Run Daslab on your keys, your servers, or under your own name
Every workflow becomes a scene — its data, its tools, the agent that runs it, and the history of everything it did. You choose how much of what runs it is yours: add your own model keys and they serve first, run the same software on your own machines, or have us run an instance that carries your name.
••••k3Xqyour provider account, your billonYour own model keys serve before ours
Add an Anthropic, OpenAI, Gemini or OpenRouter key to a workspace and it goes to the top of that provider's list. Every call tries the keys in order. A key that is rate-limited, out of quota or revoked is skipped for five minutes and the next key answers; once output is flowing nothing switches mid-reply.
Daslab plan credits sit in the same list. When your plan runs out, they fail the way a spent key does, and the call continues on your own key instead of stopping.
Metering follows whoever served. A call your key answered is on your provider bill and not counted against the plan; a call our credits answered is counted, at list price, the same as for any workspace.
••••9fA2served 2,310 callson••••XXCwserved 12 callsonThe order decides how you pay
There is no separate setting for it. Put our credits on top and your keys take the overflow; put your keys on top and ours are the safety net; switch ours off and every call runs on accounts you hold.
Use the plan you pay for; your own account picks up what the plan no longer covers.
Your provider bill carries the work; ours answers only when your key is spent or down, and is metered when it does.
offNo model traffic runs through a Daslab account. If your key fails, the call fails with your provider's error.
The API, the website, sign-in, webhooks and background jobs in one process. No cluster to keep in sync.
PostgreSQL, Redis, and at least one model key. The schema is created and migrated on start.
Any model key variable may hold several keys, comma-separated. The server fails over between them on quota, spending cap and revoked-key errors.
The same software runs on your own machines
Daslab Cloud is our deployment of the binary you can run yourself. Point it at a Postgres and a Redis, give it a domain, and the workspaces, scenes, traces and approval gates are the same ones, on your infrastructure.
On your own instance the last entry in every list is your server's keys. Workspaces inside it add their own keys above those exactly as on Cloud. The self-hosting guide has the whole setup, including the OAuth apps that have to be registered in your name. Self-hosted deployments are part of the Enterprise plan.
We can run an instance under your name
Some teams want what the self-hosted path gives them without operating it: their name and domain on the product their users open, model keys they hold, and the data where their contracts say it lives. We set that instance up with you and run it, and it says "Powered by Daslab" where the product says who built it.
The split is the one the self-hosting guide already draws. What is yours stays yours; what you would rather not operate is ours. How whitelabel works, and talk to us and we go through the table together.
Which key served a call is recorded with it, next to cost and tokens. The workspace's key settings add it up per key.
Plan credits stop at the cap. A workspace with its own keys keeps working on them; one without stops until it tops up.
Model usage on our credits is charged at the provider's price. On your key we never see the bill.
Which plan do I need to add my own keys?
Any plan, including Free. Keys are added per workspace and shared by every scene in it.
What happens when my key runs out mid-month?
The next key answers. If that is Daslab plan credits and they are on, the call is metered against your plan; if you switched our credits off, the call fails with your provider's error and the job tells you so.
Can I see which key a job used?
Yes. Each model call records the key that served it, and the workspace's key settings show calls and cost per key for the last 30 days.
Do you read my key?
The server uses it to make the call. The settings page and the API show only its last four characters.