What's new in CodeNeter.

Snappier performance

We trimmed a lot of behind-the-scenes work, so chats feel lighter and more responsive during long tasks.

Live context meter

• The context meter updates after every response instead of waiting for the task to finish.
• Auto sessions now show the real amount of room the conversation has.

Fixes and improvements

• Context usage stays accurate after summaries and when switching sessions.
• Fixed small issues with model selection and settings navigation.

Smoother long sessions

• Typing stays fast even in very long conversations.
• Resizing the terminal no longer scrambles the transcript.

Cleaner edit previews

• Edit previews are compact and focused on what changed — press Ctrl+O to expand recent ones.
• Very large edits show a quick preview instead of slowing things down.
• Softer diff colors that are easier on the eyes.

Simpler approvals

Approval prompts now show just what matters, so you can decide faster.

Live context usage

Auto sessions now show how much room the conversation has left, updating live as you work.

Fixes and improvements

• File previews stay accurate when you reopen a session.
• Finished edits take up less room in the conversation.
• Under-the-hood security updates.

Terminal

• Long-running commands — builds, installs, dev servers — no longer block the chat or get cut off. They move to the background and keep running.
• The assistant can start a command in the background and check back on its output and status when you need it.

Agent sessions

• Longer coding runs keep more of their reasoning and previous work intact.
• Follow-up requests pick up session history more reliably after the agent has been working for a while.
• Draft text in the composer is preserved while you move through a chat.

Progress you can see

• A clearer loading state appears while the agent is preparing its next action.
• Parallel file work now streams progress into the chat instead of feeling stuck.
• Paused plan work resumes more cleanly.

Repo work

• Searching across a project is faster and easier to scan.
• Large edits and command runs behave more predictably during longer tasks.
• Activity labels stay compact so the transcript remains readable.

Models

• Default open-source model choices were refreshed to newer versions.
• Older model versions remain available when you want to compare results or fall back.

CodeNeter in the activity bar

A new CodeNeter icon lives in the activity bar on the left of your editor. Click it to open the launcher and start a session without hunting through menus.

Sign in with Codex

Connect your existing ChatGPT or Codex sign-in to use it inside CodeNeter. No separate API key, no extra billing — bring the subscription you already have.

Use your ChatGPT or Codex subscription

Connect your existing ChatGPT or Codex sign-in to use it inside CodeNeter. No separate API key, no extra billing — bring the subscription you already have.

Resume instead of restart

Hitting the turn limit no longer ends the run. The agent pauses cleanly and you can continue from where it stopped — no lost context, no lost work.

Live edit progress

Big edits now stream their progress in the chat. You see what is changing line by line instead of waiting for a single "done" message.

Parallel exploration

When the agent is getting its bearings in your codebase, it can now look at multiple files at once. Wait time drops; transcript order stays intact.

Image preprocessing in the editor

Attached images are resized and recompressed before they leave your machine. Images already within limits are sent untouched.

Consistent across editor and CLI

The editor and the CLI now follow the same image rules — same dimensions, same size thresholds, same output format. What works in one works in the other.

Clipboard image paste

Paste an image into the CLI and it shows up as an inline reference. Use as many as you want in a single prompt.

Resume on exit

When a session ends, the CLI prints the command to bring it back. Run it later and pick up where you left off with full history intact.

Tighter composer layout

Status messages now sit closer to the input, and the helper row stays anchored beneath it.

Faster exploration

Searching across large repos is dramatically faster, and results come back cleaner — no more surprises from quoting or shell rules.

Commands stay bounded

Commands now run with a clear time budget and better platform guidance. They finish or fail more predictably instead of hanging in the background.

Forgiving edits

Edits tolerate line-ending differences and prefer small, uniquely-anchored matches. The agent lands on the line you meant, not the one that looked similar.

Approval before every command

Long-running commands no longer get auto-approved. Every one goes through the same approval flow as a foreground command.

Same behavior everywhere

Exploring, reading, editing, and running commands all behave identically across the editor and the CLI. The agent's skills travel with you.

Tested across releases

A dedicated test suite covers the agent's core actions in the CLI, so behavior stays stable as the CLI evolves.

History survives the summary

Summaries used to flatten the agent's earlier work into prose, losing the link between what it did and what it learned. Now the structure is preserved — continuations behave as if the original conversation was still there.

Tighter follow-up behavior

The summary was tuned to track what you actually asked and stop inventing next steps. The agent picks up your real request, not a guess.

Cleaner chat layout

Spacing between the transcript and composer was rebalanced so the chat feels less cramped on tall screens.

Aligned summary behavior

The CLI now captures the agent's earlier work the same way the editor does. Conversations summarize the same way no matter where you started them.

Dedicated CLI release tracking

The CLI gets its own changelog from this release onward, so CLI-specific changes are easy to follow.