Claude Code needs a developer. Ona is the developer.
Claude Code is a fast, thin interface that gives engineers direct leverage on a strong model. Many customers run it inside Ona today. But enterprises also need the layer underneath: reproducible environments, runtime enforcement, a powerful agent harness, and orchestration at scale.
See it in your VPC
15-minute walkthrough. No slide deck.
Capability is solved. Governance isn't.
Thin harnesses are powerful, but they assume the user is the safety system. Enterprises need agents that work for the median engineer — with reproducible environments, runtime enforcement, and guardrails that don't depend on individual expertise.
Reproducible execution
Dev Containers define the runtime. Tasks and Services start databases and dependencies deterministically. Prebuilds snapshot the environment after setup. Any agent running inside Ona — Claude Code included — starts from a known-good state and closes the loop: change, test, iterate, PR.
Fleet-scale parallelism
Run hundreds of agents across repos simultaneously, each in its own isolated environment with concurrency limits. Write five specs, step out to lunch, come back to merge-ready code.
Your infrastructure, your rules
Deploy inside your VPC. Network policies restrict agent connectivity. Team-level cost caps prevent surprise bills. The agent harness, tools, and services in your network improve context and capability — things a CLI on a laptop can't standardize.
What the thin harness doesn't cover
Claude Code is a thin harness optimized for the individual. Ona is the platform that makes any agent enterprise-grade.
Ona
Claude CodeInside your VPC. Code never leaves your network.
CLI runs on developer laptops. Claude Code Web runs on Anthropic’s VMs. Your code goes to their infrastructure either way.
Command dozens of agents on independent EC2s in your VPC. Parallel execution across repos.
Desktop "parallel sessions" use Git worktrees on one machine — same filesystem, no real isolation. No cross-repo orchestration.
Ephemeral VM per task. Mistakes are disposable.
Local machine or Anthropic’s VMs. Desktop worktrees share the same filesystem and environment. No per-task isolation.
Command deny lists and kernel-level security guarantees. Every agent runs in an isolated VM.
Permission prompts and bash tool restrictions. Runs on developer laptops.
Org guardrails, command deny lists, policy controls.
managed-settings.json for Enterprise tier.
Run agents across hundreds of repos simultaneously.
One terminal, one repo. Desktop worktrees share a filesystem and create merge conflicts. No real concurrency.
Agents work while your laptop is closed.
Remote Control is a browser window into a local CLI session. If the laptop sleeps, the session dies.
Event-triggered agents respond to PRs, CVEs, or schedules. Sentry triage overnight, docs drift detection, CVE remediation — each in isolated environments, across repos.
No built-in automation. Requires a developer to pull up the repo, configure access, and workshop a fix one-on-one.
Org-level skills pushed to every repo at once. One change, thousands of repos.
AGENTS.md per repo. Each maintained by an individual engineer. Claude Code still needs to be babysat — human attention required for every session.
Team-level cost caps, usage insights across agents and engineers, org-level policies.
Token usage tracking. No team-level caps or org-wide cost visibility.
Every human and agent action logged. SIEM export.
180-day audit logs on Enterprise plan.
Large-scale rollouts, enterprises, parallelization.
Individual developers, power users, interactive coding sessions.
Where it runs
Inside your VPC. Code never leaves your network.
CLI runs on developer laptops. Claude Code Web runs on Anthropic’s VMs. Your code goes to their infrastructure either way.
Multi-agent orchestration
Command dozens of agents on independent EC2s in your VPC. Parallel execution across repos.
Desktop "parallel sessions" use Git worktrees on one machine — same filesystem, no real isolation. No cross-repo orchestration.
Isolation
Ephemeral VM per task. Mistakes are disposable.
Local machine or Anthropic’s VMs. Desktop worktrees share the same filesystem and environment. No per-task isolation.
Security
Command deny lists and kernel-level security guarantees. Every agent runs in an isolated VM.
Permission prompts and bash tool restrictions. Runs on developer laptops.
Governance
Org guardrails, command deny lists, policy controls.
managed-settings.json for Enterprise tier.
Parallelism
Run agents across hundreds of repos simultaneously.
One terminal, one repo. Desktop worktrees share a filesystem and create merge conflicts. No real concurrency.
Background execution
Agents work while your laptop is closed.
Remote Control is a browser window into a local CLI session. If the laptop sleeps, the session dies.
Automations
Event-triggered agents respond to PRs, CVEs, or schedules. Sentry triage overnight, docs drift detection, CVE remediation — each in isolated environments, across repos.
No built-in automation. Requires a developer to pull up the repo, configure access, and workshop a fix one-on-one.
Org-level context
Org-level skills pushed to every repo at once. One change, thousands of repos.
AGENTS.md per repo. Each maintained by an individual engineer. Claude Code still needs to be babysat — human attention required for every session.
Cost controls
Team-level cost caps, usage insights across agents and engineers, org-level policies.
Token usage tracking. No team-level caps or org-wide cost visibility.
Audit trail
Every human and agent action logged. SIEM export.
180-day audit logs on Enterprise plan.
Best for
Large-scale rollouts, enterprises, parallelization.
Individual developers, power users, interactive coding sessions.
Background agents are where the value compounds
The biggest shift isn't which agent you use interactively. It's what happens when agents run continuously in the background.
CVE remediation at scale
Org-wide migrations
Platform standardization
Developer onboarding
End-to-end engineering workflows
Claude Code makes one dev faster. Ona makes the org faster.
Many customers run Claude Code inside Ona today. The real answer for enterprises is “both” — Claude Code for interactive leverage, Ona for the platform underneath.
Fast, thin interface for interactive coding. Strong engineers get direct leverage on a strong model.
Orchestrates the environment, runs tests and services from config, ships across repos in parallel. Closes the loop autonomously.
Run Claude Code today, switch to whatever wins next quarter. Ona is the layer underneath: orchestration, governance, cost controls, all inside your enterprise boundary.
Common questions
Don't standardize on an interface. Standardize on the layer underneath.
Identity, permissions, guardrails, audit logs, cost caps, and repeatable verification. The layer that turns any agent into an enterprise-grade workflow.