Harness · Codex

Codex on the Software Lifecycle Cloud

Codex on KERNSTACK means OpenAI’s coding agent producing branch-shaped work — locally through its CLI or as delegated tasks — against a repository whose entire runtime is declared in stack.json. Branches are the contract: Codex writes them, KERNSTACK gives agent branches their own deployable environments, and a person approves the plan before anything reaches production.

Agent SandboxesHosted Gitstack.json IaC
Talk to us Read Technical Documentation → Software Lifecycle Cloud →

Delegated tasks, branch discipline

Codex is at its best when work arrives as well-scoped tasks whose results can be reviewed as diffs. A KERNSTACK project makes those diffs unusually informative: a task that “adds invoice export” shows up as a worker, a workflow step, and a gateway route in stack.json, not just code files. The reviewer sees the operational consequence of the change in the same review as the implementation.

Delegated and sandboxed attempts belong on agent branches. On KERNSTACK hosted git, agent/ branches deploy to the matching agent- environment, so a Codex-produced change can be clicked through at a real URL — by the reviewer, in a browser — before anyone discusses merging it.

From task result to production, with a gate between

The path from a finished task to production is deliberately unglamorous: kernstack plan renders the change set, a person reads it, kernstack deploy applies it. Destructive entries require an explicit confirmation flag, and a project deploy gate can require review-approved source, so the agent’s confidence is never the shipping criterion.

  • Pull any agent branch locally with kernstack git checkout agent/ and run the exact code the task produced
  • Move work worth keeping onto feature/ before discarding the sandbox — kernstack use offers this at the switch
  • Deploy an agent branch outside its own environment and the CLI asks for confirmation; the boundary is mechanical

Supported Models & Workflow

Supported Models & Engines: GPT family — your choice in Codex

FAQ

Frequently asked questions

You review two artifacts instead of one: the code diff, and kernstack plan, which states the platform consequences — resources created, updated, deleted. A task that quietly drops a table announces it in the plan under “Will delete” before deploy, not in a postmortem.

Yes — that is what agent environments are for. Work on an agent/<id> branch deploys to the agent-<id> environment, which behaves like production in shape but touches neither main nor production data paths you have not pointed it at. Reviewers get a URL, not a screenshot.

The GPT-family models your Codex setup provides. KERNSTACK sits below the editing layer: it receives branches and plans, so model selection, task phrasing, and parallelism stay Codex-side decisions that never leak into the platform contract.

Connected in the Project Graph

Related capabilities, architectures, and comparisons

Category Deep DiveCloud Infrastructure for AI Coding AgentsWhy coding agents need a sandboxed lifecycle cloud instead of raw AWS IAM keys.Platform CapabilityKERNSTACK Agent SandboxesIsolated execution environments with branch isolation and credit ceilings.coding-harnessClaude Code on the Software Lifecycle CloudClaude Code on KERNSTACK: a terminal agent edits the repo, reads kernstack plan --json, and ships only through a human-approved deploy gate.coding-harnessCursor on the Software Lifecycle CloudCursor on KERNSTACK: agents edit the repo in the IDE while stack.json, plan diffs, and agent branches keep every change reviewable.coding-harnessPi on the Software Lifecycle CloudPi on KERNSTACK: the provider-agnostic terminal agent edits any model into the same governed flow — branch, plan --json, approved deploy.collaboration-channelTelegram on the Software Lifecycle CloudTelegram on KERNSTACK: bind a group to the project, invite partners by link, and get @agent answers and demo links in the chat you already use.collaboration-channelWhatsApp on the Software Lifecycle CloudWhatsApp on KERNSTACK: bind a WhatsApp conversation to a project topic and get @agent replies and demo links where your partners already are.

Ship your stack on the Software Lifecycle Cloud.

Closed preview on dedicated European infrastructure. Bring a repository or start from a blueprint.

Talk to us See pricing & credits →