Claude Code has
a waste problem.
peddra is the fix.

the whole app, free for 7 days.
no card, no account, nothing leaves your Mac.

try peddra free

after the week: $19.99 a month, or keep the basic meter for free.

how it works

how peddra fixes it

every message you send makes Claude read your whole chat again, from the top.
every agent you spawn gets handed the same instructions again.
here is how peddra ends that.

01
your Mac auditions first.
before a paid token moves, peddra hands the work to Apple Intelligence on silicon you already own. when your Mac can do the job costs nothing.
requires a Mac with Apple Intelligence turned on. without it, the audition starts at the cheapest paid model instead.
02
cheaper models, proven equal.
when the Mac can't, peddra dispatches the smallest model that passes machine checks: never a guess, never a vibe. on an 80-task audit of real work, the cheap tier passed 94.9% of the time.
03
the re-read tax, ended.
a long chat makes Claude re-read everything, every turn, at frontier prices. peddra flags it red, a cheap model does the reading, and your next chat starts from the short version. resuming cold: $5.49. the short version: $0.18.
on a Max plan the real price is your limits: resuming cold pushes about 760K tokens back through them. the short version reads on your Mac, off your limits entirely; even the cloud fallback barely dents them.
04
the setup toll, watched.
every message you send pays a setup charge before your first word: your tools, your instructions, your connectors all ride along, every time. peddra measures yours and tells you the day it jumps.
05
the gone-cold warning.
Claude keeps a working chat cheap for a few minutes at a time. walk away too long and the next message re-buys the whole thing. peddra gives you a quiet heads-up before that happens.
the features

what peddra does

long chats stop being expensive.

Claude re-reads your whole chat on every message, so a long chat quietly becomes your most expensive habit. peddra watches every chat and flags the moment one turns wasteful. then your Mac reads the full history, checks your docs tell the same story, and writes the prompt that starts a fresh chat with nothing lost.

01
flagged red
the chats list, one chat carrying a red status light

your chat gets long. peddra flags it red and shows what every message now costs you.

02
docs confirmed
STATEchat and docs compared, line by line
DECISIONSkept only what still matches
NEXTwrite the fresh-start prompt
confirmed, nothing lost

your Mac reads the whole chat and makes sure your docs tell the same story, so nothing is lost when you move.

03
the ready prompt
a chat's ready prompt with its copy button

peddra hands you the prompt for a fresh chat. copy, paste, keep working, for pennies.

resuming cold: $5.49. the fresh start: $0.18.

your Mac gets asked first.

peddra starts every job on the AI already inside your Mac. Apple Intelligence is more capable than most people give it credit for: we watched it read a million-word chat for the cost of electricity. when your Mac can't do a job, or you don't have Apple Intelligence at all, that work goes to haiku, Claude's lowest-cost tier, and every result is checked, never trusted. you don't need Apple Intelligence to use peddra; having it just makes more of the work free.

the ladder
1 . your Mac . free
2 . haiku . pennies
3 . a bigger model . only when checks demand it
every step checked, never guessed.
this message carries
tools attached
system instructions
active connectors
watched, every message.

the setup you pay for on every message.

before your first word lands, every message hauls your whole setup along: tools, instructions, connectors. you never see it, and you pay it every time. peddra measures your setup toll and tells you the day it jumps, so a new connector never quietly doubles your cost.

never reheat a chat at full price.

Claude keeps your working chat warm for a few minutes at a time. step away too long and your next message re-buys everything. peddra taps you before a working chat goes cold, so a coffee break never costs you the whole conversation again.

peddra
this chat is going cold. your next message after the break re-buys it.
the fix
your long chat becomes a short one

peddra reads the chat on the left and writes the short version on the right, before you need it. your next chat starts from there.

the chat

you: can we get the card layout tightened up before the review

claude: sure, I will drop the outer box shadow and use the bg1/bg2 grounds only, per the spacing rule

you: also check the odometer glow isn't doubling up with the stat block

claude: confirmed, only one glow on screen, the odometer's; the stat block below stays matte

you: good, ship it and move to the clock alignment

claude: clock is right-aligned now, matches the statement's convention

you: the loop detector keeps flagging single-image batches as loops, can you check the threshold

claude: found it, the count comparison used >= instead of >, fixed and re-checked

you: that should stop the false positives, run it against last week's batch

claude: ran it, zero false positives on the sample set, three true loops caught

you: nice, add a test for the boundary case at exactly five images

claude: added, and confirmed it stays green at four and flags at five

you: the preview endpoint is still reading from the old articles table after the migration

