Skip to main content
Parlé is model-agnostic: your organization connects its own API key for each AI provider it wants to use, and Parlé uses that key to run conversations. This is sometimes called BYOK (“bring your own key”) or BYOAK (“bring your own API key”). Using your own key means:
  • You are billed directly by the provider, at the provider’s own rates — Parlé does not mark up model usage.
  • Your data stays under your provider account’s terms, not Parlé’s.
  • Your organization controls which models are available to your team, and can rotate or remove a key at any time.
Parlé also offers Parlé Managed models that need no key of your own — these are billed from your organization’s Parlé credit wallet instead, always in US dollars. Managed models do not appear in the setup steps below because there is nothing to connect.

​
Features That Run Only on Parlé Managed Models

Two features always run on Parlé Managed models, whatever provider keys your organization has connected:
  • Research tasks, where the agent works through a piece of research on its own over many steps.
  • Live artifacts, which Parlé’s artifact builder creates and repairs for you.

​
Why these features don’t use your own keys

Both features run unattended through many steps, and both rely on detailed instructions that Parlé writes, tests and keeps up to date. They only work well on models that are capable enough for that kind of work and have been tested with those instructions. Running them on Parlé Managed models lets Parlé make sure that:
  • A capable model is always available, whichever providers your organization has connected and whichever models it has enabled.
  • The instructions and the model work together. Parlé tests every change to its instructions on the models that will actually run them. A model reached through your own key could be older, smaller or set up differently. The task could then stop partway through, or finish with weaker results and no clear error.
For the same reason, these features have no model picker. Parlé chooses the model. Their usage is charged to your organization’s credit wallet in US dollars, like any other Parlé Managed usage.

​
What your organization needs

An Org Admin sets how your organization gets model access in Org settings → Models. To use research tasks or live artifacts, it must be set to one of these:
  • Parlé credits only
  • Our keys first, Parlé credits as backup: your conversations still use your own key wherever one is set. Parlé credits are used only for models you have no key for, and for research tasks and live artifacts.
The credit wallet (Org settings → Budgets & alerts) also needs funds. Live artifacts also need the Business Elite plan or above. See Plans & Billing.

​
If your organization uses only its own keys

When model access is set to Our own keys only, Parlé does not run research tasks or live artifacts on your keys. It refuses them and tells you what to change:
“Research tasks run on Parlé-managed models billed to your credit wallet, and your organization uses only its own API keys. To use research tasks, an organization admin can switch to hybrid access in Organization settings and top up the credit wallet.”
Live artifacts show the same kind of message. In practice:
  • You can’t start, restart or resume a research task. If a task is already running when your organization switches to its own keys only, it stops before its next model call. You can resume it after switching back.
  • You can’t create a live artifact or keep building one. The artifact builder’s changes are refused.
  • Nothing is deleted. Existing research tasks and their results stay readable and can still be cancelled. Your artifacts are kept too.
To turn these features back on, an Org Admin switches model access to Our keys first, Parlé credits as backup and tops up the credit wallet.

​
Connecting a Provider

Only Org Admins can add, rotate, remove, or test a provider key.
1

Open the Models tab

Go to Org settings → Models. Each provider has its own card.
2

Paste your API key

Find the provider’s card and paste your key into the API key field. See the tabs below for where to get one, and what to check on the provider’s side first.
3

Select Connect

Parlé makes a small verification call to the provider before saving, so you find out immediately if the key is wrong rather than at chat time.
4

Set the quote currency

If this provider bills you in a currency other than US dollars, say so now — before any rates or spend are recorded. See Setting your quote currency below.

​
What the verification call tells you

To replace a key later, select Rotate key on a connected card, paste the new value, and select Save key. To disconnect a provider entirely, select Remove — its models immediately stop being available to your team.

