aiva_billing_report
ChatGPTReturns aggregate voice_minutes and agent_count per organization for a date range. Does NOT include cost, rate, or currency figures — this is a usage total, not a dollar bill.
aiva_check_access
ChatGPTConfirm connectivity to the AI Voice Agents API and return the account id for the configured credentials. Run this first if other aiva_* tools fail.
aiva_get_agent
ChatGPTGet an agent's full configuration (prompt, voice, tools, transfer rules) by id.
aiva_get_agent_sip_status
ChatGPTCheck whether an agent is currently registered to its PBX (registration state, expiry, last register). Use this as the authoritative source for whether an agent is currently live/registered — aiva_list_agents' status/is_active fields do not reflect registration state. Start here when an agent isn't answering calls.
aiva_get_metrics
ChatGPTUsage metrics for a date range: call volume, resolution rates, hourly distribution, per-agent performance. Optionally scoped to one organization or reseller. The response includes the resolved organizationId/resellerId the numbers belong to — verify it matches the tenant you asked about before presenting these numbers to a customer. Accounts pinned to specific agents (rather than a whole organization) are refused here — use aiva_report_minutes with their agent ids. Dates are YYYY-MM-DD calendar dates.
aiva_get_outbound_call
ChatGPTGet one outbound call's status and result. The AI summary is included when the organization's recording policy mode is 'full', or with identifiers masked under mode 'redacted'.
aiva_list_agents
ChatGPTList AI phone agents, optionally filtered to one organization. Paginated. For scoped accounts this is the intended STARTING POINT: omitting organizationId returns the account's own agents — a single-organization account is scoped upstream automatically; a reseller territory pages through the platform list with other tenants' rows removed, so a page can return 0 rows while hasNext is true (keep paging before concluding there are no agents). Returns AI Voice Agent records only, each with a non-null agentId — if this ever returns records without one, treat it as an error and do not report them as agents. The status field reflects account/subscription state and does NOT indicate whether the agent is currently registered or answering calls (a record can be status "active" with is_active false); use aiva_get_agent_sip_status as the authoritative source for actual liveness/registration state.
aiva_list_call_logs
ChatGPTList call logs for an AI voice agent: caller, timing, duration, resolution (handled/escalated), sentiment — plus full transcript, AI summary, and recording URL when the agent's organization has recording policy mode 'full'. Under mode 'redacted', transcript and summary text is included with identifiers masked and recording URLs withheld; otherwise those fields are stripped and the response says so. Paginated, newest first; pass 'since' for time-bounded questions like "today".
aiva_list_organizations
ChatGPTList organizations (tenants) under a reseller. external_id typically links the organization to its PBX domain. Omitting resellerId defaults to this account's own reseller when one is linked; a company account (office manager or personal seat) with no reseller gets its own organization(s) instead, so omitting resellerId is always the right call for a scoped account.
aiva_list_outbound_calls
ChatGPTList outbound calls placed by an agent, with status and timing. The AI result summary is included when the organization's recording policy mode is 'full', or with identifiers masked under mode 'redacted'. Paginated.
aiva_list_resellers
ChatGPTList resellers under the account (accountId defaults to the authenticated account). Platform administrators only — not available to scoped accounts, which should start from aiva_list_agents instead.
aiva_report_minutes
ChatGPTVoice minutes and call counts for a date range, aggregated per agent or per organization (with per-agent breakdown). Use for usage and billing questions.
fax_check_access
ChatGPTConfirm connectivity to the fax platform. Run this first if other fax_* tools fail.
fax_get_fax
ChatGPTGet one fax's details by messageId: status, pages, duration, result code, and whether the stored document has been purged. For the document itself, fax_get_fax_document returns redacted page images when the account's policy mode allows it.
fax_get_fax_document
ChatGPTRetrieve ONE page of a fax document as an image. What comes back depends on the fax account's recorded-content policy: under 'redacted', personal identifiers (names, SSNs, birth dates, phone numbers, record and member numbers) are automatically blacked out before anything is returned, and a page that cannot be read reliably enough to mask (handwriting-heavy, skewed, or blank pages) is withheld entirely rather than delivered unmasked; under 'full', the page is delivered as-is with nothing masked. Either way the page may contain sensitive personal information: redacted output is minimized, not de-identified. Requires policy mode 'redacted' or 'full'. One page per call: take the page count from fax_list_faxes' pagesCount (or fax_get_fax) and fetch the pages you actually need. A policy refusal from this tool is a compliance setting — relay it to the user; do not retry.
