Providers
Configure keys with /setup. Choose a model with /model. Failover keeps free paths alive when paid routes die.
Supported providers
| Provider | Get a key | Typical use |
|---|---|---|
| Gemini | aistudio.google.com | Default free-key path - prefer gemini-flash-latest |
| OpenRouter | openrouter.ai/keys | Broad catalog - use :free without credits |
| Groq | console.groq.com | Fast open-weight; strong failover |
| Mistral | console.mistral.ai | Prefer Codestral for tool calls |
| NVIDIA | build.nvidia.com | Integrate API (optional) |
Setup flow
/setup # add / update keys (missing-only walk)
/model # pick provider + model idKeys persist in user config. /setup can also clear / rebind the active provider if its key was removed.
Failover behavior
- Hard faults (no credits, unavailable, empty free replies) switch models.
- Paid OpenRouter credit exhaustion burns that paid route - not every free id.
- A single OpenRouter 429 does not wipe the rest of the free catalog.
- Routing prefers free / Groq over expensive dead-ends.
- Dead model IDs are skipped on later turns.
If all configured models fail, the finish path lists free /model options to try next.
Free-tier tips
- Start with Gemini flash-latest or OpenRouter
:free. - Add Groq as a second key for rate-limit resilience.
- Avoid pointing the agent at paid Claude/R1 routes without credits.
Token caps
On free / Groq paths, VritraAI may lower agent max_tokens to avoid “not enough tokens / afford” errors from providers.