xDev AI
A VS Code extension for managing Claude Code, Codex, and Gemini CLI accounts, chatting with AI providers, and coordinating tasks across API profiles.
Installation
Use VS Code 1.95 or later, or a code-server release that includes VS Code 1.95 or later. Search Extensions for xdev-asia.xdev-ai.
- VS Code Desktop. Install from the VS Code Marketplace.
- VS Code Remote. Open your remote workspace and install xDev AI under the remote host in the Extensions view.
- code-server. Install from Open VSX through the Extensions view or run
code-server --install-extension xdev-asia.xdev-ai on the server.
For manual installation, download the .vsix from GitHub Releases and run Extensions: Install from VSIX… in your workspace. On code-server you can also run code-server --install-extension <path-to-vsix>.
See what's new under Extensions → xDev AI → Changelog, or read the release history on GitHub.
The extension runs on the machine that hosts the workspace. For remote workspaces, that machine needs access to the provider's API and resolves any localhost addresses in custom endpoints. VS Code Remote · code-server installation
The extension ID changed to xdev-asia.xdev-ai in v0.2.1. If you installed the v0.2.0 preview (xdev.xdev-ai), disable it before using the new version, then add your API profiles and sign in to each Claude Code account again in the new extension.
Manage AI accounts
Open xDev AI in the Activity Bar. The sidebar contains Terminal Accounts and API Profiles. To move a view to the right, run View: Move View from the Command Palette, select the view, and choose New Secondary Side Bar Entry.
- Click Add AI Account (+) next to Terminal Accounts. Choose Claude Code, Codex, Gemini CLI, or API Profile. For a terminal account, give the profile a name and complete sign-in with the provider. Gemini opens its authentication menu in the terminal. Codex uses device code sign-in on remote hosts and code-server, which requires device code sign-in to be enabled for the account.
- Select an account to open its terminal. Use the xDev AI status bar item, Open AI Terminal on the Terminal toolbar, or xDev AI: Choose Account in the terminal profile menu to pick an account without opening the sidebar.
- Right-click a terminal account to rename it or sign in again. Click Check AI Usage next to a terminal account to open its session and see the provider's usage command.
- In API Profiles, click + to add a profile. Select a profile to open xDev Chat with it, or right-click it to make it active or edit it.
- Click the trash icon next to a terminal account or API profile, or use its right-click Delete command. Deleting a terminal account closes its matching terminals in the current window and removes the profile directory, including saved sign-in data and CLI history. Deleting an API profile removes its saved key from xDev AI.
Install each CLI on the workspace host. If xDev AI cannot find one, enter its absolute executable path in User Settings or Remote Settings.
| Terminal |
Settings key |
Usage command |
| Claude Code |
xdevAi.claudeExecutable |
/usage |
| Codex |
xdevAi.codexExecutable |
/status |
| Gemini CLI |
xdevAi.geminiExecutable |
/stats model |
Each account has its own configuration directory on the workspace host. The CLIs manage their sign-in, credentials, and history there. Terminal and API profiles are configured separately. Claude Code authentication · Codex authentication · Gemini CLI authentication
View account quota
Hover over xDev AI in the VS Code or code-server status bar. The hover lists terminal profiles with the plan name, percentage used, and reset time where available. Click Refresh quota to update the hover or open a profile's terminal for its full usage view. API key profiles are managed separately and have no subscription quota in the hover.
- Codex. ChatGPT profiles show rate-limit windows from Codex App Server. The extension refreshes these readings periodically in trusted workspaces. Codex rate limits
- Claude Code. Open a new Claude session from xDev AI in a trusted workspace. On supported plans, the hover shows the latest quota sample after Claude replies. This uses the session's status line and requires Node.js on the workspace host's
PATH. If you already use a custom Claude status line, open /usage in the terminal. Claude Code status line
- Gemini CLI. Use the account's hover link to open its terminal and enter
/stats model for model usage and any quota the CLI displays. Gemini CLI commands
Open a terminal with bypass permissions
Right-click a signed-in terminal account and choose Open AI Terminal with Bypass Permissions, or choose xDev AI: Bypass Permissions from the terminal profile menu. Confirm the mode for the session. Use a trusted workspace, and use a separate VM or container when running Codex without its sandbox.
