The AI coding agent that works inside VS Code
Evidence snapshot:
Kendr Code is a Visual Studio Code extension that runs a standalone agentic coding workspace in the editor: chat in the secondary sidebar, durable sessions in the Activity Bar, Git-aware and worktree-isolated work, and a bundled runtime that keeps Kendr Desktop optional.
Connect Ollama, LM Studio, vLLM, llama.cpp, OpenAI, Anthropic, Google AI, xAI, or Kendr Hosted, then work in Agent, Ask, or Plan mode against the routes that are ready and compatible.
Every change stays reviewable: open each edited file in the VS Code diff editor, edit the plan as Markdown before implementation, restore checkpoints, and approve tool activity before it runs.
Extension facts
- Extension ID
- kendr.kendr-code-vscode
- Publisher
- Kendr
- Published version
- 0.10.10
- Required editor
- Visual Studio Code 1.90 or newer
- Platforms
- Windows, Linux, and macOS on x64 and arm64
- Kendr Desktop
- Optional; the extension bundles its own authenticated runtime
- Inline completions
- Disabled by default
- Distribution
- Visual Studio Marketplace
What Kendr Code does in VS Code
Kendr Code keeps the conversation beside the editor and the durable session list in the Activity Bar. Streaming responses, tool activity, plans, todos, approvals, changed files, checkpoints, memories, previews, and delegated-agent progress stay in one conversation.
- Chat opens from the Kendr logo in the secondary sidebar, and any conversation can be opened in a retained editor tab.
- Conversations from every Kendr project are grouped into Pinned, Needs attention, Recent, Older, and Archived.
- A settled conversation can be forked into another project without replacing the original.
- Type @ to add workspace files, watch usage against the context window of the selected model, and use /compact to condense an inactive conversation.
Models and providers
The extension can create model connections through its own standalone backend or read the same safe catalog from Kendr Desktop. The picker groups routes by connection, labels each Local or Cloud, and shows only ready, enabled, policy-eligible routes compatible with the selected mode.
- Local and self-hosted: Ollama, LM Studio, vLLM, llama.cpp, and any OpenAI-compatible endpoint.
- Hosted and accounts: Kendr Hosted, OpenAI, Anthropic, Google AI, xAI/Grok, and reviewed provider subscriptions.
- Hosted models without native tool calling still reach Agent mode through the constrained Kendr coding-action transport.
- The OpenAI Codex preview appears in Ask and Plan, and does not enable full Agent mode or replace an OpenAI API key.
Review, approvals, and Git isolation
Kendr Code is built for work that has to be reviewed. Sessions are Git-aware, and a new conversation can run in the current workspace or an isolated worktree chosen directly in the composer.
- Open each changed file in the native VS Code side-by-side diff editor.
- Open the plan of an agent as an editable Markdown document, then start implementation from the plan you reviewed.
- Restore a checkpoint or rewind the conversation when a direction does not work out.
- Inline approvals and structured questions come from the coding agent rather than running silently.
Security and data boundaries
Kendr Code does not infer project trust from an open folder. File changes, approvals, capability changes, and checkpoint restores are handled by the audited authority paths of the Kendr backend.
- The standalone service binds only a random loopback port, uses a new random bearer token for every launch, verifies the bundled executable against its packaged build manifest, and stops when its owning extension host exits.
- Account sessions and local API tokens are kept in VS Code SecretStorage and are not exposed to the chat webview.
- Browser sign-in uses the OAuth authorization code flow with PKCE, and the callback returns only to the authenticated loopback runtime.
- Attachments are restricted to files inside the selected workspace, reject binary and unavailable files, and use bounded context budgets.
Editor, platform, and remote support
Standalone release packages are produced for Windows x64 and arm64, Linux x64 and arm64, and macOS x64 and arm64. Release automation publishes checksums, a CycloneDX SBOM, and SLSA/in-toto provenance with each release set.
- VS Code 1.106 and newer show the dedicated right-sidebar chat launcher; earlier supported versions keep Sessions and a chat fallback in the Activity Bar.
- For Remote SSH, dev containers, or WSL, install the matching platform package in the remote workspace extension host.
- Retry-safe requests use bounded deadlines and server-directed backoff, and interrupted event streams resume from the last durable cursor.
Commands
Every Kendr Code command is available from the VS Code Command Palette.
| Command | Action |
|---|---|
| Kendr Code: Open Chat | Open or focus the right Kendr chat |
| Kendr Code: New Conversation | Create a local or isolated-worktree session |
| Kendr Code: Search Conversations | Search and switch durable sessions |
| Kendr Code: Show Workspace Changes | Open the Git diff of the selected session |
| Kendr Code: Open Plan for Review | Open the current plan as an editable Markdown document |
| Kendr Code: Start Plan Implementation | Save the reviewed plan and start its implementation |
| Kendr Code: Compact Conversation Context | Condense the working context of an inactive conversation |
| Kendr Code: Add Model or Provider Account | Connect a subscription, local runtime, hosted service, or API provider |
| Kendr Code: Detect Local Models | Find supported local model servers on standard endpoints |
| Kendr Code: Toggle Inline Completions | Enable or disable predictive ghost text |
Install from VS Code Quick Open
Open Quick Open with Ctrl+P, paste the command, and press Enter. The bundled runtime starts automatically after the reload.
ext install kendr.kendr-code-vscodeFrequently asked questions
Is Kendr Code a standalone VS Code extension?
Yes. Kendr Code bundles its own authenticated runtime, so it works without Kendr Desktop. It first reuses a saved kendr serve connection or a running Kendr Desktop session, and otherwise starts the bundled runtime automatically.
Can Kendr Code run local models?
Yes. It connects Ollama, LM Studio, vLLM, llama.cpp, and any OpenAI-compatible endpoint, and the Detect Local Models command probes the standard loopback endpoints so you do not have to type a URL. The model picker labels every route Local or Cloud.
Do I need a Kendr account to use the extension?
Sign in to Kendr to enable entitled Kendr Hosted models. Local runtimes and your own provider accounts or API keys are connected from the extension, and credentials stay backend-side, with tokens in VS Code SecretStorage rather than in settings or the chat webview.
Which VS Code versions and platforms does Kendr Code support?
VS Code 1.90 and newer, with standalone packages for Windows, Linux, and macOS on x64 and arm64. VS Code 1.106 and newer show the dedicated right-sidebar chat launcher, and earlier supported versions keep Sessions and a chat fallback in the Activity Bar.
Does Kendr Code work with Remote SSH, dev containers, and WSL?
Yes. Install the matching Kendr Code platform package in the remote workspace extension host so its bundled backend runs in the same environment as the workspace. Operating a separate kendr serve process remains available for managed or remote runtimes.
How does Kendr Code review changes before they land?
Each changed file opens in the VS Code side-by-side diff editor, plans open as editable Markdown documents before implementation starts, and checkpoints can be restored. File changes, approvals, capability changes, and checkpoint restores stay explicit, and project trust is never inferred from an open folder.
Where does my code go when Kendr Code runs?
That follows the route you select: a local runtime keeps inference on your machine, while a hosted route sends the request to that provider. Attachments are limited to files inside the selected workspace, and inline completion is disabled by default and sends only a bounded prefix and suffix.
Sources and evidence dates
- Kendr Code on the Visual Studio Marketplace (distribution, checked 2026-09-04)
- Kendr download and release options (documentation, checked 2026-09-04)
- Kendr security controls (documentation, checked 2026-09-04)