GLM Models for GitHub Copilot Chat

Bring Z.AI's GLM models into GitHub Copilot Chat with your own API key (BYOK) — GLM-5.2, the 1M-context flagship built for long-horizon coding, plus a curated lineup (GLM-5.1, GLM-5, GLM-4.7, GLM-4.5 Air). No new sidebar or chat UI: the models appear in the picker you already use, with agent mode, tool calling, and thinking through Copilot's native provider path.

Unofficial, community-built extension. Not affiliated with, endorsed by, or sponsored by Zhipu AI, Z.AI, GitHub, or Microsoft. "GLM", "Copilot", and "Visual Studio Code" are trademarks of their respective owners. You bring your own GLM API key and pay your own usage.

Features

• GLM-5.2 flagship, right in the picker. A 1M-token context window, step-by-step thinking that streams live into Copilot's native thinking UI, and a per-model Thinking Effort control (None / High / Max) — pick None for faster simple edits. Switch models mid-chat without losing history.

• Powers Copilot up, doesn't replace it. GLM models appear alongside GPT and Claude. Because the extension uses Copilot's native Language Model Provider API, agent mode and tool calling keep working as usual.

• Dual API. Use your GLM Coding Plan subscription or the pay-as-you-go Standard API — each available International (z.ai) or Mainland China (bigmodel.cn). See Coding Plan vs Standard API.

• Live usage & balance tracking. A status-bar readout plus a full GLM: Show Usage Details panel. Coding Plan shows session (5-hour) and weekly (7-day) token limits, monthly web searches, and reset countdowns. Standard API shows cash balance and token resource packages. The status bar turns red when quota hits 100% or balance reaches 0. Works on both International (z.ai) and Mainland China (bigmodel.cn) endpoints for both API modes; no baseUrl override.

• Keys stay in your OS keychain by default. GLM: Set API Key stores your key via VS Code SecretStorage (macOS, Windows, Linux). The extension also honors a settings fallback for CI or automation, so do not put real keys in workspace settings.
• Zero core runtime dependencies. Pure VS Code API and Node.js built-ins — the core extension needs no Python, Docker, or local server. (The optional GLM Vision integration installs an integrity-locked local npm package, so it requires Node.js 18+ with npm.)
• Add any model. Newly released, fine-tuned, or proxy-hosted GLM models via glm-copilot.customModels.
• See images, even on text-only models. Install the official GLM Vision MCP server with one click and turn on image input: pasted screenshots are analyzed by GLM Vision (GLM-4.6V) and injected as text — code, logs, errors, and UI included — so any GLM model can reason about them. Agent mode also gets the full vision tool set. See Image input (vision).

Getting Started

Prerequisites

• VS Code 1.116 or later
• An active GitHub Copilot subscription (Free, Pro, or Enterprise)
• A GLM API key from z.ai or bigmodel.cn, or a GLM Coding Plan subscription

Installation

Install from the VS Code Marketplace, or search "GLM Models for GitHub Copilot Chat" in the Extensions panel (Cmd/Ctrl + Shift + X):

code --install-extension yijiazhen-qi.glm-for-github-copilot-chat

Upgrading from an older Marketplace listing? Uninstall it first — the old and new IDs can both register the same glm-copilot.* commands and settings:

code --uninstall-extension YijiazhenQi.glm-for-copilot-chat
code --install-extension yijiazhen-qi.glm-for-github-copilot-chat

Usage

1. GLM: Set API Key (Command Palette, Cmd/Ctrl + Shift + P) → paste your key. GLM key format is {id}.{secret}.
2. (Optional) GLM: Open Settings to choose your API mode and region.
3. Open Copilot Chat, pick a GLM model (e.g. GLM-5.2, the flagship), and start chatting.

Use GLM: Set API Key or GLM: Clear API Key to update or remove the key later.

Models

Model	Tier	Context	Max Output	Available on	Tools	Thinking
GLM-5.2	Flagship	1M	128K	Coding Plan + Standard	Yes	Yes (effort)
GLM-5.1	Prior flagship	200K	128K	Standard only	Yes	Yes
GLM-5	Prior flagship	200K	128K	Standard only	Yes	Yes
GLM-4.7	Fast coding	200K	128K	Coding Plan + Standard	Yes	Yes
GLM-4.5 Air	Lightweight	128K	96K	Coding Plan + Standard	Yes	Yes

The picker shows only the models your selected API Mode can serve, so you never pick one your plan can't use. GLM-5.2, GLM-4.7, and GLM-4.5 Air work on both; GLM-5 and GLM-5.1 are Standard-only. Need a model not listed — newer (GLM-5-Turbo), older (GLM-4.6), or proxy-hosted? Add it with glm-copilot.customModels.

