You describe the outcome.Our AI agents write the code and ship it.
Planned across your repos, verified before you review, and deployed to your cloud.
Describe a feature, a migration, or a fix. You approve what reaches production.
From outcome to production
You describe an outcome
A prompt or a backlog item — “Add a health check to every service.” The planning agent reads your repos and plans the change across every affected service.
The planning agent turning one outcome into a coordinated plan across your repos.
Native MCP server + open-source skill — works with the agents you already use
How the work flows
Context, then a plan, then agents doing the work.
The same three steps a good engineer takes: understand the system, agree on what needs to be done, then build it.
1 · Context
Your agents start with your system, not a blank repo.
Before anyone writes a line of code, Deployment.io builds a shared picture of your org: every repository, every environment and database, everything actually running in your cloud, and which repo ships each piece of it.
- Every repo in your org
- Connect GitHub, GitLab, or Bitbucket once and the catalog builds itself — languages, services, and entry points included.
- What's actually running
- The scan happens from inside your own cloud: environments, services, databases. Your credentials never leave your account.
- The agent knows which repo owns what
- It won't go editing the wrong service because two repos have similar names. Where it isn't sure, it asks instead of guessing.
- Corrected once, reused forever
- Fix a low-confidence guess and it sticks. Every session and every task afterwards starts from the corrected answer.
It refreshes on demand, so it never quietly drifts out of date.
Service → repo
- checkout-apiacme/checkouthigh
- billing-workeracme/billinghigh
- notify-svcacme/platform?low
Low-confidence matches wait for a human to confirm them — once.
2 · Assistant
Think it through before anything gets built.
Start a session and you're talking to an agent that has already read the relevant repos. It asks the questions a senior engineer would ask instead of guessing — and turns the conversation into a spec you can actually hand off.
- Think out loud without risking anything
- A session only reads and discusses. Nothing you explore becomes a branch, a commit, or a pull request until you decide it should — so you can chase a bad idea for ten minutes and lose ten minutes.
- Catch the vague request before it costs you a day
- The conversation becomes a written spec: the goal, what counts as finished, what's off the table. You find out the ask was ambiguous now, not after an agent spent an afternoon building the wrong thing.
- Never locked to one model
- Plan with one, build with another. When a better coding model ships next quarter, point the work at it — nothing to rewire, and you can bring your own provider key.
When the spec reads right, convert it — it lands in the Tasks backlog as a draft, ready for whoever starts it.
Goal
Move session auth onto rotating refresh tokens without logging anyone out.
Acceptance criteria
- Existing sessions stay signed in through the cutover
- Refresh tokens rotate on every use
- Rate limit is enforced per user, not per IP
Out of scope
SSO, password reset emails.
3 · Tasks
A board your team shares with its agents.
The same board you'd use to track work with people — except the assignee is a coding agent. Drop a card in the backlog, pick who runs it, and watch it move to done.
Rotate refresh tokens
Drop legacy /v1 routes
Add SBOM to release
Node 18 → 20, all services
Fix flaky checkout test
Cache invalidation bug
Every card runs in an isolated container in your cloud and ends as a pull request you review.
Assign work to an agent, not a person
Match the model to the job: something cheap for a dependency bump, something strong for a migration. Claude Code, Codex, or your own provider key.
Your code never lands on our infrastructure
Every task runs walled off inside your own AWS account. Your code, your secrets, and your build artifacts stay there — we never hold a copy.
Lands in the workflow you already have
Every task finishes as a pull request in your repo, so your existing review, CI, and branch rules just apply. Nothing new for the team to adopt.
Approvals stay with your team
Tool calls that need permission queue up for a human. Nothing reaches production without a sign-off.
What our agents ship
First-class support for the workloads teams actually run in the cloud.
Static Sites
Our agents ship React, Next.js, Vue, and other static sites with HTTPS, a global CDN, cache invalidation, and rollbacks built in.
e.g., “Migrate our docs from Hugo to Astro and ship it.”
Web Services
Go, Node.js, Python, Ruby, and Rust services deploy with auto-scaling, health checks, and zero-downtime rollouts.
e.g., “Bump Node from 18 to 20 across all services and roll out.”
Multiple Environments
Stand up dev, staging, and production environments — each in its own cloud account if you want. Our agents deploy to any of them.
e.g., “Stand up a staging environment for the checkout-v2 branch.”
The platform underneath
Built for agents. Controlled by you.
The features that make agent-led deploys safe, fast, and predictable.
Native MCP tool surface
Agents call real deploy actions over MCP. Not chat — actual deploys.
Zero-config auto-detection
Our agents read your Dockerfile, port, health check, and build command from the repo. No YAML.
Human approval gates
Production deploys pause for sign-off. Approve from Slack or the dashboard.
RBAC for humans and agents
Scope what each teammate and each MCP token can deploy, read, or approve.
Auto-deploy on git push
Wire a branch to an environment once. Every push triggers a deploy.
Custom domains, HTTPS, Slack alerts
Bring your domain. Certs, CDN, and load balancing wired automatically. Alerts in Slack.
Security & control
Your cloud. Your data. By architecture, not promise.
Deployment.io runs inside your own cloud. We don't proxy your traffic, store your code, or hold your secrets.
Read the full security modelYour cloud, your data
Code, secrets, build artifacts, and runtime traffic never leave your cloud.
Open-source runner
The runner that touches your infrastructure is open source. Audit it, fork it, run your own build.
Revoke access anytime
We connect via a scoped role you control. Detach the role and we lose all access — instantly.
Approval gates by default
Production deploys require human approval. Role-based access controls scope what each teammate (and each agent) can do.
Ship outcomes. Keep the keys.
Describe what you want. Every change is tested before you review it; every production deploy waits for your sign-off.