[WIP] viking

Floating overlay that shows 3–4 ways to write the snippet you're asking for. Pulls live context from Context7 (language docs) and fff-mcp (codebase grep), plus a screenshot of your screen. Any OpenAI-schema API works.

Setup

npm install

Required env (any OpenAI-compatible provider):

export LLM_API_KEY=sk-...                       # or OPENAI_API_KEY
export LLM_BASE_URL=https://api.openai.com/v1   # optional, defaults to OpenAI
export LLM_MODEL=gpt-4o                         # optional
export VIKING_CWD=/path/to/your/project         # codebase fff greps against

fff-mcp is expected at ~/.local/bin/fff-mcp. Change the path in electron/main/agent/config.ts if yours lives elsewhere. Context7 runs via npx, no install.

Run

npm start

Hotkeys

key	what it does
⌘ I	open the prompt. If results are visible, opens a follow-up referencing the active option.
⌘ S	open settings → provider section (base URL, API key, model). Window-scoped.
⌘ K	open settings → keymaps section. Window-scoped.
⌘ 1..9	switch between rendered options
⌘ C	copy the active option (defers to native copy if you have text selected)
q	hide the overlay (ignored while typing in an input — use esc then)

Settings (provider + keymaps) are saved to <userData>/viking-settings.json and override the defaults from electron/main/agent/config.ts.

Edit hotkeys in electron/main/agent/config.ts.

Where to tweak

• electron/main/agent/config.ts — hotkeys, model, base URL, MCP paths, default language
• electron/main/agent/prompts.ts — system + user prompt
• electron/main/agent/shared-types.ts — zod schemas (Context, Option, LLMResponse)
• electron/main/agent/llm.ts — how context is gathered and the LLM is called