Open Agents.

Code in a durable cloud Session, create GitHub-triggered or scheduled Automations, and inspect each Run before acting on it.

Open Source

Why Vercel? It's the identity provider for Open Agents — one account to sign in, no separate password to create.

  1. 1.Connect GitHub

    Choose the repositories Open Agents may access.

  2. 2.Start a Session

    Open a durable interactive workspace with repository, branch, and sandbox context.

  3. 3.Create an Automation

    Configure manual, GitHub, or scheduled coding steps subject to permissions and readiness.

  4. 4.Inspect a Run

    Review execution status, trigger, evidence, and available recovery controls.

open-agents
Illustrative example

Build the auth flow with GitHub OAuth

12 tools3/4
42s

app/api/auth/route.tsapp/api/auth/callback/route.tsmiddleware.ts
Request changes or ask a question...
Configured model1%

Start a durable coding Session.

A Session keeps repository, branch, sandbox, and conversation context together while you direct and review the coding work.

  • File ops, search, shell, and task delegation built in
  • Helper roles can explore and implement scoped work
  • Choose from configured models

Illustrative example

grepauth patterns in src/
readlib/session.ts
writeapp/api/auth/route.ts
writeapp/api/auth/callback/route.ts
editmiddleware.ts
bashbun run typecheck
auth flow complete. 2 routes, middleware, callback. typecheck passes.

Cloud sandboxes, not local machines.

Every session runs in an isolated Vercel sandbox with its own branch. Changes stay on that branch until you commit — turn on auto commit & push in Settings if you'd rather it happen automatically.

  • Ephemeral environments with full git integration
  • Inactivity can hibernate the sandbox; snapshots support recovery
  • Snapshot and restore filesystem state

Illustrative example

sandboxfeat/auth
active
full filesystem, network, runtime access
branch
feat/auth
snapshot
—
cost
varies by usage

Create an Automation, then inspect its Run.

Automations coordinate configured coding steps and triggers. Each attempt creates an inspectable Run with status and available evidence.

  • Manual, GitHub webhook, and scheduled triggers
  • Configured permissions and readiness gates apply
  • Run history exposes status, evidence, and recovery controls

Illustrative example

automationruninspectable
convert messages
run agent loop
persist messages
record usage
refresh diff cache
auto-commit + push
example complete · 6 steps recorded

A clear path
from idea to evidence.

Connect a repository, work interactively, automate repeatable coding tasks, and inspect what actually ran.

001

Connect GitHub

Choose which repositories Open Agents may access before starting repository-scoped work.

002

Start a Session

Work interactively with repository, branch, sandbox, and conversation context kept together.

003

Create an Automation

Configure coding steps and start them manually, from GitHub events, or on a schedule.

004

Inspect a Run

Review an execution attempt's status, trigger, evidence, and available recovery controls.