EveryAPI · VS Code AI Assistant

One key, every model. Use the EveryAPI gateway (240+ models — Claude, GPT, Gemini, DeepSeek, …) two ways inside your editor:

A built-in AI assistant — its own chat sidebar, inline code completion, workspace-aware context, and a status-bar HUD. Works in VS Code and its forks (Cursor, VSCodium, Windsurf, …).
A Copilot Chat model provider — EveryAPI models also show up in VS Code's built-in Copilot Chat picker.

Both surfaces share a single key stored in your OS keychain.

The assistant

Open the EveryAPI icon in the activity bar:

Chat sidebar — pick any model from /v1/models (with live pricing), stream replies, run slash commands (/explain /fix /test /docs). Conversation persists per workspace.
Workspace context — chips above the input attach your code to each turn so the model can actually see it:
- the active file (or just the selection),
- @-attached files via the + file picker,
- a Workspace overview (file tree + package.json + README excerpt) for project-level questions.
Inline completion — press Alt+\ in the editor for a ghost-text suggestion from your selected model. Manual-trigger only, so it never spends your balance on every keystroke.
Status-bar HUD — current model + remaining balance, with today's tokens/spend in the tooltip.
Onboarding & settings — connect with a key, pick a default model, set a low-balance alert, and see usage (today / 7-day / top models).

Copilot Chat provider

If you use GitHub Copilot Chat, EveryAPI also registers as a Language Model Chat Provider (VS Code v1.104+): open Copilot Chat's model picker → "Manage Models" → EveryAPI, and its models appear alongside Copilot's own.

⚠ Copilot Business / Enterprise: third-party model providers are gated on a tenant admin enabling "Models from other providers" in your Copilot policy. Individual plans have it on by default. (This only affects the Copilot-Chat surface — the built-in assistant works regardless.)

Setup

Install from the Marketplace / OpenVSX (search "EveryAPI") or code --install-extension everyapi-ai.everyapi-vscode.
Open the EveryAPI sidebar → paste your sk-everyapi-… key (created at https://app.everyapi.ai) → pick a default model.
Chat, or hit Alt+\ in code for an inline completion.

The key is stored via vscode.SecretStorage — your OS keychain, never settings.json, and never the webview.

Commands & keybindings

Command	Default key	What it does
EveryAPI: Set API Key…	—	Connect / replace the key (also the Copilot "Manage Models" entry)
EveryAPI: Clear Stored API Key	—	Disconnect
EveryAPI: Refresh Model List	—	Re-fetch models, pricing, and balance
EveryAPI: New Chat / Clear Conversation	—	(also in the sidebar title bar)
EveryAPI: Open Settings	—	The in-sidebar settings panel
EveryAPI: Inline Completion	Alt+\	Ghost-text completion at the cursor

Configuration

Setting	Default	What it does
`everyapi.baseUrl`	`https://api.everyapi.ai/v1`	Gateway base URL. Override for staging / self-hosted.
`everyapi.defaultMaxOutputTokens`	`4096`	Fallback when the gateway doesn't expose a per-model output limit.
`everyapi.toolCalling`	`true`	Report models as tool-capable so they're selectable in Copilot Chat's Agent/Edit modes. Turn off if a model errors on tool definitions.

Architecture

apps/vscode/
├── package.json                # activity-bar view + webview, commands, keybindings, config,
│                                 languageModelChatProviders
├── esbuild.mjs                 # bundles src/ → dist/extension.js (CJS, `vscode` external)
├── assets/icon.png             # EveryAPI brand mark
├── media/                      # webview assets (served, not bundled)
│   ├── chat.css / chat.js      # the sidebar UI (vanilla JS; renders host snapshots)
│   └── everyapi-activity.svg   # activity-bar icon
└── src/
    ├── extension.ts            # activate(): wires the assistant + Copilot provider + commands
    ├── store.ts                # host-owned state: key (secrets), models/pricing, wallet,
    │                             usage, conversation (globalState), settings
    ├── chatView.ts             # WebviewViewProvider: streams turns, owns context state
    ├── context.ts              # workspace context: active file/selection, @-files, overview
    ├── statusBar.ts            # model + balance HUD
    ├── inlineComplete.ts       # manual-trigger InlineCompletionItemProvider
    └── provider.ts             # EveryApiProvider — the Copilot Chat model provider

The gateway client (/v1/models, streaming /v1/chat/completions, wallet/usage/pricing) is the shared @everyapi-ai/gateway workspace package, bundled in.

Development

cd apps/vscode
bun install        # via the monorepo workspace
bun run typecheck  # tsc --noEmit
bun run build      # tsc --noEmit && esbuild → dist/extension.js
bun run dev        # esbuild watch mode
bun run package    # → everyapi-vscode.vsix

Press F5 from apps/vscode/ to launch a fresh VS Code window with the extension loaded.

Releases

Merges to main that touch apps/vscode/** auto-release via CI: the version bump level is derived from the conventional-commit messages (feat → minor, fix/chore → patch, type!:/BREAKING → major), then the .vsix is built and published to both the VS Code Marketplace and OpenVSX. Publishing on both reaches the whole ecosystem — every VS Code fork that isn't Microsoft's pulls from OpenVSX.

Roadmap

[x] Built-in assistant: chat sidebar, inline completion, workspace context, status-bar HUD.
[x] Marketplace + OpenVSX publish (auto from CI).
[x] Per-model capability flags — image input from input_modalities; tool calling via everyapi.toolCalling.
[ ] Cross-surface shared conversation (VS Code ↔ Raycast ↔ menubar) — needs a backend conversation store; today the conversation is local per workspace.
[ ] @-mention autocomplete inside the input (today the picker is a QuickPick).

License

MIT — see LICENSE in the repository root.