Cut the bill.Keep the performance.
StateSync runs the coding agents you already pay for and gets the same work done for a lot less. Same agent, same model, same quality. Nothing to set up.
Windows, macOS and Linux. 14 days free for the first 50 places.
+86%
more work on the same subscription
Measured onTokenBenchWe measured this with the TokenBench harness, which is built to track real coding work as closely as we can measure it. The figure is a pooled average across short and long tasks alike: the saving is smallest on the shortest tasks, and the harness's tasks are shorter than much real coding work, so it is a conservative one. Your own experience may differ.
Claude Code
Codex
OpenCode
Cursor CLI
Copilot
Pi
Auggie
Factory Droid
Grok
Measured, not promised.
55 real tasks, each done with StateSync and without. Same model, same repositories, quality checked on every one.
See the evidenceTokenBench harnessTotals across all 55 tasks, not an average of each task's own saving, so a large task counts for what it actually cost. Both arms used Claude Sonnet 4.6. StateSync ran each task once, and the plain agent's runs of the same tasks are from 22 to 24 August 2026. The set is the 55 of TokenBench's 73 tasks where the plain agent made more than 8 model requests, and /evidence says why.
Why your agent costs so much.
With coding agents, early work stacks up and you pay for it again on every reply. StateSync keeps that pile small, so the same task costs less.
Nothing to set up
Install StateSync, open a folder, start working. No config files, no plugins, nothing to wire up.
- Without StateSync
- 342,090 tokens, 10 requests
- With StateSync
- 254,166 tokens, 8 requests
See every taskOne real task from TokenBench's earlier August run, drawn from the same 55 tasks: a bug fix in a Python service, of median length, and the closest of those to that run's own saving. By its last request the plain agent was sending 39,373 tokens to have 185 new ones read. Neither agent handed work to a subagent on it. A request is one time your agent sends the conversation to the model; a single instruction usually takes several.
Watch the savings in real time.
Tasks from start to finish, with and without StateSync. Single runs rather than averages. The pooled figures are on the evidence page.
Both agents were given the same prompt
One data table
The task and user tables moved onto one shared table.
Claude Sonnet 5 on both sides, the same starting code and the same settings.
- 1Move the Tasks and Users tables onto one shared, typed data-table module.
- 2Change no behaviour: search, filters, sorting, bulk delete and row actions all keep working.
- 3Add column visibility per table that survives a reload, with a Reset columns action.
- 4Keep the existing tests green and add tests for the new module.
One data tableThe task and user tables moved onto one shared table.
Bring your own agent.
Your subscriptions, your keys, your models. StateSync launches the agents you already pay for and stays out of the way.
- Claude Code
- Codex
- OpenCode
- Cursor CLI
- Copilot
- Pi
- Auggie
- Factory Droid
- Grok
- Switch mid-conversation
- Hit a weekly limit, or want a different agent for the next step? StateSync moves the conversation to another agent, or to another of your Claude Code or Codex accounts, with the same project and history and nothing repeated. Work that matters does not have to wait for a limit to reset.
- Installed in one click
- StateSync works with the agents already on your machine, and the ones that are not are one click away. Minimal setup either way.
Everything is local.
Code, chats, prompts, voice. Everything stays on your computer and works exactly as it did before.
What StateSync collects, in fullStateSync never receives your code, your prompts, your chats, your commands or your voice. What reaches us is counts: how much was used, which models ran, and how often StateSync ran. The privacy policy lists every field.
Voice built in.
Hold a key and talk. StateSync cleans up what you said and drops it straight into the prompt. The recording never leaves your computer.
- What you said
- “um so can you uh fix the flaky test in the auth module and then run the suite again”
- What StateSync wrote
Can you fix the flaky test in the auth module and then run the suite again?
Save more than you spend.
Pick the plan that matches what you already spend on coding agents.
Every plan has every feature. Plans differ in usage each week and in the number of signed-in devices.
Early access pricing
- First 50 places$8now
- Next 50 places$10
- After that$12
Prices shown for the Core plan. Every plan steps by the same proportion.
Enough for one $20 a month agent plan.
Early access price. Rises to $10 after the first 50 places and $12 after 100.
6 times Core. Enough for a $100 a month plan.
Early access price. Rises to $25 after the first 50 places and $30 after 100.
25 times Core. Enough for a $200 top tier plan.
Early access price. Rises to $62.50 after the first 50 places and $75 after 100.
Enough for several top tier plans at once.
