VS Code

Overview

VS Code has no configuration file that registers chat models on its own. Models are contributed by an extension that registers a Language Model Chat Provider and returns its models in code. The Erebine vscode bundle ships both halves: a chatLanguageModels.json manifest (the data) and a ready-to-run provider extension under erebine-lm-provider/ (the code that loads it).

The extension reads the manifest from your workspace, registers each model in the VS Code chat model picker, and proxies requests to your Erebine Chat Completions endpoint. Streaming and tool calls travel through the standard OpenAI-compatible adapter.

Prerequisites

• VS Code 1.104 or newer (the Language Model Chat Provider API)
• A Erebine account with an active endpoint
• An API key with inference scope
• For packaging the extension: npx @vscode/vsce (no global install required)

Downloading the Bundle

From your Erebine dashboard, open the workspace Client tab and download the VS Code configuration bundle. The bundle is generated for the selected workspace and endpoint, so the manifest already carries your baseURL, model list, and a freshly minted API key.

Finding Your Endpoint URL

The endpoint URL follows the format:

Replace WORKSPACE_ID with your ws_-prefixed workspace ID and ENDPOINT_SLUG with the slug shown on your endpoint's detail page. The project ID (proj_ABC123) is visible in your dashboard URL and project settings. The workspace prefix scopes the request to that workspace; this is the form the downloaded manifest uses.

To route across every generative endpoint in a workspace instead of naming one slug, use the semantic-routed base URL. The router picks the endpoint that best matches each request:

Replace WORKSPACE_ID with your ws_-prefixed workspace ID. See Semantic Router for the full request shape.

The router picks the endpoint per request, so the serving model can change between turns. The X-Erebine-Routed-Model response header names what served each turn, and Pin a Conversation covers how a multi-turn thread is held on one model.

Bundle Contents

The downloaded VS Code bundle contains:

Model Manifest

Save chatLanguageModels.json at .vscode/chatLanguageModels.json in your project. Its shape mirrors VS Code's LanguageModelChatInformation so the provider maps each entry across one-to-one.

Provider Extension

The erebine-lm-provider/ folder is a complete, plain-JavaScript VS Code extension. There is no build step: package.json points main at extension.js, which runs directly in VS Code's Node runtime.

The extension's package.json declares the provider:

On activation the extension calls vscode.lm.registerLanguageModelChatProvider("erebine", provider) and implements the three provider methods:

• provideLanguageModelChatInformation reads .vscode/chatLanguageModels.json from the active workspace and returns one LanguageModelChatInformation per models[] entry.
• provideLanguageModelChatResponse translates the VS Code messages and tools into OpenAI shape, POSTs to provider.baseURL + /chat/completions with Authorization: Bearer <apiKey> plus any provider.headers carried in the manifest, and streams text and tool-call frames back through the progress reporter.
• provideTokenCount returns a length-based estimate.

Installing the Extension

Pick whichever path fits your workflow.

Quick try (Extension Development Host)

Open the erebine-lm-provider folder in VS Code and press F5. A second VS Code window launches with the provider active. Open your project in that window so it can see .vscode/chatLanguageModels.json.

Install for everyday use

From inside the erebine-lm-provider folder:

Reload VS Code, then open the chat model picker. The Erebine models appear under the Erebine provider.

MCP

The bundle also carries the Erebine MCP server config for the workspace tool catalog (memory, intelligence, artifacts, code, governed execution). VS Code reads MCP server definitions from the top-level servers field.

Replace <workspace_external_id> with the ws_-prefixed workspace identifier (list candidates with erectl workspaces list); the project id belongs only in the URL path. Save it at .vscode/mcp.json and reload the window so the Copilot Chat MCP picker refreshes. The MCP block coexists with the provider extension: chat completions flow through the extension while tool-call traffic reaches the workspace surface over MCP. See the general MCP integration documentation for the full tool catalog and the security gates that apply to every MCP request.

Field Reference

The following tables describe each field in chatLanguageModels.json.

// rootProvider Block

// models[]Model Fields

Each entry in models[] maps to a VS Code LanguageModelChatInformation shown in the chat model picker.

Supported Features

The provider extension talks to the Erebine API over the OpenAI-compatible Chat Completions endpoint.

Streaming (SSE)

The extension requests stream: true and reports each text delta to VS Code as a LanguageModelTextPart:

• Content is read from choices[].delta.content
• The data: [DONE] sentinel terminates the stream
• Cancelling the chat request aborts the underlying HTTP request

Tool Calling

When VS Code supplies tools, the extension forwards them as OpenAI tools with tool_choice: "auto", accumulates streamed tool_calls frames, and emits each completed call as a LanguageModelToolCallPart:

• VS Code tools map to { type: "function", function: { name, description, parameters } }
• Streaming tool calls accumulate arguments across chunks before being reported
• Prior-turn tool results are passed back as OpenAI tool-role messages

Limitations

Text Only

capabilities.imageInput is false for every model. Image and audio inputs are not part of the current integration.

Reference Implementation

extension.js is generated reference code, not a published extension. It is validated for syntax but is intended to be loaded in the Extension Development Host or packaged locally, and adapted to your needs.

API Key in the Manifest

For zero-setup, the manifest embeds the workspace API key and the extension reads it from there. A hardened build should move the key into VS Code SecretStorage and keep only baseURL + models in the manifest.

Augmentation Headers

The downloaded manifest ships one augmentation header under provider.headers: X-Erebine-Augment-Corrective-Retries: on. The extension forwards it on every request, letting Erebine retry a malformed or empty model turn a bounded number of times before returning it, so a single bad generation does not surface as a failed request. Remove that entry from provider.headers for strict passthrough, or add other X-Erebine-Augment-* headers to opt into more; see Passthrough and Augmentation.

max_tokens

Your max_tokens is sent to the model verbatim. If it does not fit alongside the input inside the model's context window, the request returns 400 context_length_exceeded rather than a quietly shortened answer. Lower the value, or send X-Erebine-Augment-Max-Output-Tokens-Clamp: on to have Erebine fit it to the remaining context. See Passthrough and Augmentation.

Troubleshooting

Models Do Not Appear in the Picker

• Confirm chatLanguageModels.json is saved at .vscode/chatLanguageModels.json in the open workspace folder.
• Confirm the extension is loaded (Extension Development Host window, or installed via code --install-extension), then reload the window.
• Check that models[] is a non-empty array.

Provider API Not Available

The Language Model Chat Provider API requires VS Code 1.104 or newer.

• Update VS Code, or follow the proposed-API fallback if your build exposes the API as a proposal.

401 Unauthorized

The API key is missing, invalid, or lacks the inference scope.

• Verify provider.apiKey in the manifest is set and current. Re-download the bundle to rotate the key.
• Check that the key has the inference scope in your dashboard.

404 Not Found

• Verify provider.baseURL includes your project ID, workspace ID, and endpoint slug, and ends with /v1.
• Check that the endpoint is active in your dashboard.

Tool Calls Not Working

• Verify the model deployed on your endpoint supports function calling, and that capabilities.toolCalling is true in the manifest.
• Check your endpoint status in the dashboard for any errors.

Verifying Connectivity

Use curl to test your endpoint independently of VS Code:

If this returns a valid response, the issue is in the extension or manifest. If it returns an error, resolve the API issue first.