Settings

Setting	Default	Description
glm-copilot.apiMode	coding-plan	Which GLM API to use: coding-plan or standard. See below.
glm-copilot.region	international	Server region for both chat API modes and GLM Vision: international (z.ai) or china (bigmodel.cn). A custom baseUrl overrides it for chat only.
glm-copilot.baseUrl	(empty)	Override the chat API base URL. Overrides apiMode and region for chat. GLM Vision still uses the official endpoint selected by region and requires a separate Vision key.
glm-copilot.maxTokens	0	Maximum output tokens per request. 0 means no explicit limit (uses API default).
glm-copilot.maxRetries	3	Automatic retries for transient chat failures (HTTP 429/5xx), not counting the first attempt. 0 disables retries (fail fast). Range 0–10. Backoff honors the server's Retry-After.
glm-copilot.thinking	enabled	Step-by-step reasoning: enabled (higher quality) or disabled (faster). Applies to thinking-capable models without a per-model Thinking Effort picker; GLM-5.2 uses None / High / Max in the model picker.
glm-copilot.visionEnabled	false	Accept pasted/attached images in chat when the verified GLM Vision package is installed. GLM Vision analyzes them first so text-only models can reason about them. If the server is unavailable, the reply shows an analysis-failure notice and continues without the image. See Image input (vision).
glm-copilot.visionPrompt	(empty)	Instruction sent to GLM Vision when analyzing images. Empty uses the built-in prompt (verbatim text/code transcription, UI/diagram/chart description, no speculation). Changing it re-analyzes images on the next turn.
glm-copilot.customModels	[]	Add your own models. Array of model id strings or objects: { id, name?, maxInputTokens?, maxOutputTokens?, toolCalling?, thinking? }.
glm-copilot.modelIdOverrides	{}	Remap a built-in model's API id (keys = picker id, values = id sent to the API). Use for regional endpoints or proxies with different names.
glm-copilot.debugLogging	false	Write verbose debug logs to the GLM output channel. View with GLM: Show Logs.
glm-copilot.usageRefreshIntervalMinutes	5	How often (in minutes) to refresh the GLM usage status bar. Minimum 1. Shows Coding Plan quota or Standard API balance for both z.ai and bigmodel.cn regions. Only when no baseUrl override.
glm-copilot.showUsageStatusBar	true	Show the GLM usage status-bar item. Coding Plan quota or Standard API balance, both z.ai and bigmodel.cn regions. Only when no baseUrl override.

Coding Plan vs Standard API

Coding Plan requires a GLM Coding Plan subscription — best for teams or high-volume coding. Standard API is pay-as-you-go via the GLM Open Platform. The endpoint follows your region either way.

Mode	Region	Endpoint	Get a key
Coding Plan	International	https://api.z.ai/api/coding/paas/v4	z.ai/manage-apikey/subscription
Coding Plan	Mainland China	https://open.bigmodel.cn/api/coding/paas/v4	bigmodel.cn/coding-plan
Standard	International	https://api.z.ai/api/paas/v4	z.ai/manage-apikey/apikey-list
Standard	Mainland China	https://open.bigmodel.cn/api/paas/v4	open.bigmodel.cn

Full API documentation: docs.z.ai.

Usage details rely on regional usage endpoints (api.z.ai for International, open.bigmodel.cn for Mainland China) that are not part of the public chat-completions API. If those endpoints are unavailable or change, the extension degrades to a status message instead of blocking chat.

Image input (vision)

Vision is powered by the official Z.AI Vision MCP server (@z_ai/mcp-server, GLM-4.6V). Setup is two steps, and nothing is installed without your say-so:

1. Install the server — run GLM: Install GLM Vision MCP Server (or the Install button in the Getting Started walkthrough). Before anything is downloaded, a modal confirmation explains that the third-party package runs locally, receives the API key selected for Vision and attached-image paths, and sends image content to Z.AI or BigModel. After consent, the flow checks the required key and Node.js/npm (Node.js 18+, 22+ recommended), then installs @z_ai/mcp-server@0.1.4 under the extension's global storage with npm ci. A bundled lockfile pins the complete dependency graph and its SHA-512 integrity hashes, and npm lifecycle scripts are disabled. VS Code launches the fixed local entry point — never a later package fetched by npx — with Z_AI_MODE set from your region (ZAI for International, ZHIPU for Mainland China). A dedicated Vision key in VS Code SecretStorage takes priority. Without one, Vision reuses your stored GLM key only when glm-copilot.baseUrl is empty. When a custom Base URL is active at server start, Vision requires a separate Z.AI or BigModel key and does not select the proxy credential. You can replace the separate key with GLM: Set Vision API Key. Base URL and key changes prompt you to restart the server so the new credential selection takes effect. GLM: Uninstall GLM Vision MCP Server disables the registration and image input, removes the dedicated Vision key and temp images, and then removes the local package. Vision stays disabled and reports any cleanup that could not finish.
2. Turn on image input — after installation, GLM: Toggle Vision for Chat Models appears in the Command Palette. Turn it on while the server is healthy, or set glm-copilot.visionEnabled. Every GLM model in the picker then accepts image input. A temporary server outage does not remove image support from the model picker.