Claude Code skips permission prompts, Codex skips approval prompts and disables its sandbox, and Gemini CLI uses YOLO approval mode. The option applies to the new terminal and uses the selected account's existing sign-in. Normal terminal and sign-in actions use the CLI defaults. Managed provider policies may restrict these modes. Claude Code options · Codex CLI options · Gemini CLI options
Sign in to Claude Code from code-server
- In Terminal Accounts, right-click your Claude Code account and choose Sign In to Terminal Account. Keep the xDev AI terminal open.
- Open the URL printed by Claude Code in your browser. If Claude Code shows Paste code here, use the URL directly above that prompt. After authorizing in the browser, return to the original terminal and paste the code, then press Enter.
- Once Claude Code confirms sign-in, xDev AI opens the Claude session in the same terminal and profile. If sign-in exits with an error, the terminal stays open with the original CLI output and retry and status commands.
On the code-server host, check claude --version and update Claude Code if your version lacks the manual code prompt. Run claude update if you installed Claude Code with its native installer. Claude Code remote OAuth troubleshooting · Claude Code updates
Coordinate work with Squad
Open xDev Squad in the bottom panel, click Open Squad on the Terminal toolbar, or run xDev AI: Open Squad from the Command Palette.
Use one chat for the squad
- Open Roles, select Lead and any workers you want to use, and assign an API profile to each. Save each assignment.
- Open Chat and send a request to Lead. If Lead proposes tasks, review the briefs and profiles, then click Approve and run or Discard plan.
- Follow the results in the same chat. Use
@Implement, @Review, or @Tests to send a message directly to a role. Press Stop run to cancel the current run and skip any remaining tasks.
- To use an assigned terminal profile, select its role in Chat and click Open terminal.
The shared chat uses text APIs for planning, implementation proposals, reviews, and test suggestions. This mode does not read or edit repository files or run tests. Lead receives recent chat entries. Approved workers receive the original request and their assigned brief. Review and Tests also receive the implementation proposal when the plan includes Implement. Use profiles that are approved for the information you share.
The most recent 60 chat entries and the pending plan are stored in local VS Code workspace state. Reloading VS Code preserves pending approval and completed replies. An interrupted run needs a new request to continue.
Use Squad in GitHub Copilot Chat
- Assign API profiles in xDev Squad → Roles. Open the Chat view in VS Code and select a Copilot agent session with extension tools enabled.
- In the chat input, type
# and select Ask xDev Squad (xdevSquadAsk). Try “Ask xDev Squad to plan an API migration.” To talk directly to a role, try “Ask xDev Squad's Review role to check backward compatibility.”
- Review the roles, profiles, and briefs in the Copilot reply. Ask Copilot to run or discard the pending Squad plan, using Manage xDev Squad Plan (
xdevSquadPlan) if needed. Choose Approve and run in the VS Code dialog to send the tasks. The dialog is required even when Copilot's tool approvals are bypassed. Results return to the same Copilot conversation.
- If the tools are missing, run Chat: Open Customizations, open Tools, and check that the xDev AI tools are enabled for the selected Copilot session.
Both interfaces use the same workspace profiles, pending plan, and Squad context. Copilot receives the responses to its Squad tool calls. Each assigned API profile receives the request and context described above. On code-server, open xDev Squad → Chat for the same workflow. VS Code extension tools
VS Code manages the chat session's agent harness and model picker. To change the native harness, start a new chat and choose a session target. xDev terminal profiles open in their own terminals. Use the API profiles assigned under Roles for xDev Squad tool calls in a Copilot session. VS Code agent harnesses
Run an AI SDLC workflow
- Open xDev Squad → Workflow or run xDev AI: Open SDLC Workflow. Enter the project brief to start Plan → Implement → Review → Test → Release.
