Omni Chat Provider

Use OpenAI, Anthropic, Gemini, Ollama, and OpenAI-compatible backends inside VS Code Copilot Chat through one extension.

The extension contributes a single vendor, OmniChat, to Copilot's language model system. Each model group in Chat: Manage Language Models points at one configured providerId, so removing a group only removes the Copilot-side mount. Your own omnichat.providers, omnichat.models, and stored API keys stay intact.

Features

Native adapters for openai, openai-responses, anthropic, gemini, and ollama
Provider-level connection settings with model-level overrides
Per-provider API keys stored in VS Code Secret Storage
Native Copilot model management via Chat: Manage Language Models
Stable internal model routing with clean picker labels
Retry, delay, custom headers, and system-prompt interception
Optional model variants via configId
Commit message generation command
Commit message model selection and prompt editing commands

How It Works

There are three layers:

omnichat.providers Defines backend connections such as baseUrl, apiMode, and shared headers.
omnichat.models Defines the actual models shown in Copilot, each mapped to a provider via provider or owned_by.
Chat: Manage Language Models Creates Copilot model groups that attach one providerId to the OmniChat vendor.

Deleting a Copilot group does not delete OmniChat settings or OmniChat secrets.

Requirements

VS Code 1.104+
GitHub Copilot Chat installed

Setup

1. Install the extension

Install the extension from VS Code Marketplace, install the VSIX manually, or install from Open VSX after publishing.

2. Define providers

Add provider backends in settings.json:

"omnichat.providers": [
  {
    "id": "openai",
    "baseUrl": "https://api.openai.com/v1",
    "apiMode": "openai"
  },
  {
    "id": "anthropic",
    "baseUrl": "https://api.anthropic.com",
    "apiMode": "anthropic",
    "headers": {
      "anthropic-version": "2023-06-01"
    }
  },
  {
    "id": "ollama",
    "baseUrl": "http://localhost:11434",
    "apiMode": "ollama"
  }
]

3. Define models

Each model must point at a provider:

"omnichat.models": [
  {
    "id": "gpt-5.4",
    "provider": "openai",
    "context_length": 128000,
    "max_completion_tokens": 8192,
    "vision": true
  },
  {
    "id": "gpt-5.4",
    "provider": "openai",
    "configId": "reasoning",
    "reasoning_effort": "high",
    "max_completion_tokens": 16384
  },
  {
    "id": "claude-sonnet-4",
    "provider": "anthropic",
    "max_tokens": 8192,
    "thinking": {
      "type": "enabled",
      "budget_tokens": 4096
    }
  },
  {
    "id": "llama3.1:70b",
    "provider": "ollama",
    "num_ctx": 32768,
    "temperature": 0.4
  }
]

4. Add or edit a provider

Run:

OmniChat: Edit Provider

This flow lets you:

Pick an existing provider and edit its apiMode, baseUrl, and API key
Add a new provider with the same form

Provider API keys are stored in VS Code Secret Storage under omnichat.apiKey.<providerId>.

5. Mount the provider into Copilot

Run:

Chat: Manage Language Models

Then:

Add a new language model group
Choose OmniChat
Enter the providerId you want to mount

That group will now expose only the models belonging to that provider.

Configuration Reference

`omnichat.providers`

Provider-level backend settings:

id
baseUrl
apiMode
headers

`omnichat.models`

Model-level settings:

id
provider or owned_by
configId
displayName
family
context_length
vision
temperature
top_p
headers
delay
extra
useForCommitGeneration
include_reasoning_in_request

API-specific model fields

OpenAI

max_tokens
max_completion_tokens
reasoning_effort
frequency_penalty
presence_penalty

OpenAI Responses

max_output_tokens
reasoning

Anthropic

max_tokens
thinking
top_k

Gemini

maxOutputTokens
topK
topP
thinkingConfig

Ollama

num_predict
num_ctx
num_gpu
top_k
min_p
repeat_penalty

Retry

"omnichat.retry": {
  "enabled": true,
  "maxAttempts": 3,
  "intervalMs": 1000,
  "statusCodes": [429, 500, 502, 503, 504],
  "retryRequestErrors": true,
  "retryNetworkErrors": true,
  "retryEmptyResponse": true,
  "timeoutMs": 120000
}

retryRequestErrors: Retry request errors such as retryable HTTP status codes (429, 5xx) and per-attempt timeouts.
retryNetworkErrors: Retry network or stream interruption errors such as terminated, aborted, ECONNRESET, and socket hang up.
retryEmptyResponse: Retry when the request succeeds but the response stream finishes without yielding any content.

