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
const base = Math.min(30_000 * 2 ** attempt, 300_000); const delay = Math.random() * base; // full jitter
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.
Save. Carry. Continue.
The console checkpoints on its own, or paste any transcript. GiState compresses it into a small, versioned state.
cp-7 · 512 tokensSwitch 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 costStart 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 · recallA 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.
Recall across sessions
Ask for the decision, get the checkpoint it came from.
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.
- DecidedFirst retry after 30 s, doubling to a 5 min cap
- Ruled outA flat 5 s retry: it caused the rate-limit incident
- Must holdRetry only on 429 and 503
- NextSettle jitter, so two workers never retry in lockstep
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
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.
- 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.
Stop starting over.
Bring a ChatGPT or Claude chat you already have, or start a new one, and keep going in any model.