Private betaLimited seats, approved by hand

Great work
continues.

Claude, GPT and Gemini in one place, and none of them starts from scratch. GiState keeps track of what you’ve decided, what you’ve ruled out and what’s next, so you can switch to whichever AI fits the next step, or come back next week, and pick up exactly where you left off.

Top up from $10 · no subscription required · handoffs billed at provider cost

gistate.com/console
Payment retry worker cp-7
RESUMED FROM CP-6 · CLAUDE SONNET 5.5
The payment worker keeps retrying too fast on 429s. Where did we land on the backoff?
CS
We settled on a 30 second first retry, doubling up to a 5 minute cap. Only 429 and 503 are retried. Jitter is still an open question.
Handed off to GPT-6.1 Sol
Add jitter so two workers never retry in lockstep.
G6
Full jitter on top of the schedule you already chose, so the 30 s and 5 min bounds still hold:
const base  = Math.min(30_000 * 2 ** attempt, 300_000);
const delay = Math.random() * base;  // full jitter
cp-73 of 10 turns to the next checkpoint
Reply to GPT-6.1 Sol…
GPT-6.1 Sol
One session, carried acrossClaude Sonnet 5.5GPT-6.1 SolGemini 3.1 Prowithout starting over
WHY NOW

The best AI keeps changing.
Your work keeps going.

The best model for the next step is rarely the one you started with. Today, switching means pasting a wall of context and hoping the new one catches up.

With GiState you just pick another model. It gets a short brief of the work so far: the goal, the decisions and why, and what’s next. Check it if you like, and the new model carries on as if it had been there from the start.

01
Context windows fill up.
Long sessions hit the limit and the model forgets the decisions that got you there.
02
Switching models means starting from zero.
Every provider keeps its own context. Nothing carries between them.
03
Coming back means re-explaining.
A new chat next week knows nothing about this one, so you paste, summarize and correct it all again.
04
Memory remembers preferences, not progress.
Knowing you prefer TypeScript is not knowing where the refactor stands.
HOW IT WORKS

Save. Carry. Continue.

01
Save

The console checkpoints on its own, or paste any transcript. GiState compresses it into a small, versioned state.

cp-7 · 512 tokens
02
Carry

Switch from Claude to GPT to Gemini mid-task. The state goes with you, and you can review it before it does.

Claude Sonnet 5.5 → GPT-6.1 Sol · at cost
03
Continue

Start a new session with “continue the pricing page” and it picks up from the right checkpoint. Or branch one to try an idea, or recall last week’s decision.

continue · branch · recall

A lineage for every session

Every checkpoint is a version. Branch an idea, merge it back, and step back to any point without losing the rest.

branch · try-jittercp-1cp-2cp-3cp-4cp-5cp-6cp-7importGPT-6.1 Sol

Recall across sessions

Ask for the decision, get the checkpoint it came from.

why a 30 second retry?
DECISION · cp-43 days ago

First retry after 30 s, doubling to a 5 minute cap. Chosen after the rate-limit incident.

See what the next model sees

Before another model takes over, read the brief it will get and fix anything that’s off.

BRIEF FOR GPT-6.1 SOL512 tokens
  • Decided
    First retry after 30 s, doubling to a 5 min cap
  • Ruled out
    A flat 5 s retry: it caused the rate-limit incident
  • Must hold
    Retry only on 429 and 503
  • Next
    Settle jitter, so two workers never retry in lockstep
EditSend to GPT-6.1 Sol

Priced to the micro-dollar

Every response reports what it cost. No subscription required and no rate limits on paid usage: your balance is the only limit, and a spend ceiling is there if you want one.

Pro is optional: $20 a month or $200 a year for more sessions, files and synced history, priority support and early access. A call costs the same on either plan. Compare plans

Chat turn · GPT-6.1 Sol$0.0031
Checkpoint cp-7$0.0008
Handoff · Sonnet 5.5 → GPT-6.1 Solat cost
Example figurescharged_usd

Hand it a goal

Work splits a goal into tasks, runs each on the model best at it and combines the results into one answer. You see the most it can cost before anything runs.

WORK · 3 TASKSup to $0.40
  • Research the two competitorsGemini 3.1 Pro
  • Draft the recommendationClaude Opus 5.5
  • Argue against itGPT-6.1 Sol

COMBINED BY CLAUDE OPUS 5.5

Ask several models at once

Compare sends one message to up to four models side by side, each on your saved state. Keep the answer you want and carry on from it.

Claude Opus 5.5

·

GPT-6.1 Sol

KEPT

Gemini 3.1 Pro

·

Get pushback before you commit

Challenge has models from other providers argue against the answer: the strongest objections, what is missing, and what would change their mind.

GEMINI 3.1 PRO · OBJECTION

A 30 second first retry hides a real outage from the dashboard for minutes.

GPT-6.1 SOL · OBJECTION

Nothing caps the queue while the provider is down.

0%
margin on handoffs
$0.000001
billing precision, in integer micros
None
rate limits on paid usage
3
model families: Claude, GPT, Gemini

Stop starting over.

Bring a ChatGPT or Claude chat you already have, or start a new one, and keep going in any model.

GiState: Great work continues