Bring your own key

Before you start

Ollama: a free model on your own machine

No key, no account, nothing leaves your computer. Good for drafts, rewrites and the résumé tools; a 7B to 8B model on 16 GB of RAM is comfortable.

1. Install Ollama. Settings → Local AI setup (Ollama) shows the one-line command for your system, with a Copy button:

   winget install -e --id Ollama.Ollama        # Windows
   brew install ollama                          # macOS
   curl -fsSL https://ollama.com/install.sh | sh  # Linux

2. Pull a model. In the same card press Install beside a model (it runs ollama pull for you), or in a terminal: ollama pull llama3.2. The card shows Ollama running ✓ once the server answers on http://localhost:11434.
3. Select it. In AI provider choose Ollama (local, no key). The server URL is filled in; leave it unless you run Ollama elsewhere. Press Use beside the model, or Use this in the model manager.
4. Optional. Set System RAM so the resource warnings are right, and press Load into VRAM to warm the model before a long session.

LM Studio

1. In LM Studio, load a model and start the local server (the OpenAI-compatible one, on port 1234 by default).
2. In AI provider choose LM Studio (local, no key). The URL http://localhost:1234/v1 is filled in.
3. Vulture listens for the server and shows Server live at … · N models loaded with the loaded models as chips. Click a chip to use it, or press ↻ Check after loading a new one.

Cloud providers with your own key

Same three steps for all of them: choose the provider, paste the key, pick a model from the preset list or type any model id. The get a key ↗ link beside the key field opens the provider's key page; model list ↗ opens its model documentation.

1. Provider. Pick it from the dropdown under Cloud.
2. API key. Paste it. The field is masked and the key is stored only in this profile.
3. Model. Choose a preset, or pick Custom / other… and type the id exactly as the provider spells it.
4. Azure only. Also fill Endpoint (https://<resource>.openai.azure.com), Deployment name and API version (for example 2024-10-21).
5. Keep it. Press ＋ Save current model so the key and model live in your library and the form clears. You can save several and switch from the console.

Custom and OpenAI-compatible endpoints

Anything that speaks POST /chat/completions or the Responses API: vLLM, LiteLLM, a company proxy, a self-hosted gateway, or a provider not in the list.

1. Choose Custom (OpenAI-compatible).
2. Base URL: the root that owns /chat/completions, for example http://localhost:8000/v1 or https://gateway.example.com/v1.
3. API type: Chat Completions for almost everything; Responses API only when the endpoint implements /responses.
4. Auth header: Authorization: Bearer by default; some gateways want api-key or x-api-key instead.
5. API key is optional here; leave it blank when the endpoint needs none.
6. Fetch available models under Multi-model endpoint asks the server for its list. Tick the ones you want, name the endpoint, and ＋ Add N models to library. They are saved as one group sharing the key and URL, so you can switch between them from the console.

Importing a VS Code style config

Open Config methods & import at the bottom of the card, paste a config and press Apply config. Groups with one key and several models import into your saved models. A ${apiKey} placeholder is accepted; add the real key afterwards with add key on the group.

{
  "name": "My gateway",
  "vendor": "custom",
  "apiKey": "${apiKey}",
  "apiType": "chat",
  "models": [
    { "id": "qwen2.5-7b-instruct", "name": "Qwen 2.5 7B", "url": "https://gateway.example.com/v1/chat/completions" },
    { "id": "llama-3.3-70b", "name": "Llama 3.3 70B", "url": "https://gateway.example.com/v1/chat/completions",
      "requestHeaders": { "x-api-key": "${apiKey}" } }
  ]
}

The same panel shows your current setup as Vulture JSON, cURL or Python, with the key masked, so you can copy a working request into another tool.

Saved models: one library, many keys

Each saved group keeps its own key and endpoint; keys are never shared between groups and never shown after saving. The active model is referenced by name, so the form stays empty and nothing sensitive sits in it.

1. Save current model puts what is in the form into the library and activates it.
2. Use beside any model switches to it; In use marks the current one; ✕ Detach at the top returns to the plain form.
3. set / add key on a group replaces or adds its key; ✕ deletes a model or a whole group.
4. In the AI Console, the model button at the bottom of the composer lists every saved model, live local servers and the built-in model. Switching mid-chat re-sends the conversation, so nothing is lost.

Pro models: no key at all

A Pro plan adds models on our bill. They appear in the same model button under Vulture Pro (shared) and in Settings under Pro AI models. Requests go through our server with our key, are counted against your plan, and are never mixed with your own keys. Your own keys keep working alongside them.

Images: FLUX, Stable Diffusion and more

Charts, flowcharts, logos and UI mockups need no key: VultureEngine renders them on your device from what your chat model writes. For photo-real or artistic images the console asks a provider you choose. Two of them are free.

1. Settings → Image generation: pick the provider. Pollinations works at once with no key.
2. Paste the key where one is needed; the get one ↗ link opens the right page. For Cloudflare paste accountId:token.
3. Pick a model, or press Fetch models where the provider lists them, or type any model id under Custom / other. Choose a default size. It saves as you type and the pill at the bottom says Ready.
4. In the AI Console open the model button and pick your provider under Image engine, or type /engine images. Your chat model stays loaded and, with Refine prompts on, expands each short prompt into a detailed one first.
5. /engine vulture or ↩ VultureEngine hands the console back to your language model.

Tavily: web search for the AI

With web search on, the AI Console, Job Helper and the Cover Letter can read current pages and cite them. Pro plans include searches on our shared keys; the Web search permission also lets you use a Tavily key of your own, which is never stored on the server and never counted against the plan.

1. Sign up at app.tavily.com (the free tier includes monthly credits) and copy the key from API keys. It starts with tvly-.
2. Settings → Web search: paste it into Your Tavily API key and press Save. The pill reads Ready with your own key.
3. In the AI Console press the Web button in the composer. Each message is searched first and the answer ends with a Sources list.

Apify: LinkedIn, Indeed, Glassdoor and any careers page

Real scrapers on your own Apify account, used from Scraped Jobs, Job Search and the AI Console's agent. Runs spend your Apify credit; Pro plans can also run on our shared tokens within the plan's daily budget.

1. Create an account at console.apify.com, then open Settings → API & Integrations and copy the personal API token. It starts with apify_api_.
2. Settings → Apify scraping: paste it and press Save. Vulture checks it straight away; the pill reads Token works · your username.
3. Use it from Scraped Jobs → Apify (LinkedIn, Indeed, Glassdoor presets or any actor) or ask the agent: "scrape LinkedIn for React roles in Berlin".

Test it

1. Open the AI Console. The status badge at the top reads Idle with the provider and model.
2. Press the Test connection chip (or type Reply with exactly: OK). A working key answers within a few seconds.
3. /stats shows the model in use, the context window and this session's tokens. Settings → AI usage keeps the totals per model.

Troubleshooting