Bring your own LLM key. We run the box.
Every BYOK page on the internet is about an IDE plugin. This one is about a colleague that runs 24/7. Your key, your model, your provider invoice. Our servers, our restarts, our 3am problem.
OpenAI, Anthropic, Google, Groq, Mistral, xAI, DeepSeek, Perplexity, Cohere, Bedrock
OpenRouter, for one key across hundreds of models
Ollama, cloud or your own local instance
Any OpenAI-compatible endpoint, by pasting its base URL
What it does
Work that lands on your desk (or doesn't, because the agent handled it).
You pay the model provider
Your key, your account, your invoice, at their list price. We never sit in the middle of it.
Keys are encrypted at rest
Stored per user in an encrypted column, delivered to your agent over the provisioning channel, never printed in a dashboard or a log.
Switch model without redeploying
Change the model in the dashboard. The agent picks up the new setting and restarts itself.
Different agents, different models
Run a cheap model on the inbox triage agent and a frontier model on the one doing research.
Self-hosted models welcome
Point an agent at Ollama or any OpenAI-compatible endpoint you run yourself.
Managed tokens if you prefer
Do not want to hold a key at all? Use our managed token allowance instead and skip the provider signup.
Skills included
Arrives knowing its job
What bring your own key actually removes
It removes the reseller. Platforms that resell inference decide which models you may use, when you may use a new one, and what the margin is. When the key is yours, the model list is whatever your provider offers on the day, and the price is whatever they charge. The hosting fee is then honestly a hosting fee.
What it does not remove
The server. An agent that answers at 3am has to be running at 3am, which means a box, a process supervisor, a restart policy, a lock file, log rotation and someone who notices when the gateway wedges. That is the part we do, and it is the part that quietly eats a weekend when you do it yourself.
You still see every token
- Usage is recorded at the model call, for every channel, not just the ones you can see in a chat window.
- Input, output, cache read, cache write and reasoning tokens are counted separately.
- Cache reads are shown, not hidden. On a long-running agent they carry most of the volume, and a total that drops them is not a smaller number, it is a wrong one.
- Cost is computed per run against the real provider rates for the model you chose.
Bring your own key, on every plan
BYO model is included from the $29 Starter plan up. There is no tier where holding your own key is a paid upgrade, because charging for the privilege of paying someone else would be a strange thing to do.
FAQ
Questions we hear a lot
Which providers can I bring a key for?
OpenAI, Anthropic, Google, Groq, Mistral, xAI, DeepSeek, Perplexity, Cohere, Bedrock, OpenRouter and Ollama, plus any endpoint that speaks the OpenAI API, which you add by pasting its base URL.
Where is my key stored?
Encrypted in the database, scoped to your account, and written to your agent as a credentials file on its own server. It is masked everywhere in the interface, and updating other settings never overwrites it.
Do you take a cut of my model spend?
No. When you use your own key, the provider bills you and we never see the transaction. Our revenue is the hosting subscription.
Can I change models later?
Any time, from the dashboard, per agent. The agent reloads its configuration and restarts on its own.
What if I do not want to manage a provider account?
Use managed tokens. Every plan includes an allowance, from 500K a month on Starter, and you skip the provider signup entirely.
Can I use a local model?
Yes. Point the agent at your own Ollama instance or any OpenAI-compatible server you host. Cloud Ollama models work the same way with an ollama.com key.
Your next hire doesn’t need onboarding.
Pick a role, plug in your API key, connect your channel. Live in five minutes. Free to start.