documentation · BYOK setup
Bring your own LLM key
repryntt does not mark up tokens. You connect provider keys; we route each task to the right model. You pay the provider directly.
Why BYOK
Your own key is optional. The $19.99/month repryntt plan includes 20 hours of work a month on our AI — one pool across our cloud and the Bot download on your own computer — with more hours at 69¢ each. When you bring a key, inference cost on it stays between you and the provider — the same dollar of OpenAI credit buys you the same number of tokens whether you call them from repryntt or directly — and on the Bot download it draws nothing from your hours.
If a provider raises prices, you see it immediately. If they cut prices, you see that too. No surprise platform invoices.
Connect a provider
- 1. Go to Dashboard → Settings → LLM Providers.
- 2. Click "Add provider" and pick one from the list.
- 3. Paste your API key. We encrypt it at rest with AES-256-GCM before persisting. You can rotate or delete it at any time.
- 4. Optional: set a per-provider monthly budget. We'll halt routing to that provider when the budget is hit.
Supported providers
| Provider | Env var (OSS) | Where to get a key |
|---|---|---|
| Anthropic | ANTHROPIC_API_KEY | console.anthropic.com/settings/keys |
| OpenAI | OPENAI_API_KEY | platform.openai.com/api-keys |
| xAI | XAI_API_KEY | console.x.ai |
| GOOGLE_API_KEY | aistudio.google.com/apikey | |
| NVIDIA NIM | NVIDIA_API_KEY | build.nvidia.com |
| Local (Ollama / vLLM) | REPRYNTT_LOCAL_LLM_URL | self-hosted (e.g. http://localhost:11434/v1) |
Provider notes
Anthropic
claude-fable-5, claude-opus-4-8, claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5
Recommended default for reasoning-heavy agents. Workbench critics use Opus by default.
OpenAI
gpt-4o, gpt-4o-mini, o1, o3
Strong general-purpose router target. Cheapest tier for high-volume routine jobs.
xAI
grok-2, grok-2-mini, grok-vision
Optional. Useful for long-context and real-time-data agents.
gemini-2.5-pro, gemini-2.5-flash
Gemini 2.5 Flash is the cheapest frontier vision model — used for tiered vision in robotics.
NVIDIA NIM
Llama 3.x, Mixtral, Nemotron
Hosted open models. Often the cheapest path for high-throughput producer jobs.
Local (Ollama / vLLM)
any OpenAI-compatible local server
Fully offline. Set the env var to your local OpenAI-compatible endpoint and pick a model in Settings.
Multi-provider routing
Connect more than one provider and the router picks per task: cheap models for cheap work, frontier for hard. You can pin specific agents to specific providers in Settings → Routing.
The Coherence Cloud critic always uses a frontier judge regardless of producer routing — that's the whole point of an independent reviewer.
Security
- Keys are encrypted at rest with AES-256-GCM under a server-held secret. Plaintext is used only at the moment we call the provider on your behalf.
- You can rotate or delete a key from the dashboard at any time; future requests fail immediately.
- Set per-provider budgets to cap blast radius if a key leaks.