Early access price. Rises to $125 after the first 50 places and $150 after 100.
Every plan starts with 14 days free, on that plan's own weekly allowance. A card is needed to start, but nothing is charged during the trial. Cancel before the 14 days are up and you are not billed.
Otherwise the plan you picked starts at the price shown above, and you can cancel online any time.
Questions
StateSync is a desktop app that runs the coding agents you already pay for and gets the same work done for a lot less. Every agent sits in one place, there is nothing to set up, and the saving is measured rather than claimed: across 55 real tasks, the same work cost 46 percent less.
Claude Code, Codex, OpenCode, Cursor CLI, Copilot, Pi, Auggie, Factory Droid and Grok. You keep your own subscription or API key for each one. StateSync runs these agents rather than replacing them, and you can switch from one to another in the middle of a task without losing the conversation.
In our TokenBench run on 55 real tasks across 8 repositories, StateSync did the same work for 46 percent less and 48 percent faster. That is 86 percent more work for the same money. Your own numbers will depend on your model and the work you give it.
No, not on average. The easy way to spend less is to hide things from the agent, and then the work gets worse. Every task in the measured run was scored, and StateSync solved slightly more of them than the agent did on its own.
Different jobs. Compaction is for the end of a task: it summarises the chat so the next task starts clean, and mid-task it throws away detail the agent still needs. StateSync works inside the task, trimming each result as it arrives so every later request is smaller. A bigger context window raises the ceiling and the price per request, and sends the same material either way.
No. StateSync never receives your code, your prompts, your chats, your commands or your voice. Your agent connects to its provider exactly as it did before, on your own account. StateSync does not change what that provider does with your data: the provider's own terms and privacy policy apply just as they do when you use it directly, including any use of your prompts and code to train its models, and StateSync has no visibility into that. What StateSync itself receives is counts: how much was used, which models ran, and a few facts about your installation. The privacy page lists every field.
Pick the plan that matches what you already spend on coding agents. Plans run from $8 to $100 a month. The trial is 14 days free for the first 50 places, 7 days after that, on whichever plan you pick, and a card is needed to start. Cancel before the trial ends and you are not billed. Otherwise the plan you picked starts at the price shown.
Not yet, and we have not measured every agent. The savings on this site were measured mainly with Claude Code and Anthropic's models. Codex gets StateSync's tools too, but savings there may be slightly smaller depending on how you use it, because of how Codex handles tools from MCP servers rather than its own. The built-in tools in OpenCode, Factory Droid and GitHub Copilot are not as refined as Claude Code's, so StateSync's tools can save as much there or potentially more, though we have not published measurements for them yet.
You get more work out of the same plan. Each task uses fewer tokens and fewer requests, so you go further before you hit a session or weekly limit.
No. Same provider, same account, same model. There is no StateSync server between your machine and your provider.
No. Signing in is a one-time code sent to your email (or Google or GitHub). There is no password to create, remember, leak or reuse. This is intentional, not a missing feature.
No. Keys stay on your machine: they are kept in your operating system's secure store (or in your agent's own config file) and passed to your agent when it runs. StateSync has no key proxy: there is no StateSync server in the path between your machine and your provider's API.
Nothing. Install StateSync and open a folder. The agents already on your machine are found for you. No config files, no plugins, no per-project settings. There are advanced settings if you want them, for things like which voice model to download.
StateSync is built not to. When StateSync is not sure what a result needs, the agent gets the whole result. If something does go wrong, please tell us at support@statesync.net, or use Support in the app's settings.
Voice dictation built into StateSync. Press a key and talk, or hold it if you prefer, and clean text lands in the prompt. StateVoice runs entirely on your machine, so the recording never leaves your computer.
Windows, macOS and Linux. StateSync is a native desktop app, and the quickstart page shows which builds are out right now.
TokenBench 55-task set. StateSync was run on 11 September 2026 against the plain agent's runs of the same tasks from 22 to 24 August 2026. 55 tasks across 8 repositories and 6 categories, Claude Sonnet 4.6, one StateSync run per task against the plain agent's 4 to 14 runs, averaged. We ran TokenBench ourselves and no independent run has been published. Every task and the harness are public on the TokenBench GitHub, and you can run the same tasks against any tool yourself.
Start saving on your next task.
The trial is 14 days free for the first 50 places, 7 days after that, on the plan you pick. Cancel before it ends and you are not billed.
Downloads could not be loaded just now. Try again in a moment.