- For an AI draft, assign the stage's role an API profile in Roles, then click Draft with AI in Workflow. The draft also appears in the shared Chat. In Copilot Chat,
#xdevSdlc can show the current stage or request a draft.
- Edit the draft or enter your own result, then click Approve stage. For terminal accounts, use Copy stage brief and open the role's terminal. Check approved results in the same workflow after reloading the workspace.
The Release stage drafts a Markdown changelog entry from approved results. Review it in Workflow, then copy the entry to your project's changelog when preparing the release.
Each AI draft sends the project brief, excerpts of earlier approved results, and enabled workspace skills to the assigned API provider. Each stage produces a text result. Run code changes, tests, and releases in your editor, CLI, or CI. Open Workflow to approve results from either chat interface. Each workspace keeps one current workflow, with a reset action to start another.
If VS Code reports that the workflow command is missing after an update, check Extensions for the installed xDev AI version on the workspace host, then run Developer: Reload Window. To diagnose a failed activation, open Output and select Log (Extension Host) or Log (Remote Extension Host).
- Under xDev Squad → Skills & Tools, add, edit, disable, or delete workspace skills. A skill is a named set of instructions that xDev AI includes in SDLC drafting when enabled.
- Use the tool switches to control Copilot access to Ask xDev Squad, Manage xDev Squad Plan, and xDev SDLC Workflow in this workspace. The SDLC switch also controls MCP access to the workflow. In supported VS Code versions, use the panel shortcuts to open native Agent Skills and Tools & MCP management. VS Code agent customizations
Control the workflow through MCP
- In a trusted workspace, open xDev Squad → Skills & Tools → Manage MCP or run xDev AI: Manage MCP. Choose Start MCP.
- In VS Code builds with native MCP support, run MCP: List Servers and start xDev AI if needed. Select its tools in Copilot Agent Chat. VS Code MCP server management
- For a terminal client, reopen Manage MCP and choose Copy Client Configuration, then choose Claude Code, Codex, or Gemini CLI. Merge the snippet into the client's private config on the workspace host. For an xDev terminal profile, use
$CODEX_HOME/config.toml for Codex or $GEMINI_CLI_HOME/.gemini/settings.json for Gemini CLI. In Claude Code, you can use claude mcp add-json xdev_ai '<server definition>' with the mcpServers.xdev_ai object from the snippet. Restart the client or refresh its MCP connections. Claude Code configuration · Codex configuration · Gemini CLI configuration
- Ask your agent to “Use xDev AI to show the current SDLC stage and submit a draft for review.” The draft appears in Workflow. Approve it there, or ask the agent to request approval in a VS Code dialog.
The MCP server exposes workflow status, start, stage context, draft submission, approval requests, and account aliases. It shares the same workspace state as the Squad panel. A confirmed approval advances the current stage. The server exposes no account credentials or shell execution tools. Stage context includes the project brief, excerpts of approved results, and enabled skills, so only connect clients you trust with that context.
MCP listens on a loopback address on the workspace host, including on code-server. The client must run on that same host. Keep the copied configuration private and out of source control. Stop MCP ends access and invalidates the token. Reloading the window stops the server. A new start issues a new address and token.
Manage a role manually
- Open Roles in the xDev Squad panel.
- Select Lead, Implement, Review, or Tests. Assign an API profile or terminal account, enter a task brief, and click Save task. Each role has one current task, and you can prepare a brief before assigning a profile.
- For a terminal account, use Copy brief and Open terminal, then paste the brief into that provider's session. Check usage opens the assigned terminal and shows its usage command.
- For an API profile, click Send to API to send the saved brief to the selected model. Read the reply under Latest API response. Include any context you want the model to see in the brief.
