Contenox for VS Code

Run local, reviewable AI workflows in VS Code. Use it to:

Chat about your code with @contenox — workspace questions and source-aware explanations, no GitHub Copilot sign-in required.
Fix problems — explain or fix diagnostics from the lightbulb or command palette.
Review before you push — summarize your workspace changes and draft a commit message from the current diff.
Get inline autocomplete from a local or cloud model you choose (needs a separate autocomplete model — see Autocomplete below).
Run tool workflows you approve — MCP, OpenAPI, and shell tools gated by human-in-the-loop approval.

The extension is the VS Code frontend for the Contenox runtime: it uses the models and providers you configure, keeps state on your machine, and asks for approval before crossing policy boundaries.

Why Contenox

Modern coding agents are useful, but too much AI-assisted work still depends on fragile prompt habits, hidden context, vendor-specific behavior, and black-box tool execution.

Contenox takes a different approach: repeated AI work should be defined, inspected, versioned, and controlled like an engineering artifact. Contenox Chains make prompts, model routing, tool access, retries, branches, budgets, and approval gates reviewable.

Contenox is runtime-first, not editor-first: it sits across your editor, terminal, project state, tools, sessions, and model/provider configuration.

What You Can Do

Chat with @contenox in VS Code Chat without requiring GitHub Copilot sign-in for Contenox-owned requests.
Ask about selected code or open files with editor context attached.
Explain and fix diagnostics from the command palette or lightbulb menu.
Review workspace changes and draft commit messages from the current diff.
Use inline autocomplete powered by a separate, FIM-capable autocomplete model you configure (see Autocomplete).
Connect local or external models, including local GGUF models and common cloud providers.
Use MCP, OpenAPI, and local tools through Contenox runtime policies.
Approve or deny gated tool actions before they run.
Keep sessions, configuration, telemetry, and workflow state local.

Local-First Ownership

The extension includes the Contenox runtime and starts it locally for the active workspace. You can use the bundled runtime, point the extension at an existing Contenox binary, or reuse an existing Contenox data directory.

In VS Code Remote Development, WSL, Dev Containers, SSH, and GitHub Codespaces, the extension runs as a workspace extension on the remote workspace host. That means the Contenox runtime also runs on the remote host, not on your local laptop. Install the Marketplace/VSIX package that matches the remote operating system and architecture, or set contenox.binaryPath to a contenox binary that exists inside that remote environment.

Contenox is designed around:

local-first usage
model and provider choice
auditable execution traces
human-in-the-loop approval
explicit tool and policy boundaries
workflows that can graduate from chat into repeatable Chains

Getting Started

Install the extension.
Run Contenox: Run Setup from the command palette.
Choose or configure a model/provider.
Run Contenox: Open Chat.
Ask @contenox about your code, or use the editor actions (Ask About Selection, Explain Diagnostics, Review Workspace Changes, Draft Commit Message) from the command palette and lightbulb menu.

Autocomplete

Contenox can provide inline ghost-text suggestions as you type. It is disabled by default; enable it after choosing an autocomplete model.

Autocomplete uses a separate model from chat, and that model must be FIM-capable (fill-in-the-middle). Set it from the CLI:

contenox config set default-autocomplete-provider mistral
contenox config set default-autocomplete-model codestral-latest

or override it per workspace with contenox.autocompleteModel / contenox.autocompleteProvider. If no autocomplete model is configured, no suggestions appear and the reason is only logged to the output channel.

Enable autocomplete from the status bar or with Contenox: Enable Autocomplete. If nothing shows up after enabling it, run Contenox: Test Autocomplete At Cursor — it reports whether the model returned usable text.

Remote Development

Contenox declares itself as a VS Code workspace extension so workspace files, git state, terminals, tools, and the Contenox runtime live on the same machine as the checked-out code.

Remote behavior to expect:

SSH, WSL, containers, and Codespaces need a Linux-compatible runtime in the remote environment.
Contenox: Run Setup opens a terminal in the remote workspace.
contenox.dataDir is resolved in the remote environment when set.
Local diagnostic telemetry is written where the runtime runs.
Use Developer: Show Running Extensions to confirm Contenox is running in the remote extension host.

If the runtime cannot start, run Contenox: Show Output; the log includes the remote name, platform, architecture, runtime path, and the exact spawn error.

CLI Compatibility

If you already use Contenox from the terminal, the extension can reuse the same state and configuration. Common settings are available through the CLI:

contenox config set default-provider mistral
contenox config set default-model codestral-latest
contenox config set default-autocomplete-provider mistral
contenox config set default-autocomplete-model codestral-latest

Contenox