Once your company turns on BYOK in DataMind Oda, you can add your own API keys from Anthropic, OpenAI, Google Gemini and OpenRouter and chat with those providers' models next to the DataMind models. It works the same in the DataMind Oda desktop app and in VS Code. Your keys stay encrypted on your computer, and Oda sends those requests straight to the provider, billed to the key's owner.
Before you start
Your administrator turns on BYOK in DataMind Oda under Admin > Oda > General in DataMind OS; see Turn BYOK in DataMind Oda on or off. Until then, Settings > Models shows only the DataMind models.
You need an API key from your own account with the provider.
Add a key
Open Settings > Models. The Your API keys section, tagged Allowed by your company, sits above the DataMind models. A provider without keys shows a tile that reads No keys; a provider with keys gets a card that lists them.
Choose Add key on the provider's tile, or on its card to add another key.
In Add API key, pick the Provider, type a Name such as Main, and paste the key into API key. The eye button, Show key, shows what you pasted.
Choose Check and save. While the button reads Checking…, Oda checks the key with the provider by reading its model list, and it saves the key once the provider accepts it.
Every key has a name. The name is required, up to 60 characters, and unique for that provider. Oda uses it, never the key, in notes and messages.
Up to 10 keys per provider. Add several to keep working when one reaches its limit. A provider with 10 keys greys out its Add key, which reads "Up to 10 keys per provider."
Only the last 4 characters show again. After you save, the key reads like sk-ant-…a1F4 with the day you added it, such as added 4 Nov. The hint under the field reads "Stored encrypted on this computer, for your DataMind account only."
A key the provider refuses is not saved. The dialog gives the reason, for example "Anthropic refused this key (invalid or revoked). Check it in the Claude Console and paste it again."
A key at its limit is still saved. It rests until it is free, as below.
Put your keys in order
A provider with several keys lists them in priority order, numbered from 1. Drag a key to a new place, or select its handle and press the up or down arrow key. Oda always uses the first key in the list that can answer, and moves down only when a key cannot.
A key with no tag is ready. Otherwise each key shows how it stands:
Key shows
What it means
In use
Oda sent this provider's latest request with this key.
Rate limited · free at 14:35
The key reached a per-minute limit and rests until the time shown. Oda uses the next key meanwhile.
Credit used up · free at 15:10
The key's credit or monthly spend is used up. It rests for 1 hour, or until you choose Check again.
Refused by Anthropic: key invalid or revoked
The provider refused the key. Oda skips it, and the line under it reads "skipped until you replace it".
Checking…
Oda is checking the key with the provider after you chose Check again.
The menu at the end of each key offers Rename, Replace key, Check again and Delete. A refused key also shows Replace key on its row.
Rename opens Rename key. Type the new name and choose Save.
Replace key opens Replace API key. Paste the new key and choose Check and save; the key keeps its name and its place.
Check again asks the provider at once. When the provider accepts the key, its rest or refusal ends.
Delete asks Delete the key "Main"? with "Oda stops using it at once. This cannot be undone." Choose Delete to confirm.
How Oda moves between keys
A per-minute limit. The key rests until the provider says it is free again, and Oda moves to the next key.
Credit or monthly spend used up. The key rests for 1 hour, or until you choose Check again.
A refused key. A key the provider calls invalid, revoked or not allowed is skipped until you replace it.
A busy provider. When the provider is overloaded or has a server error, Oda tries the same key again, as for any model call; see Model calls and failures.
Only before the answer starts. Oda moves to another key before the first words of an answer arrive. A failure in the middle of an answer is retried like any other model call.
A note above the message box. When Oda moves on, a short note names the keys, for example Key "Main" hit its limit. Using "Backup 1". or Key "Main" was refused. Using "Backup 1".
Rest ends with the app. Oda remembers which keys rest while it runs. When you start it again, it tries your keys from the top.
When every key is at its limit
A key frees within a minute. Oda waits and shows a bar above the message box, such as Waiting for a free Anthropic key · 42s, counting down, then carries on.
Every key rests longer. The turn stops with the card All your Anthropic keys are at their limits and "The first one is free again at 14:35. Your conversation is kept; nothing was lost." Choose how to go on:
Continue with and your company's default DataMind model, such as Continue with Kartlos Standard, switches the chat to that model and carries on.
Retry carries on with the same model.
Manage keys opens Settings > Models.
The buttons follow the chat.Continue with and Retry show on the chat's latest turn while Oda is not working on it, and Continue with shows when your account can use DataMind models.
Oda never moves you to a DataMind model by itself. You choose it.
Scheduled tasks wait up to 15 minutes. A run on your key waits for a free key, then fails with All your Anthropic keys stayed at their limits for 15 minutes.
Choose which models the picker shows
Under your keys, each provider with a key lists its current models that can use tools and read images, for example under Anthropic models from your keys, with "Only models that can use tools and read images are listed. Refresh picks up new ones." Rows show Tools and Images, and the model's id.
The three newest start shown. The other models stay off until you turn them on with their switch, or with Show all. The card counts them, such as 3 of 12 shown.
Long lists fold. A link such as 8 more opens the rest, and Show fewer folds them again.
Search and refresh.Search models narrows the list, and Refresh picks up models the provider added since. A new model starts shown when it is newer than every model Oda has listed for you; an older one starts off.
In the chat picker, models on your keys come after the DataMind models, in their own groups such as Anthropic · your key, and the model button adds · your key to the model's name. A provider appears there once you have saved a key for it. See Models in the chat picker for the switches and Model selection and defaults for choosing a model.