​
Setting Your Quote Currency

The quote currency is the currency your agreement with a provider is quoted and billed in. Parlé supports USD, CAD, EUR and GBP, and every connected provider card shows a Quote currency line. It defaults to USD. Where you set it depends on the provider:
1

Open the Azure AI Foundry card

Go to Org settings → Models and find Declared deployments on the Azure AI Foundry card.
2

Choose the deployment's Quote currency

Each deployment row has its own Quote currency list. Pick the currency your Azure agreement bills that deployment in. The change saves as soon as you select it.
3

Enter your rates in that currency

Under Deployment rates, the form now reads “Enter prices in GBP, exactly as your price sheet quotes them — Parlé stores this deployment’s rates in GBP.” Type the figures exactly as they appear on your price sheet. Do not convert them yourself.
Why it matters. Every message sent to that deployment is costed from those rates, recorded in that currency, and converted at Parlé’s monthly exchange rate wherever it is measured against a project cap or shown in a report. Each rate is stamped with the currency it was entered in and keeps it, so your rate history still reads correctly if the deployment’s currency changes later. See Currencies and Exchange Rates.
If you leave a deployment at USD but type rates from a GBP or EUR price sheet, nothing errors — the figures are simply read as dollars. Recorded spend is then wrong by the exchange rate, project caps are measured against the wrong amount, and the Foundry reconciliation statement will not match your Azure invoice. Set the currency before you enter rates.
Azure list prices are USD only. For a deployment in any other currency Parlé does not apply or compare Azure’s list prices — the rate card says “Azure publishes list prices in USD only, so Parlé does not compare this GBP deployment against them.” Moving a deployment away from USD also stops any Azure list price Parlé was applying to it, so the deployment reads as not priced until you enter your own rate card in the new currency. Do both in one sitting.

​
When the quote currency can be changed

You can change a quote currency freely until your organization has a live rate of its own on file under it. After that the change is refused, so an existing rate can never end up describing a different price sheet from the one it was typed from. The message names the count and the currency, for example:
“This deployment has 1 live rate on file in GBP, entered by your organization. Retire it and re-enter the rate card in EUR before changing the quote currency.”
To change it anyway, retire your rate on the deployment’s row under Deployment rates (Use list price, or Clear estimate on a provisioned deployment), change the Quote currency, then enter the new rate card. Retired rates stay in the rate history in the currency they were entered in, and spend already recorded keeps the currency and rate it was charged at — changing the currency changes nothing that has already happened. Two more things worth knowing:
  • Rotating a key keeps its quote currency. Removing a key and adding a new one for the same provider also carries the previous currency over, rather than resetting to USD.
  • Every change is recorded in the audit log, with the old and new currency.

​
Provider-Specific Setup

Each provider has account-side setup you need to do before pasting a key into Parlé. These steps happen on the provider’s own website.
Create a key in the Anthropic Console under API Keys. Copy it before closing the dialog — Anthropic only shows the full key once.Make sure the workspace the key belongs to has billing set up and available credit. A key from a workspace with no credit will connect, but requests will fail.

​
Enabling Models and Setting a Default

Connecting a key does not automatically make every model from that provider available to your team. You choose which models to enable, and one becomes the organization’s default model for new conversations.
1

Review the Available models list

On a Connected provider card, each row shows the model’s name, context window size, and whether it supports tools and vision.
2

Tick the models you want and save

Select Save models.
3

Optionally turn on Auto-allow new models

When the provider releases a new model, it is enabled automatically instead of waiting for an admin to opt in.
4

Set the organization default

Next to any enabled model, select Set default. The first model you enable becomes the default automatically; you can change it at any time.
Models your provider account cannot actually see — most commonly an OpenAI model not enabled for the connected project — appear greyed out and unselectable, with a note explaining they are not available on your account. Enable them on the provider’s side first, then revisit this list. See Model Selection for restricting which of these models a given project may use.

​
Testing a Connection

Test connection sends a real, one-word message to the provider using your organization’s key and a chosen model — the same code path a real conversation uses. It is the most reliable way to confirm a key and model work end to end, because some provider-side restrictions (like OpenAI’s model-access and account-verification gates) only show up on an actual request, not on a simple key check.
1

Select Test connection

On a Connected provider card.
2

Confirm the prompt

It tells you which model will be used and that this makes a real request on your own provider account, at a fraction of a cent.
3

Read the result

Success shows the response time in milliseconds. Failure shows a plain-language reason plus the provider’s own error text, so you can act on it directly.
Connection tests are not available for Parlé Managed models — there is no API key of your own to test.

​
What Happens to Your Key

Parlé stores your API key encrypted at the org level, resolved at request time. The key is never displayed again in the UI — only a masked placeholder appears in settings to confirm a key has been saved. It is used solely to forward API requests on your behalf.
Only Org Admins can add or remove provider connections. Project admins and members cannot view, copy, or modify API keys in any part of the interface.

​
Troubleshooting

Still stuck? Contact Parlé support with your organization name, the provider, and (if shown) the error detail from the connection test.