CALL-E

Use this when the user asks for the status or result of a call. Poll while result_status is pending; unavailable or not_applicable results are final even when result and error are null.

Query a call run by run_id. This tool does not initiate calls or modify run state. After calling 'run_call', poll get_call_run for realtime progress updates. Poll every 1-3 seconds while activity is changing, then slow down; do not call plan_call or run_call again.

Use this when the user asks what was said on their call. Return available transcript only; missing transcript does not mean no call occurred.

Use this when the user wants to view an existing call and its card controls. Never places another call.

Create or update a saved phone-call plan from the user's request; this tool does not dial. In 'user_input', pass only the user's call-related request or clarification answer, preserving its facts, constraints and corrections. Omit unrelated conversation and unnecessary personal information. Ask only for missing details needed to carry out the requested call. Do not guess region/language or reformat ambiguous phone numbers. If 'ready_to_run' is false, prefer letting the user fill the missing details in the plan card UI instead of restating every question in chat. If the user answers in chat instead, call this again with 'user_input' set to the user's response. When 'ready_to_run' is true and a plan card UI is available, stop after 'plan_call' and let the plan card continue execution. Only call 'run_call' yourself if no plan card UI is available or the user explicitly asks to start or retry the call in chat. Do not guess parameters.

Create or update a saved preparation for one immediate phone task. Never places a call. The server determines missing information. Send task and only known phone/region/locale. When status is ready, call start_call with its preparation_id and revision without asking the user to confirm again. When status is needs_input, ask the returned clarifications in chat (the card can also answer them), then revise with preparation_id, its revision, and user_input. To reopen, pass preparation_id only. Never author questions or guess missing facts.

Executes the planned call. Use only after 'plan_call' returns 'ready_to_run=true'. Pass the 'confirm_token' exactly as received. In ChatGPT, do not call 'run_call' in the same turn if a plan card UI is available and has already started the call. Use 'run_call' there only when no plan card UI is available or the user explicitly asks to start or retry the call in chat. If the call starts, it runs asynchronously. Do not perform extra operations; the server will notify on completion. Do not call 'run_call' more than once for the same 'plan_id'. Once started, wait for the activity card updates.

Use this to place the call for a ready preparation, with the preparation_id and revision returned by prepare_call. Repeating it returns the same call and never dials twice. Follow the result with get_call.

Widget-only batch telemetry ingestion tool. Use this from the app UI to submit a telemetry batch for a single plan or run. Events must be sent in batches, not one-by-one.