One allowance for playground, MCP and API.
Your access renews at 00:00 UTC.
Sign in to System One
Opening your workspace…
SYSTEM ONE CLOUD · FREE EARLY ACCESS
Open your workspace
Verify your email to use Jev in your agent, application or playground. No payment card, waitlist or provider key. Free Jev access is available each UTC day, subject to usage limits.
Have a private invitation?
We use your email only to send sign-in codes. Privacy
AN AGENT IS REQUESTING ACCESS
The client name is supplied by the client. Check its client ID and return address before granting access.
- Client ID
- Return address
- Engine
Dashboard
Track requests and available allowance across your connections.
Connect ChatGPT or Claude Code
Connect your agent to run a check and see its activity here.
Up to 1 GiB of input each UTC day
Measured valueNot measuredSavings need complete tasks compared with and without System One. Usage and price scenarios are not savings.
Response times
View activity →No decision requests yet.
Scenario estimate: input tokens only
The same reported input tokens priced at another model's list rate, projected to 1,000 requests. Not recorded spending and not whole-task savings.
Updates every 5 seconds while visible.
Allowance and request limits
reported tokens.
- Request rate
- 60 requests per minute
- Running at once
- Up to 10 requests
- Input per evaluation
- Up to 32 KiB
- Model allowance
- Each Jev attempt counts against free access. Reused results and fixed rules use no model allowance.
Each attempt reserves a full Jev context, then settles against reported input tokens. Failed attempts and attempts with unreported usage keep that reservation.
Per-connection daily limits also apply. There are no automatic paid overages.
Recent requests
| When | Connection | Tool | Client | Caller model | Effort | Time | Evaluations | Outcome |
|---|
Run a check in the playground, through MCP or through the API to see its receipt here.
Try a claim check
Does a successful build mean the release is live? Run the example here, or ask your connected agent to run it.
Running the example uses your shared allowance. Nothing runs until you request it.
How to use the result
Connect once
Add the MCP with a scoped connection. The optional skill tells your agent when a check helps.
Run one real check
Your agent calls sysone_check with a claim and the evidence it observed. No setup calls come first.
Use the answer
Your agent reads the probabilities, keeps uncertainty visible and continues its task.
The tool returns a model estimate, not a verified fact. Keep the evidence and review uncertainty. Use code for exact rules. The optional skill guides your agent; connecting the tool gives it access to evaluations.
Read the agent prompt
Use my connected System One MCP to call sysone_check once with {"claim":"The release is live.","evidence":"The build passed, but deployment has not run."}. This is a setup example, not evidence from our current task. Report the supported and contradicted probabilities, the latency and model calls, and what the evidence establishes. No other call is needed first. If the tool is missing or fails, report the error; read the sysone://status resource only to diagnose it. Do not deploy anything or retry a failed check.
Workspace
Set your workspace name and the services your agents can use.
Add System One to an agent
Choose your agent, then sign in once. You review its requested access in this workspace.
Copy the MCP address
Use this address in any compatible remote MCP client. It contains no credential.
Connections and API keys
Save this key privately. It is shown only now; your provider key stays on the server. Use it with the API or SDK.
| Connection | Services | Status | Control |
|---|
No connections yet. Add the MCP URL to your agent to get started.
Recipe selection
Choose the saved recipes your agents discover. Enabled services can still accept custom questions. Changes apply after you save.
Add the URL below in ChatGPT, Cursor, Claude, or Codex after you save. Approve the connection in this workspace.