OpenNextEdit

Bring AI-assisted editing to the models and infrastructure you choose. OpenNextEdit adds two complementary workflows to VS Code: fast inline completions that keep you typing and next-edit suggestions that anticipate the change you are likely to make next.

Point it at a local or self-hosted OpenAI-compatible endpoint, choose prompt formats matched to your models, and tune completions and next edits independently. You get a native editor experience without giving up control of the model, its context, or where your code is sent.

Requirements

VS Code 1.92 or newer.
An OpenAI-compatible server with a /completions endpoint.
A model compatible with one of OpenNextEdit's prompt formats.

Install

Install OpenNextEdit from your editor's extension marketplace or from a .vsix file. To build a .vsix locally:

npm install
npm run package
code --install-extension open-next-edit-*.vsix

Configure

Run this command from the command palette:

OpenNextEdit: Open Config File

OpenNextEdit creates a starter TOML file if one does not exist. Update the model and URL for each feature you want to use:

active_profile = "local"
log_level = "info"

[profiles.local.inline_completion]
enabled = true
url = "http://localhost:1234/v1"
model = "Qwen/Qwen2.5-Coder-3B"
edit_format = "qwen"

[profiles.local.next_edit]
enabled = true
url = "http://localhost:1234/v1"
model = "sweepai/sweep-next-edit-v2-7B"
edit_format = "sweep"

Use api_key = "..." in either feature table if your endpoint requires authentication. Changes to the TOML file are reloaded automatically.

The default config path is:

Linux/macOS: ~/.config/opennextedit/config.toml
Windows:     %APPDATA%\OpenNextEdit\config.toml

Set opennextedit.configPath in VS Code settings to use another file.

Prompt formats

Feature	Supported `edit_format` values
Inline completion	`qwen`, `starcoder`, `gemma`
Next edit	`sweep`

Choose the format that matches how your model was trained.

Common settings

Each feature has independent model, generation, timeout, and context settings. The generated starter config contains every available option.

Setting	Purpose
`enabled`	Enables or disables the feature.
`url`	OpenAI-compatible API base URL, usually ending in `/v1`.
`model`	Model identifier sent to the endpoint.
`max_tokens`	Maximum generated tokens.
`max_context_tokens`	Context budget; currently enforced for inline completion.
`request_timeout_ms`	Maximum request duration.
`debounce_ms`	Time typing must settle before an inline request.
`edit_grouping_time_ms`	Time editing must settle before a next-edit request.
`context.lines_before_cursor`	Current-file lines collected before the cursor.
`context.lines_after_cursor`	Current-file lines collected after the cursor.
`context.max_files`	Maximum number of additional editor files included.
`context.clipboard_max_age_ms`	Maximum age for clipboard context; `0` disables it.

Inline completions

Type in an editor and pause briefly. OpenNextEdit requests a completion and VS Code displays it as ghost text.

Press Tab to accept the suggestion.
Press Escape to dismiss it.
Use VS Code's Accept Next Word Of Inline Suggestion or Accept Next Line Of Inline Suggestion commands for partial acceptance.

Next-edit suggestions

After the active document changes, OpenNextEdit waits for editing to settle and requests the most likely next change. Run OpenNextEdit: Request Next Edit to request one immediately.

The editor marks only the changed tokens and shows the replacement beside them. Hover over a change to see the complete replacement.

Press Tab once to jump to the edit.
Press Tab again to apply it.
Press Escape to dismiss it.

If an inline completion is already visible, the next edit waits in the background. Pressing Escape rejects the inline completion and shows the pending next edit if the document and cursor have not changed. Accepting the inline completion invalidates the pending edit and requests a fresh one.

Commands

OpenNextEdit: Open Config File
OpenNextEdit: Show Status
OpenNextEdit: Request Next Edit
OpenNextEdit: Jump to Next Edit
OpenNextEdit: Accept Next Edit
OpenNextEdit: Dismiss Next Edit

OpenNextEdit: Show Status opens the output channel. The inline-completion and next-edit status bars show whether each feature is idle, waiting, requesting, or has a suggestion. They can be hidden with opennextedit.showInlineCompletionStatusBar and opennextedit.showNextEditStatusBar in VS Code settings.

Context and privacy

Prediction requests can include source from the current, visible, and recently visited files, recent edits and cursor actions, diagnostics, language-server results, and recent clipboard text. Use an endpoint you trust, and set clipboard_max_age_ms = 0 under a feature's context table to exclude the clipboard.

At debug log level, the output channel includes assembled context, prompts, and model responses. These logs can contain source code and clipboard contents.

Troubleshooting

If suggestions do not appear:

Run OpenNextEdit: Show Status and check for a configuration or request error.
Confirm that active_profile names an existing profile.
Confirm that the relevant feature has enabled = true.
Confirm that the active file belongs to an open workspace folder.
Confirm that the configured server accepts requests at /v1/completions.
Confirm that edit_format matches the selected model.

Open Next Edit

Varchas Gopalaswamy