the EV roadmap
from blank key to useful assistant.
There is no single perfect provider. Start with one model API, add search when EV needs fresh information, then layer in vision and local tools as your workflow grows.
01
the shortest path
01
get a key
Create one provider key and keep it in EV settings.
02
make one request
Test a simple voice or text prompt before adding tools.
03
add context
Connect search and screen awareness only when you need them.
A good first setup is OpenAI + Tavily: one provider for conversation and multimodal understanding, one provider for current web results.
02
choose your API stack
Use the key links to create credentials and the docs links to learn each provider’s request format. EV is designed around bring-your-own-key workflows, so review pricing and rate limits before enabling a provider.
03
a practical build order
- 1
conversation
Start with a chat or realtime model. Confirm EV can hear, understand, and answer.
- 2
tools
Add the tools EV should call locally: opening apps, reading the screen, or controlling the cursor.
- 3
fresh information
Add Tavily Search for questions where a static model answer is not enough.
- 4
routing
Try Groq or OpenRouter when you need faster responses or want to compare models without rewriting your app.
- 5
specialists
Use Hugging Face or Gemini for open models, embeddings, long context, or vision-heavy workflows.
04
keep your keys safe
Never put an API key in client-side code, a public repository, a screenshot, or a URL. Use EV’s local settings or a secret manager, set spending limits where the provider supports them, and rotate a key immediately if you think it leaked.
see the installation checklist