claude: right, it never got repointed. want me to cut it over now or keep the dual read during rollout

you: keep dual read for now, editors are still mid-review on the old drafts

claude: understood, I will leave the fallback in place and just fix the slug lookup

you: does the redirect map still hold for the old slugs

claude: yes, checked against the sitemap, all 214 old slugs resolve to their v2 equivalents

you: good, once editors confirm previews match production we can retire the old tables

claude: noted in NEXT.md as the next session's first task

you: can you pull the current pricing table before my call in ten minutes

claude: here it is: three tiers, free, pro, and team; team adds seat-based billing

you: does team still show the old per-seat number

claude: yes, still $18 per seat, matches what is in the pricing page

you: ok that is what I needed, thanks

claude: good luck on the call

760K tokens
the short version
## STATE
The content model migration is halfway done: articles and authors have moved to the new schema, but the preview endpoint still reads from the old tables and returns stale drafts.

## DECISIONS
Kept the old preview endpoint live during the migration instead of cutting over at once, so editors could keep working; the new schema's slug field is now the single source of truth for routing.

## NEXT
Point the preview endpoint at the new schema, run the redirect map against the sitemap, then retire the old tables once editors confirm previews match production.

## FILES TOUCHED
src/cms/schema.ts, src/cms/preview.ts, src/cms/migrations/0007_articles.sql, NEXT.md
about 1K tokens
760K tokens becomes about 1K.
reopening the old chat: $5.49 . making the short version: $0.18 at worst, free when your Mac does it.
the biggest read we have measured: a 1,130,000-word chat, chopped into 892 pieces, read end to end by a Mac's own model in 5.8 hours. tokens spent: 0. cost: electricity.

measured on the founder's ten most expensive idle sessions, stated vs resuming each cold.

the app does this for you, before you ask.
the app
what it shows you
the peddra app: the live sidebar, chat list, an open chat's transcript, and its ready prompt
1
2
3
4
5
the live meteryour limits, from Anthropic's own numbers
every chat a lightgreen, yellow, red
the ready promptwritten before you need it
every agent and its modellive
the ledgerwhat peddra kept, counted once
  1. 1the live meter
    your limits, from Anthropic's own numbers
  2. 2every chat a light
    green, yellow, red
  3. 3the ready prompt
    written before you need it
  4. 4every agent and its model
    live
  5. 5the ledger
    what peddra kept, counted once
true user data
one account, fully metered
67.0%
needed
33.0%
avoidable
every dollar counted once: what had to be spent, and what did not.
$17,460 consumed at list rates
50,430 messages
63 sessions . 15 projects
$2,730 the long-chat leak
178K free Mac AI runs a week
one real account, metered by the product, valued at API list rates (what the tokens would cost if bought directly). cash paid was a subscription. that price gap is the point.

these are one real user's measured numbers. find yours in the first two minutes of your free week.

the honest numbers
tested against real sessions, not a pitch
the short-version test
20/20
complete and usable short versions
$0.18
haiku short version
$1.13
sonnet short version
against a $5.49 mean cost to resume the same chat cold.
measured on the founder's ten most expensive idle sessions, stated vs resuming each cold.
the rerun test: 80 real tasks, redone on a cheaper model
94.9%
machine checks passed
64.5%
of audited spend saved
on the founder's own 80-task audit.
one set of real chats, independently verified, estimates labeled. this does not generalize beyond this sample.
the honesty is the marketing: both experiments are one founder's own data, verified by independent reimplementation, never generalized beyond the sample they measured.

your Mac works free. Claude doesn't. peddra keeps score.

every number on this page computes on your own Mac, from your own transcripts. nothing is uploaded, nothing is copied, and the meter stays free forever.

the price
what it costs
the free meter tells you what's wrong.
the full app fixes it while you work.
seven days to decide, free.
the menu bar meter
what you keep if you don't pay. free forever.
your limits, from Anthropic's own numbers
today's spend, by model
every agent and its model, live
no account, no data leaves your Mac

the router, which auditions every delegated task the same way, comes next.

peddra

seven days of everything.
no card, no account.
decide with your own numbers.

try peddra free

after the week: $19.99 a month, or keep the basic meter for free.

Apple silicon Macs, macOS 14 or later
works with or without Apple Intelligence.
already sure? buy pro directly