Create your first Burki assistant: a browser-first walkthrough

Your first milestone is a useful browser conversation with an unpublished assistant. Telephone service, real actions and production acceptance come afterward. There is no guaranteed thirty-minute launch: the task, account readiness and integrations determine the work.

Prepare a short business brief

Use a fictional home-services company for practice. Write its name, approved service area, opening hours and the request it should capture. For example: “Collect the caller's repair issue and preferred callback time. Do not quote prices or promise availability.” Avoid real customer details while learning the workflow.

Create and review the draft

Open Assistants and start creation from your description. A free starter produces an editable template. Optional AI workflow generation is a separately metered operation; browser trial time does not pay for it. Review any quoted charge before choosing it.

Read the resulting conversation instructions and supported workflow steps. Remove invented business facts. Check that the greeting identifies the assistant appropriately, the questions are short and the outcome matches your brief. New assistants default to GPT Live; existing saved assistants can retain another configuration.

Practice the selected version

Choose the browser test for the draft you are editing. Check the remaining allowance and readiness before starting. Eligible personalized trials provide up to three minutes pooled across sessions, with a two-minute limit per call, subject to the displayed remaining allowance and shared trial capacity. They do not include a free telephone number or a telephone campaign.

Try a repair request, correct the callback time and ask for an unsupported price quote. The useful result is accurate capture and an honest boundary, not a claim that a real booking or message was sent. Free-trial external actions are simulated where supported.

Review, correct and retry

Inspect the conversation and available action evidence. Did the final details reflect the correction? Did the assistant invent a price? Change the specific instruction that failed and test again within the remaining allowance or funded usage. Testing must not silently publish the draft.

Add real capabilities deliberately

A real calendar action needs its supported connection, appointment type, required fields and funded usage. Confirm the selected slot before creation and check the returned booking. Telephone use needs funds, a supported number and route, and a published configuration ready for that channel. Human handoff requires separate acceptance of the selected route and its failure behavior.

Use pricing before funding or acquiring a recurring number. Keep the initial task narrow until the instructions, connected action and telephone path have each been verified.