Delay

Global:

"omnichat.delay": 1000

Per model:

{
  "id": "gemini-flash",
  "provider": "gemini",
  "delay": 1500
}

System prompt handling

"omnichat.systemPrompt.mode": "passthrough",
"omnichat.systemPrompt.content": ""

Modes:

passthrough
replace
append
disable

Commit message generation

"omnichat.commitLanguage": "English",
"omnichat.commitMessageModel": "openai/gpt-5.4::reasoning",
"omnichat.commitMessagePrompt": ""

commitLanguage: 生成 commit message 的语言
commitMessageModel: 指定用于生成 commit message 的模型，格式为 providerId/modelId 或 providerId/modelId::configId
commitMessagePrompt: 追加的自定义 commit prompt

Commands:

OmniChat: Generate Commit Message
OmniChat: Stop Generating Commit Message
OmniChat: Select Commit Message Model
OmniChat: Edit Commit Message Prompt

The generator prefers staged changes and falls back to unstaged changes when nothing is staged. It will also trim oversized diffs to fit the selected model budget.

Token counting

provideTokenCount now uses a provider-aware strategy:

gemini: native countTokens endpoint when available
anthropic: native token count endpoint when available
openai / openai-responses: local tiktoken estimation with message-structure overhead
others: improved heuristic fallback for text, thinking parts, tools, and images

Build

Install dependencies:

npm install

Compile:

npm run compile

Package VSIX:

npm run package

自动 bump 版本：

pnpm run bump           # patch
pnpm run bump minor
pnpm run bump major
pnpm run bump 0.2.0
pnpm run bump -- --dry-run

如果执行前 Git 工作树是干净的，脚本还会自动创建一条 chore: bump version to x.y.z 提交，并打上 vx.y.z tag。

GitHub Actions

The repository includes:

.github/workflows/marketplace.yml
.github/workflows/openvsx.yml

Marketplace workflow:

Build and package extension.vsix
Publish that VSIX to VS Code Marketplace using MARKETPLACE_TOKEN

Open VSX workflow:

Build and package extension.vsix
Publish that VSIX to Open VSX using OPENVSX_TOKEN

Required secrets

Set these repository secrets before publishing:

MARKETPLACE_TOKEN
OPENVSX_TOKEN

If you only use one marketplace, only that secret is required for that workflow.

Triggering publish

Both publishing workflows run on:

Manual workflow dispatch
Git tag pushes matching v*

VS Code Marketplace Publishing

This extension uses the proposed chatProvider API, so Marketplace publishing must explicitly allow it.

Local publish:

npm run package
npm run publish:marketplace -- --pat <MARKETPLACE_TOKEN>

This uses:

npx @vscode/vsce package
npx @vscode/vsce publish --allow-proposed-apis chatProvider --packagePath extension.vsix

Before publishing:

Create a publisher in Visual Studio Marketplace
Make sure package.json.publisher matches that publisher exactly
Create a PAT for Marketplace publishing
Save it as the GitHub secret MARKETPLACE_TOKEN

Open VSX Publishing

Local publish:

npm run package
npm run publish:openvsx

This uses:

npx @vscode/vsce package
npx ovsx publish --packagePath extension.vsix

You still need a valid Open VSX token in your environment as OVSX_PAT.

Omni Chat Provider

ODtian

Omni Chat Provider

Features

How It Works

Requirements

Setup

1. Install the extension

2. Define providers

3. Define models

4. Add or edit a provider

5. Mount the provider into Copilot

Configuration Reference

omnichat.providers

omnichat.models

API-specific model fields

OpenAI

OpenAI Responses

Anthropic

Gemini

Ollama

Retry

Delay

System prompt handling

Commit message generation

Token counting

Build

GitHub Actions

Required secrets

Triggering publish

VS Code Marketplace Publishing

Open VSX Publishing

`omnichat.providers`

`omnichat.models`