Skip to content

Kilo Code makes Codestral its default autocomplete model

The v7.8.30 pre-release retires Mercury Edit 2 and Next Edit, drops the edit surface from its SDK, and adds hosted workspace snapshots to the CLI.

Draftship Editorial3 min read

Kilo Code published v7.8.30 as a pre-release on 16 November, and the change most people will notice first happens inside the editor. The release retires the Mercury Edit 2 autocomplete models and Next Edit, and makes Codestral the default autocomplete model. The Kilo Code Codestral autocomplete switch also catches saved settings: any existing Mercury or Inception selection is moved to Codestral rather than left pointing at a retired model. The full item list is in the v7.8.30 pre-release notes.

What the release confirms for the VS Code extension

The retirement is not only a settings default. The SDK no longer exposes client.kilo.edit, and @kilocode/kilo-gateway no longer exports ./edit or ./edit-prompt. That is an interface removal, not a rename, and the release notes do not describe a replacement path for either entry point.

Dependency versions were bumped alongside it: @kilocode/[email protected], @kilocode/[email protected], @kilocode/[email protected], @kilocode/[email protected], @opencode-ai/[email protected], @opencode-ai/[email protected] and @kilocode/[email protected]. The same dependency set appears under the CLI section.

Hosted workspaces get more visibility in the CLI

Two CLI changes concern hosted workspaces rather than the editor. The first exposes change snapshots, revision-fenced per-file reads, explicit refresh, and verified linked-pull-request diffs through Session Host and the generated SDK. The second exposes hosted workspace pull request summaries with explicit refresh, and the notes say partial observations are preserved along with distinct authorization, connectivity and rate-limit states. In plain terms, the CLI is being given a way to report what changed in a hosted workspace and to distinguish "no access" from "network down" from "rate limited" rather than collapsing them into one failure.

Smaller fixes in the same release

  • Cloud Agent CLI authentication now works by exchanging login credentials for a Cloud Agent control token.
  • hono moves to 4.13.18 and @hono/node-server to 2.1.7, and node-forge to 1.4.3, both credited to Dependabot security alerts.
  • The JetBrains Agent Manager gets a unified Add menu for local and Cloud tasks, along with a long list of Cloud workspace behaviours: opening workspaces in the correct organization, closing and reopening sessions, deleting whole workspaces with retry, creating a Cloud workspace with an empty prompt, and keeping session titles in sync.
  • Session cost now shows what Kilo Gateway charged, with the provider's inference cost shown only for requests that use your own API key.
  • Questions stay open while background tasks finish, so answers are not dismissed mid-typing.

What this changes for people building apps

If your tooling or scripts called client.kilo.edit, or imported ./edit or ./edit-prompt from the gateway package, this release breaks that call. The notes do not offer a migration step, so the practical move is to pin the previous version while you check your imports. This is inference rather than something the notes state, but a removed export normally surfaces as an import error at build time rather than a silent fallback.

Autocomplete itself should keep working for most users, because saved Mercury and Inception selections are redirected to Codestral instead of being cleared. What changes is which model answers, and the notes do not say how the two compare.

Limits and open questions

The source does not name a Codestral version, does not say whether the retired Mercury or Inception models remain selectable for anything other than autocomplete, and does not describe what replaces Next Edit, if anything. There is no deprecation timeline, no migration guide for the SDK removal, and no benchmark comparing the old and new defaults. The release is also marked pre-release, so behaviour may shift before it is promoted.

For AI app builders

Editor-level model swaps like this one matter when your workflow depends on a specific autocomplete behaviour or a specific SDK method. If you would rather spend the time on the app than on the toolchain, Draftship is an AI app builder where you describe an app in plain language and it writes, builds and hosts it, with a built-in database, user sign-in and hosting included. Paid plans add a choice of AI model, private projects, source code access, custom domains and GitHub push. The model choice is documented under using the AI chat and choosing a model. Draftship is not affiliated with Kilo Code, and nothing here suggests Draftship supports the models or SDK changes described above.

Primary sources

This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.