In other model lists, such as the model of a scheduled task or a subagent, models on your keys are grouped by the model's maker, such as Anthropic.
The whole chat stays on your key. Compaction and the chat's title use the model you chose, and so do subagents unless you give them a model of their own. Inline and next-edit suggestions in VS Code keep running on Kartlos Standard.
Choose how much the model thinks
Models on your keys that can reason show Effort at the bottom of the model picker, as the DataMind models do: Low, Medium, High and Max. A chat on your key starts at Medium.
The bar belongs to the chosen model. Moving the pointer or the arrow keys over other models leaves it on the model you chose, so you change its effort in one move.
Oda speaks each provider's language. It turns the level into that provider's own reasoning setting. Where a provider's highest level is High, Max thinks as much as High.
On Anthropic models, a new effort level starts a fresh prompt cache, so the request right after the change reads less from the cache.
See what your keys cost
Requests on your keys show their cost in the chat statistics, next to Cache hit, and as Cost (your API keys) in Session statistics and in a turn's usage details.
OpenRouter reports the exact cost, and Oda shows it as reported.
For Anthropic, OpenAI and Google Gemini, Oda works the cost out from the tokens and the provider's list prices, and shows it as an estimate, for example ≈ $0.42.
A model with no known price shows its tokens only.
The cost stays on your computer, like everything else about the requests on your key.
What stays on your computer
Your keys. They are encrypted on this computer and sent only to their provider. They never reach your DataMind server, they are kept apart from Oda's settings files, and Oda removes them from error messages and from Copy Details.
The requests. The model requests on your key go straight from your computer to the provider. DataMind OS receives nothing about them: not the content and not the usage. They are not recorded in Observability and do not count toward your AI limits.
Your company's rules still apply. Oda adds your company's Organization instructions to every request, and it refuses a scheduled run on your key while scheduled tasks are not enabled for your account.
Everything else works as before. DataMind models, Mind tools, web search and page reading keep going through your DataMind server.
Keys belong to your account. Your keys work only for your DataMind account on that DataMind server. Someone else who signs in to Oda on the same computer sees none of them.
When your company turns it off
Within 5 minutes, Oda sees the change and the models on your keys leave the picker.
A turn that is running finishes its current step and stops before the next one with the card Your company turned off your own API keys and "Models from your own API keys are no longer available in Oda. You can continue this conversation with a DataMind model." Choose Continue with and your company's default DataMind model to carry on.
A scheduled task on your key fails with Own API keys are turned off by your company. Pick a DataMind model for this task.
Your keys stay, encrypted on your computer and unused. Settings > Models reads Your company turned off your own API keys. with, for example, "Your 2 saved keys stay encrypted on this computer and are not used. They come back if the company turns BYOK on again."
Delete saved keys removes them. It asks Delete your saved API keys? with "Every key saved on this computer for your DataMind account is removed. This cannot be undone." Choose Delete saved keys again to confirm.
When Oda cannot reach your DataMind server, it keeps the last setting it read for up to 24 hours, then pauses your keys until it can check again. A turn on your key then stops with Your own API keys are paused and Continue with, and Settings > Models reads "Your own API keys are paused: DataMind Oda could not reach your DataMind server."
When you sign out, or your account loses access to Oda, your keys stop at once. If your server no longer accepts your account, Settings > Models reads "Your DataMind server no longer accepts this account. Sign in again to use your own API keys."
Questions, answered
Does DataMind see my API key?
No. The key is encrypted on your computer and sent only to its provider. Your DataMind server never receives it.
Who pays for requests on my key?
The key's owner, through the provider's own billing. These requests do not count toward your AI limits in DataMind OS and are not recorded in Observability.
Why did Oda switch to my second key?
Your first key reached a limit or was refused. The note above the message box reads, for example, Key "Main" hit its limit. Using "Backup 1"., and Settings > Models shows when the first key is free again. Oda goes back to it as soon as it can answer.
A key reads Refused by its provider. What do I do?
The provider no longer accepts that key, for example because it was revoked. Choose Replace key, paste a working key and choose Check and save. Until then, Oda skips it.
Why is a model from my provider missing?
Oda lists only models that can use tools and read images. Of those, the three newest start shown, so turn the others on in Settings > Models. For a model the provider added recently, choose Refresh.
Does Oda move me to a DataMind model when my keys run out?
Never by itself. When every key rests for longer than a minute, the turn stops and you choose: Continue with your company's default DataMind model, Retry on the same model, or Manage keys.
Why does the cost start with ≈?
Oda worked it out from the tokens and the provider's list prices, because that provider does not report a cost with each answer. Your provider's own bill is the final amount. OpenRouter reports its cost, so it shows without ≈.
Can I get a deleted key back?
No. Delete asks first, and once you confirm, the key is gone from this computer. Add it again with Add key if you still have it.
Can I use my keys on another computer?
Add them there too. Keys stay on the computer where you added them, for your DataMind account only.
What happens to my keys when my company turns BYOK off?
They stay encrypted on your computer and Oda does not use them. Delete saved keys on Settings > Models removes them; if your company turns BYOK on again, they come back.
Do my company's instructions apply to chats on my key?
Yes. Oda adds your company's Organization instructions to every request, whichever model you chat with.
Can someone else on my computer use my keys?
No. Your keys belong to your DataMind account on your DataMind server. Another account that signs in to Oda on the same computer sees none of them.