fill¶
Question: How full is the context, what is in it, and when is it full?
Context is 34% full; at this pace it is full in 2h12m. A cache miss now costs
$0.57.
▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒███░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 34%
▒ Always loaded (instructions and tools) 30%
█ Conversation so far 4%
░ Free 66%
Do: nothing, there is room.
The split is the first request of the main thread (instructions and tools) against everything added since. The proxy does not record how much of the rest is tool output.