One person, many agents.
Ode is a cloud harness for running AI agents. Send an email, assign a git issue, or hand off a task from the dashboard. Ode runs the agent in an isolated sandbox, asks before anything risky, and keeps a full record of what happened.
Fix the login redirect loop
- Working in sandbox
From task to done in three steps
No chat window to babysit. Hand work off, let the agent run, and come back to a result worth reviewing.
Hand it off
Send an email, assign a git issue, or start a task from the dashboard. That is the whole setup.
The agent runs
Ode spins up an isolated sandbox with the tools and permissions you set, and the agent gets to work.
You approve and merge
Risky steps pause for your approval. Review the diff and the audit log, then merge with confidence.
Meet your agents where you work
Three channels, one harness. Try the tabs in the run view above, then read how each channel behaves.
Email your agent address with a task. Ode creates a run, the agent works, and you get a reply with the result. Approval requests arrive the same way and wait for your answer.
To: atlas@ode.rest
Subject: Fix the login redirect loop
Run finished. Reply attached with the patch and test results.
Approved reply sent
Git
Assign an issue to the agent or mention it on a pull request. It works on a branch inside a sandbox and opens a pull request. It never pushes to main.
fix/rate-limit-backoff → main
3 checks passing
+87 −22 across 5 files
Pull request opened, never pushed to main
Dashboard
Start a coding task, watch the run live, approve risky steps, and review the diff before anything merges.
Live run · billing webhook refactor
Step 4 of 6 · paused for approval
Safe by default
Agents you cannot trust are agents you cannot use. These four rules hold for every run.
You decide, the agent acts
Anything destructive or irreversible waits for your explicit approval.
Untrusted input stays untrusted
Emails, issues, and repo contents are data. They can never grant new permissions.
Nothing happens off the record
Every tool call, input, output, and approval lands in the audit log.
Isolated by default
Each run gets its own sandbox, with credentials scoped to you.
Why not just use a chat window
Chat works while you watch. DIY works if you build and maintain it. Ode is the always on layer in between.
Questions, answered
The short version of how Ode works and what it promises.
Join the waitlist
We are onboarding solo developers and small teams first. Leave your email and we will invite you when a spot opens.