Hey Boris, quick things to check: - Make sure your tool calls are marked as internal in your prompt (e.g. “this is an internal tool, don’t say this to the user”). - Re-save your assistant and refresh the session —> sometimes after an update, assistants need to be re-synced to respect the prompt.

Steven it's this bit that trips many individuals up. You don't ask your client to create their own OpenAI account. Here's how it works on Assistable: - You (your agency) make an OpenAI API key under your own account. That's where charges apply. - You may enable rebilling for subaccounts in Assistable. With rebilling enabled: If rebilling is disabled, then the subaccount will try to use the subaccount's wallet balance in Assistable. If such balance is zero and no rebilling happens, then it will make the bot not work (what happened). So the possible configurations are: 1. Your OpenAI key + rebilling on -> most highly-recommended, you're always in control, customers just pay usage to you. 2. Subaccount balance only -> they replenish their Assistable wallet directly, without rebilling you. You don't have to ask your customers to create OpenAI accounts — you just take care of everything on your agency side.

Hi Jon, the assistant will actually switch languages, but there are a few limitations you should keep in mind: - Recognition: It notices when someone starts speaking in another language and switches over, but usually takes a sentence or two to adjust. - Consistency: Once it switches, it tends to stick with that language. Rapid back-and-forth can get messy. - Quality: For big languages (Spanish, French, German, Portuguese), it sounds very natural. For less common ones, it can feel a little rough. - Voices: Using a multilingual-friendly one gives the smoothest experience.

Hi Eric that error indicates that request going from your end to the model became too large with your current limits. In particular: - Your org has a tokens per minute (TPM) of 30,000 limit. - It attempted through 45,709 tokens, and this went beyond the limit. - The model doesn’t like it because it cannot handle more than your limit admits. A couple of solutions to this problem: 1. Shrink prompt size — If your assistant prompt or knowledge base is quite extensive, edit it down or break it into several steps so each request remains small. 2. Output shorter — If generated responses are longer than desired, mend settings such that the model outputs remain short. 3. Increase your rate limits — You may appeal a higher TPM limit from OpenAI if you frequently require bulkier requests: https://platform.openAI.com/account/rate-limits Short answer: Either broaden the request or lift the limit of the account — otherwise the model will continue spitting out this error once the limit has been reached.

Hi Boris. Your “best” mix can vary depending on what style you want with your AI, but the following are some of the general rules we’ve discovered to be true, - Friendliness: Close to 0.9–0.95 feels a bit too rapid yet just right. Close to 1.0 or greater feels too rapid, and much farther back makes the AI too slow. - Sensitivity to Interruption:Good value of 0.5–0.65. Smaller values make the AI more patient (waiting until a caller fully finishes), while higher values make it cut off a little too early. Your current settings (0.98 speed, 0.6 interruption sensitivity) are good but if you desire smoother, more natural-sounding speech, reducing response speed a tiny bit (0.9–0.92) often has the advantage of making speech feel more natural. It's okay to get creative with your audiences since some business areas such as a quick pace (i.e., sales) and others prefer a longer duration of time (i.e., wellness and medical).