01
Connect Claude + your repo
morning CLI on your machine. Your subscription. No model we ship.
Stop trusting “done.”
Make a different process prove it.
You already run Claude overnight. SessionTwin freezes what “done” means before the work starts — then a process the agent cannot reach signs the verdict on your machine, with your subscription.
Your Claude, your machine
Break the ledger yourself
CODE-READY — D1 prove still open
Real login UI · floating cards are labeled illustrations · D1 prove still open
Uses the tools you already pay for — we ship no model of our own
- Claude Code
- morning CLI
- GitHub
- Your machine
Cursor stays Blocked until LIVE-PROVEN. Codex and Gemini are ungoverned — not hireable.
Live product · sessiontwin.ai/login
One place for what needs you. Everything else runs.
Hire AI employees. They drive Claude on your repo. You only show up for the exceptions — approvals, conflicts, first-time calls. The morning is a brief with evidence, not twenty terminals to audit.
One approval surface
Employees drive Claude
Empty means Not staffed
Company OS UI · labeled fiction until your workspace fills it
Overnight on your machine. Wake up to evidence.
Connect morning. Use the Claude subscription you already have. We record the chain — we never become the model vendor, and we never train on your answers.
Your Claude sub
morning CLI
Evidence, not chat
Ledger · live at sessiontwin.ai/web/ledger
Don’t take our word. Attack the log.
Every step is signed into a tamper-evident ledger. Try to edit, delete, or re-sign an entry. Your browser catches it — not our server. If we can’t prove it here, we don’t deserve the night.
Frozen before work
Agent can’t grade itself
You check the math
Three steps
From “I hope it worked” to a signed morning
02
Freeze what “done” means
Write the checks first. Hire an employee onto Claude. Then leave the bar alone.
03
Open Needs You
A separate process signs the verdict. You only wake for what has no rule yet.
Built for people who got burned by a confident lie.
The worker never becomes the governor. If we can’t prove a claim, we say so.
Your machine
Code stays on your disk
Runs with your Claude subscription. Credentials stay local. We keep the evidence trail — not your keys.
Verifier
Homework the author can’t grade
The process that wrote the code has no tool that calls the verifier. Dirty workspaces can’t fake a pass.
Twin
It copies your judgment — not a model
Re-reads answers you already gave. Delete one and it unlearns. Nothing is trained on you.
Straight answers
What do I actually get?
A company layer on top of Claude: hire employees, freeze contracts, get a signed verdict in the morning, and one place for what still needs you. Not another chat. Not another agent.
Is this another coding agent?
No. We ship none. SessionTwin sits above Claude, Cursor, and GitHub. The moment we favor a model we stop being the referee.
Can I really leave it overnight?
That’s the product. Today it is CODE-READY, not LIVE-PROVEN — D1 prove is still the gate. We won’t pretend a qualifying night has already happened.
Can I use Cursor?
Cursor stays Blocked until LIVE-PROVEN. Hire Claude Code today. Codex and Gemini are ungoverned — not hireable.
Do you train on my work?
No. The Twin re-reads your past answers; it does not train a model. Delete an answer and it unlearns.
How do I start?
Sign in, connect morning, freeze a contract, hire onto Claude Code. Paying users stay blocked on the founder — this page invents no customer count.
Tonight: stop babysitting Claude.
Tomorrow: open the brief.
Connect your machine. Freeze the bar. See if you’d trust the morning.
Not true yet: LIVE unattended autonomy as proven, and the scheduler. A mission needing a second round waits for something to restart it. CODE-READY is not LIVE-PROVEN. Paying users stay blocked on the founder — this page invents no customer count. The rest →