- Update task status as work progresses. For Blocked, save a reason to show it under Needs attention. Use Hand off to another role to leave a note for that role in Recent activity.
The board keeps assignments, briefs, the latest API response per role, and the 30 most recent activity entries in local VS Code workspace state. Status changes and handoffs are manual.
Chat with AI providers
- Run xDev AI: Open Chat from the Command Palette.
- Run xDev AI: Add API Profile and choose a provider. Give the profile a recognizable name, then enter a model ID and API key. For an OpenAI-compatible endpoint, also enter its base URL, such as
http://127.0.0.1:11434/v1 for Ollama. You can leave the key blank if the endpoint accepts unauthenticated requests.
- Send a message with
Ctrl+Enter or Cmd+Enter. Use the profile menu or xDev AI: Switch API Profile to change accounts. Each profile keeps its most recent 12 exchanges while the chat tab is open.
- Run xDev AI: Manage API Profiles to rename or remove a profile, change its model, or replace its API key. The Đoạn chat mới (New chat) button clears the selected profile's conversation.
The chat tab authenticates with API keys. Get a key and model ID from your provider. OpenAI · Anthropic · Google AI Studio
VS Code SecretStorage holds the API profiles. You may need to add your profiles again on a different VS Code installation or server. The chat sends your message and the selected profile's conversation history to its provider. Custom base URLs must use HTTPS or loopback HTTP.
Use the TypeScript library
The chat client also runs independently in Node.js 20.3 or later. Run npm ci and npm pack from the repository root to create an xdev-ai archive. Install that archive in your Node.js project, then set AI_MODEL and OPENAI_API_KEY for this example.
import { chat } from 'xdev-ai'
const answer = await chat(
{ provider: 'openai', model: process.env.AI_MODEL!, apiKey: process.env.OPENAI_API_KEY },
[{ role: 'user', content: 'Explain how to use AbortSignal.' }]
)
console.log(answer)
Call xDev AI from another VS Code extension
Add "extensionDependencies": ["xdev-asia.xdev-ai"] to the calling extension's manifest and install xDev AI in the same extension host. listAccounts() returns each profile's ID, name, provider, and model. Pass { accountId } to select a profile. When you omit it, chat() uses the active profile.
import * as vscode from 'vscode'
const extension = vscode.extensions.getExtension('xdev-asia.xdev-ai')
if (!extension) throw new Error('Install xDev AI to continue.')
const ai = await extension.activate()
const [account] = await ai.listAccounts()
if (account) {
const answer = await ai.chat(
[{ role: 'user', content: 'Describe how a VS Code extension host works.' }],
{ accountId: account.id }
)
console.log(answer)
}
Build and test
npm ci
npm test
npm run package
Release an extension version
GitHub Actions tests and packages every pull request and push to main. Add the version's notes to CHANGELOG.md under a heading such as ## 0.6.1 - 2026-11-19. Push a version tag such as v0.6.1 that matches package.json. The workflow uses that changelog entry for the GitHub Release, attaches the tested VSIX, and publishes the same VSIX to both registries.
- VS Code Marketplace. Create or select the
xdev-asia publisher and add a Trusted Publishing policy in Manage Publishers & Extensions. Set the repository owner to xdev-asia, repository to vscode-ai-profiles, and workflow to publish.yml. Leave the environment blank.
- Open VSX (code-server). Declare the extension license in
package.json and include the license file. Sign the Open VSX Publisher Agreement and create the xdev-asia namespace. Publish the first VSIX through the Open VSX upload page, or save an Open VSX access token as OVSX_PAT in the repository's Actions secrets to publish from CI. After the first version is active, the namespace owner can register the same owner, repository, and workflow for xdev-asia.xdev-ai under Trusted Publishers. Remove OVSX_PAT after registration to use short-lived credentials for subsequent releases. Open VSX trusted publishing
License
xDev AI is licensed under the Apache License 2.0.