Kimi K3 (Moonshot API) - No Bugs Given

Kimi K3 (Moonshot API) by No Bugs Given

Use Moonshot AI's Kimi models directly inside VS Code Copilot Chat.

This extension is a native Language Model Chat Provider for VS Code. It adds Moonshot AI's Kimi models to the standard GitHub Copilot Chat model picker — no separate chat window, no custom UI, no proxy server. Your prompts go straight from VS Code to Moonshot's official API, authenticated with your own API key.

VS Code  →  this extension  →  Moonshot API (api.moonshot.ai)

That's the whole path. There is no No Bugs Given server, no telemetry, no analytics, and no third party ever sees your key or your code.

Features

Native model picker integration — Kimi models appear alongside your other models in Copilot Chat via VS Code's stable Language Model Chat Provider API (BYOK).
Streaming responses — tokens render in Copilot Chat as they arrive.
Agent mode & tool calling — full tool-call lifecycle: VS Code sends tool definitions, Kimi requests tool calls, VS Code executes them, and the results flow back into the conversation.
Vision / multimodal input — paste or attach images on all three models.
Edit mode, agent mode, ask mode — anywhere VS Code lets you pick a model.
Cancellation — pressing Stop aborts the HTTP request cleanly.
Secure key storage — the API key lives only in VS Code Secret Storage (OS keychain), never in settings.json, workspace files, or logs.
Multi-turn conversations with full history preservation.
Token budgeting via a documented, CJK-aware token estimator.

Supported Models

Model	API ID	Purpose
Kimi K3	`kimi-k3`	Flagship coding, reasoning and agent model — 1M token context
Kimi K2.6	`kimi-k2.6`	Coding, reasoning and general AI work — 256K token context
Kimi K2.5	`kimi-k2.5`	General Kimi model — 256K token context ⚠️ deprecated by Moonshot

Note on Kimi K2.5: Moonshot's model documentation lists kimi-k2.5 as discontinued (August 31, 2026). It is still exposed here for completeness and for compatible gateways, but requests may fail — prefer Kimi K3.

All three models support image input and tool/function calling per Moonshot's documentation. Reasoning traces from Kimi K3 / K2.6 thinking mode are consumed by the extension but not dumped into chat — only the final answer streams.

Getting Started

Quick Start

Install the extension.
Obtain a Moonshot API key from the Moonshot platform console.
Open the Command Palette (Cmd/Ctrl+Shift+P).
Run Kimi: Set API Key.
Paste your API key (input is masked).
Open Copilot Chat.
Open the model picker.
Select Kimi K3, Kimi K2.6, or Kimi K2.5 under Kimi K3 (Moonshot API) - No Bugs Given.
Start chatting — or switch to Agent mode for tool-using workflows.

On first install the extension shows a one-time welcome pointing you at Kimi: Set API Key.

Setting Your API Key

Your Moonshot API key is stored exclusively in VS Code's Secret Storage (backed by macOS Keychain / Windows Credential Manager / Linux libsecret). It is never:

written to settings.json or any file
included in logs or the Output panel
sent anywhere except the configured Moonshot API endpoint

Command	What it does
`Kimi: Set API Key`	Secure, password-masked input; stores the key and refreshes the model picker
`Kimi: Manage API Key`	Menu: Set/Replace key, Test connection, Reset key, Open docs, Get an API key, Endpoint settings
`Kimi: Reset API Key`	Confirmation prompt, then deletes the stored key
`Kimi: Test API Connection`	Lightweight `GET /models` check — validates key + connectivity without spending chat tokens

Kimi: Manage API Key is also wired as the provider's management action in VS Code's language-model management UI.

If no key is configured, the provider stays silent while VS Code enumerates providers in the background, and offers a one-click setup prompt only when you explicitly interact with it.

Selecting a Kimi Model

After setting your key, open Copilot Chat and click the model picker (bottom of the chat input). You'll see:

Kimi K3 (Moonshot API) - No Bugs Given
├── Kimi K3
├── Kimi K2.6
└── Kimi K2.5

Pick one and chat as usual. The selection works in Ask, Edit, and Agent mode.

Agent Mode / Tool Calling

In Agent mode, VS Code passes its registered language-model tools to Kimi. This extension implements the complete tool-call lifecycle over Moonshot's OpenAI-compatible function calling:

VS Code tool definitions → Moonshot tools / tool_choice (required is honored).
Streamed tool-call deltas are reassembled into complete calls (id, name, JSON arguments).
Calls are emitted to VS Code as LanguageModelToolCallParts — never flattened into text.
Tool results return as proper tool-role messages with matching tool_call_ids.

Commands

Command	ID
`Kimi: Set API Key`	`nobugsgivenKimi.setApiKey`
`Kimi: Reset API Key`	`nobugsgivenKimi.resetApiKey`
`Kimi: Manage API Key`	`nobugsgivenKimi.manageApiKey`
`Kimi: Test API Connection`	`nobugsgivenKimi.testApiConnection`

Security

Key storage: VS Code SecretStorage only.
Network: requests go only to the configured API base URL (default https://api.moonshot.ai/v1). The host being contacted is shown in diagnostics and connection tests.
Logging: the Kimi - No Bugs Given output channel never logs keys or Authorization headers; every log line passes through a key-masking sanitizer as defense in depth.
No telemetry. No analytics. No proxy. No No Bugs Given server.

Settings

Setting	Default	Description
`kimi.apiBaseUrl`	`https://api.moonshot.ai/v1`	Advanced: API base URL for development/testing or Moonshot-compatible gateways. Contains no credentials.
`kimi.diagnostics`	`false`	Verbose (sanitized) request/response diagnostics in the output channel.

Troubleshooting

Symptom	What to do
Models don't appear in the picker	Run `Kimi: Test API Connection`. If the key is missing/invalid, run `Kimi: Set API Key`.
"Invalid API key" (401)	Replace the key via `Kimi: Manage API Key` → Replace API Key.
"Rate limit reached" (429)	Wait for the suggested retry window; check your Moonshot plan's rate limits.
Requests time out / network error	Verify connectivity to `api.moonshot.ai`, or check `kimi.apiBaseUrl` if you customized it. The failing host is named in the error.
`kimi-k2.5` returns model-not-found	Moonshot has discontinued K2.5 — switch to Kimi K3.
Want to see what's happening	Enable `kimi.diagnostics` and open the Kimi - No Bugs Given output channel.

Billing: usage is billed by Moonshot AI against your API key, entirely separate from your GitHub Copilot subscription.

Requirements

VS Code 1.104.0 or newer (the first release with the stable Language Model Chat Provider API).
A Moonshot API key.
Access to GitHub Copilot Chat (models contributed through this API are available to Copilot Chat users; Business/Enterprise admins may gate BYOK providers via policy).

Privacy

Prompts, file context, and tool results you send in chat are transmitted directly to Moonshot AI under your own API key and are subject to Moonshot's policies. This extension collects nothing, stores nothing besides the secret key, and phones nowhere else.

Development

npm install
npm run compile   # type-check + build to out/
npm test          # unit tests (no credentials needed)
npm run lint      # eslint
npm run package   # produces a .vsix via @vscode/vsce

Press F5 in VS Code to launch an Extension Development Host.

Optional live integration tests (auto-skipped without a key):

KIMI_INTEGRATION=1 MOONSHOT_API_KEY=sk-... npm test

License

MIT — No Bugs Given (see LICENSE in the repository)