Each time image input changes from off to on, the extension waits until VS Code can see GLM Vision > analyze_image, then shows approval guidance. To stop the repeated confirmation before each analysis:

• Select Manage Tool Approval in the notice.
• In the workspace Tool Approval manager, find GLM Vision > analyze_image.
• Enable without approval only for that tool.

The extension does not change approval settings or approve tools for you. Organization policy can still require confirmation.

With image input on, pasted or attached images just work — even for text-only chat models:

• Analyzed by GLM Vision — each image is written to a local temp file and analyzed through the server's analyze_image tool on the selected Vision key, and the analysis is injected into the request as text before it reaches your selected chat model. The instruction is configurable via GLM: Edit GLM Vision Prompt or glm-copilot.visionPrompt; changing it re-analyzes images on the next turn.
• Cached — analyses are content-addressed and cached in memory for the session (never written to extension storage), so conversation history is not re-analyzed every turn.
• Validated — images larger than 5 MB, or outside png / jpeg, are skipped with a clear notice before any tool call.
• Visible progress, graceful failure — a status line streams into the thinking block while analysis runs. If analysis fails or the server stops, the model keeps image input enabled, but the reply continues without the image and opens with a short "image analysis failed" notice instead of erroring out. Turning vision off or uninstalling the package removes image input from the model picker.

In agent mode the server's full tool set is also available directly to the agent — UI-to-code, screenshot OCR, error-screenshot diagnosis, diagram/chart analysis, UI diffing, and video understanding. For images on disk, the agent calls these tools itself with the file path; no extra setup needed.

Commands

Command	Description
GLM: Set API Key	Set or update your GLM API key
GLM: Get API Key	Open the key management page for your selected API mode
GLM: Clear API Key	Remove your stored API key
GLM: Open Settings	Open the extension settings
GLM: Show Logs	Open the GLM output channel
GLM: Refresh Usage	Refresh GLM usage/balance now (Coding Plan quota or Standard API balance; both z.ai and bigmodel.cn)
GLM: Show Usage Details	Open the GLM usage panel (Coding Plan quota or Standard API balance; both z.ai and bigmodel.cn)
GLM: Install GLM Vision MCP Server	Review the impact, then install the integrity-locked local package. Visible when not installed
GLM: Uninstall GLM Vision MCP Server	Remove the GLM Vision MCP server and revert models to text-only. Visible when installed
GLM: Set Vision API Key	Store or replace the key used only by the official GLM Vision server. Visible when installed
GLM: Edit GLM Vision Prompt	Open the multiline setting for the image-analysis instruction. Visible when installed
GLM: Toggle Vision for Chat Models	Turn image input for chat models on/off. Visible while GLM Vision is installed

Frequently asked questions

Is this an official GLM or GitHub extension?

No. Unofficial, community-built, open source — not affiliated with Zhipu AI, Z.AI, GitHub, or Microsoft. It just lets you use your own GLM API key inside Copilot Chat.

Do I still need a GitHub Copilot subscription?

Yes. This adds GLM models to Copilot Chat; it doesn't replace Copilot. You need an active Copilot subscription (Free, Pro, or Enterprise) and your own GLM API key.

Where does my API key go?

GLM: Set API Key stores the chat key in VS Code SecretStorage (the OS keychain on macOS, Windows, Linux), and chat requests send it only to your configured GLM endpoint over HTTPS. A settings fallback exists for CI or automation, but avoid putting real keys in settings.json, especially workspace settings that could be committed. GLM Vision uses a separate SecretStorage key when you run GLM: Set Vision API Key. If glm-copilot.baseUrl is set when Vision starts, Vision requires the separate key and does not select the proxy credential. Restart Vision after a Base URL or key change.

Can I use a proxy or self-hosted endpoint?

Yes, for chat. Set glm-copilot.baseUrl to any OpenAI-compatible endpoint; it overrides apiMode and region for chat. GLM Vision still uses the official endpoint selected by region. When Vision starts with the custom Base URL active, it requires a separate Vision key and does not select the proxy credential.

Is GLM-4.6 still supported?

GLM-4.6 was superseded by GLM-4.7 and the GLM-5 series in v0.2.0. Add it with glm-copilot.customModels if your account still serves it.

Contributing

Contributions welcome — see the contributing guide and Code of Conduct. All PRs require code-owner review and are never auto-merged.

• Bug? Open a report · Feature? Request it
• Help? Support or Discussions · Security? Policy

Changelog

See CHANGELOG.md.

License

MIT © GLM Models for GitHub Copilot Chat contributors