fax_list_accounts
ChatGPTList the fax accounts visible to this session: id, name, account number, and parent. Fax accounts are SEPARATE from phone-system domains — a customer can exist on the phone system with no fax account at all, and there is no automatic mapping between the two. Fax-account scope is independent of phone-system domain scope and is not auto-derived from a customer/domain name — this session's fax account may be a different real-world entity than its phone-system domain. Always check the returned name field before attributing any fax data to a specific customer; do not assume the account returned here matches the domain/tenant you were asked about. Use this FIRST to answer 'does this customer have fax service?' and to find the accountId the other fax tools take (either its id or its account number works there). Much cheaper than searching the number directory. One row is marked isThisAccount — this session's own account, where fax history lands when no accountId is given (for a customer session, its account; for a reseller session, the reseller's own account, with its customers listed beside it). On that row, name/number/parentId may be absent (only id and isThisAccount are guaranteed there), and when the name IS present it may resemble subaccount names — match on the flag, never the name. Results cover only accounts within YOUR scope: zero matches means no such customer under this account, not platform-wide absence.
fax_list_fax_numbers
ChatGPTList fax numbers (DIDs) with the account that uses each one and the users assigned to it — use this to find the number to give out, or to check who receives a number's faxes. A number belongs to the account that MANAGES it (managedBy), which is often not the account that bought it (ownedBy). With accountId, lists that account's numbers, paged with offset/limit. Without accountId, a customer or reseller session covers its own fax account first and then the accounts beneath it, up to 10 accounts and 100 numbers per account in one call; the response lists the accounts covered and says when one needs its own call. For a platform administrator, omitting accountId instead pages the platform-wide directory, which includes fax numbers belonging to OTHER accounts/customers, not filtered to any single tenant — and it is slow at platform scale. For a single-customer question, always call fax_list_accounts first to resolve accountId, and never report an unscoped result as belonging to one specific customer.
fax_list_faxes
ChatGPTFax history as line items: direction, status, page counts, duration, fax number, result codes, and timestamps. Use for delivery troubleshooting and volume questions. OMITTING accountId is the right default: history then covers this session's own fax account (the row fax_list_accounts marks isThisAccount), which is where most traffic lands; a reseller's customers' history needs their accountId. Naming a SUBaccount narrows to that subaccount alone and legitimately returns zero for most of them — an empty result there means 'no faxes on that subaccount', NOT 'no fax access'; retry without accountId before concluding anything is wrong. Fax-account scope is independent of phone-system domain scope and is not auto-derived from a customer/domain name — the account this history belongs to may be a different real-world entity than the domain/tenant you were asked about; check the account's name (fax_list_accounts) before attributing any fax data to a specific customer. Larger limits can exceed the response size limit even on accounts with routine daily fax volume — start with a small limit (10-20; the default is 20) and page with offset if you need more. Document content is available only via fax_get_fax_document (redacted page images, policy-gated). Rate limited upstream to 8 calls/minute.
fax_list_notification_settings
ChatGPTPer-user fax notification preferences: whether send/receive emails are sent, and whether those emails include the fax document as an attachment. Use this to review or explain how a user is currently alerted about faxes.
fax_list_webhooks
ChatGPTList configured fax webhooks: which events fire (e.g. inbound fax received), whether each is enabled, which number it is tied to, and where it posts (Slack, Microsoft Teams, Zapier, or a custom endpoint). Webhooks are read per fax account: name one with accountId, or omit it to cover this session's own accounts. Destination URLs are shown truncated because a webhook URL can be used to post into the channel.
report_problem
ChatGPTRecord the user's problem report about this integration — a wrong or incomplete answer, missing data, a confusing error, slowness — together with the last few tool calls of this session for technical context. Use when the user says something like "report that as a problem" or expresses that a result was wrong; describe the problem in their words. This appends one record to the integration operator's feedback log and touches no phone system, fax, or billing data.
voice_check_access
ChatGPTConfirm connectivity and show the scope/permissions of the credential this session uses (credential values are redacted). For any seat — reseller, company or personal — it shows the seat's own role: the seat tier and, when a phone-system login is linked, that login and its scope, plus whether reads currently travel as that login (readsAs) or are made by the platform within this account's scope (configuredLogin) — never key material. Run this first if other voice_ tools return 401/403 — but an error here does not predict other tools failing (scope is enforced per call), and questions about which PRODUCTS this account has need no call at all: absent tool families (fax_, aiva_*) mean that product is not enabled for this account.
voice_count_cdrs
ChatGPTCount call records and sum minutes for a domain (or one user) within a datetime range — use for volume/usage questions instead of paging through CDRs. 'total' counts call RECORDS, not distinct calls: one call can produce several (queue legs, transfers). 'seconds' is the sum of each record's whole span (call-batch-total-duration-seconds — ringing, queue and auto-attendant time included), so it runs higher than talk time; 'minutes' is that total divided by 60 and rounded up once, not per call. The response labels each figure. For a local day (e.g. "calls today") pass "today" as both datetimeStart and datetimeEnd — the server reads it in the domain's own time zone. Counts every call record in the range regardless of direction or disposition — there is no type/direction filter here; for a breakdown (inbound vs outbound, missed) use voice_read_cdrs with its type filter, and compare two periods with two calls.
