Shared workspaces instead of disposable chats
Threads stay attached to the machine, path, approvals, previews, and execution history that produced them.
Run Codex where the code already lives. Pair the machine, bind the real path, and supervise previews, approvals, and long-running sessions from one browser control plane.
Workspace live
checkout-recovery
Active run
Model
Codex with browser supervision
Tools
Diff, approvals, previews, terminal
Continuity
Delta resume + event replay
Timeline context
server-authoritativeApprovals
Preview
Install/update
Cold buyers need to understand the difference immediately. Eridai is not another hosted AI tab. It is a control plane for running AI work on real machines, with enough structure to review, preview, and reopen that work later.
Threads stay attached to the machine, path, approvals, previews, and execution history that produced them.
The desktop stays on the real Mac, Linux host, or VPS. The browser stays the control plane.
Open explicit localhost previews, approve actions, and inspect events without switching tools.
Desktop replay, presence leases, and browser delta resume keep long-running work recoverable.
The repo path, machine identity, preview exposure, approval history, event stream, and reconnect story are part of the product model. That is why Eridai can feel serious where lighter tools still feel temporary.
Cold-buyer test
A serious buyer should see this page and understand that Eridai is for running real engineering work with stronger context and control, not for generating one more isolated answer.
Proof-bound machine auth, browser cookies with CSRF, and user-scoped actions across workspaces, machines, sessions, approvals, and previews.
Desktop install and update flows are brokered by the server and verified with signed release metadata instead of raw unauthenticated downloads.
Localhost previews are saved exposures, not implicit tunnels. The browser opens an isolated preview origin backed by the existing desktop socket.
Machine choice, path binding, repository inspection, and session continuity stay part of the same working surface rather than spread across tabs.
The public site should show a real path into the product, not vague promises. Eridai starts with install and pairing, then moves into workspace-bound execution, approvals, previews, and reopenable supervision.
Step 01
Generate a signed bootstrap command, install locally by default, and keep the runtime on the machine you already trust.
Step 02
Device pairing is explicit. The desktop starts pairing, the browser approves it, and the server keeps the user scope authoritative.
Step 03
Workspaces and machines stay separate. Bind the path you want, keep standalone chats when needed, and avoid fake migrations.
Step 04
Run the agent, inspect previews, manage approvals, and reopen the same operational context later from the browser.
Eridai keeps the ownership boundary intact. The desktop owns local execution and localhost relay to its own machine. The server owns persisted truth and authorization. The browser stays presentation and supervision only.
Refresh and connect flows are challenge-based. Runtime scope comes from the server database, not client claims.
Preview access is created, opened, and revoked through saved server records rather than implied tunnels.
Deleting or rebinding a machine does not rewrite historical sessions. Old commands keep the context they were created with.
Install and update flows verify signed metadata and checksums before switching the active desktop release.
The public site can issue a short-lived bootstrap command right now. Install the desktop, pair the machine, and continue the workflow in the browser.
Signed bootstrap
Generate a fresh token any time. The command is public-install only; pairing and browser approval still gate the machine into the product.
If the choice is between a cleaner chat surface and a real operational control plane, Eridai should look like the tool for the second job. That is the bar this page is designed to clear.