voice_get_agent_log
ChatGPTAgent activity log for a datetime range: logins, logouts, and status changes — use to audit staffing/adherence (who was available when). Returns per-status durations per agent, including the busy reason codes set via voice_set_queue_agent_status (break, lunch, meeting, web, other, and the domain's custom statuses — request fields CUST1–CUST8 for those; the default field set is "AM,UM,L,B").
voice_get_agent_stats
ChatGPTPer-agent call-center statistics for a datetime range: calls handled, talk time, availability. Pass an agent for one agent's numbers, or a queue for that queue's agents (agent wins if both are given); omit both for every agent in the domain.
voice_get_call_trace
ChatGPTReconstruct a finished call's full journey — every leg, transfer, queue hop, and SIP response — by call-id (from voice_read_cdrs). Platform administrators and resellers only; a reseller can trace calls on its own tenants, and a call that also reached another provider's tenant (an on-net call or transfer to one) is refused. Pass the call-orig-call-id or call-term-call-id exactly as the CDR row has it. The trace server and time window are taken from the call record, so datetimeStart, datetimeEnd and servers are optional. Traces can be large. Check the CDR's "is-trace-expected" first: when it is false no trace was retained for that call, and this returns empty.
voice_get_dnis_stats
ChatGPTCall-center statistics grouped by the number the caller dialed (DNIS) for a datetime range — use to compare lines/campaigns feeding the queues. Optionally limit to one queue. For a local day pass "today" (or "yesterday", or a date) as both bounds — the server reads it in the domain's own time zone. Some platform versions answer with one empty entry per queue even for busy days; when agent statistics show queue calls in the same range the response says so (upstreamReportingGap) and explains how to count calls per dialed number from the call records instead. Limited to one queue, the note only says that queue's activity is unconfirmed, since the check is domain-wide.
voice_get_domain
ChatGPTGet configuration details for a specific domain. The response mixes static configuration with live capacity metrics; "limits-max-*" fields are a licensing/plan ceiling that may read 0 or stale even when many live objects of that type exist — do not treat them as live object counts. May time out on first call for larger/heavier domains; safe to retry once before treating as a failure — this is a read-only call and cannot have partially applied anything.
voice_get_queue_stats
ChatGPTCall-center statistics for a datetime range: calls handled, abandoned, wait and talk times, service level. Pass a queue for one queue's numbers, set perQueue for a per-queue breakdown, or omit both for a domain-wide aggregate. Some platform versions report nothing here even when queues handled calls; when that happens the response says so explicitly (upstreamReportingGap) and names the reads that do carry the numbers — never read an empty result as 'no queue activity' without checking that note. For one queue the check is domain-wide, so the note only says that queue's activity is unconfirmed and names the call records that settle it.
voice_get_recording
ChatGPTGet recording details for a call by call-id. Take the call-id from a CDR's call-term-call-id column (voice_read_cdrs): the recording resolves by the terminating leg, so the orig-side id returns 404 for calls that do have a recording. Requires policy mode 'metadata' (details without audio access) or 'full' for the domain. Under 'full' the audio comes as playback-url, a short-lived link (until playback-url-expires-at) that plays it in a browser; every play is recorded in the audit trail, and an expired link is replaced by looking the recording up again. A call captured by more than one recording rule is one row, with every rule listed in call-recording-match-patterns. A personal account pinned to one extension can read recordings of calls it took part in; any other call is refused.
voice_get_recording_policy
ChatGPTShow the recording/transcription access policy in effect (deny/metadata/redacted/full). Pass a domain to see that domain's mode — use this to explain why recorded-content access was blocked. Platform administrators additionally see every domain's override.
voice_get_transcription
ChatGPTGet the transcription of a recorded call. Requires policy mode 'full' or 'redacted' for the domain — under 'redacted', all transcript, summary, topic and sentiment text is returned with personal identifiers automatically masked. All arguments come from the same CDR row (voice_read_cdrs): jobId from "call-intelligence-job-id", date as the call's year+month (YYYYMM), and callId — which is the crux. A transcription resolves by the call-id of the leg its job was attached to: the ORIGINATING leg for outbound calls, the TERMINATING leg for inbound. Because you cannot tell which in advance, pass BOTH the row's call-orig-call-id and call-term-call-id, comma-separated, and this tool uses whichever the job was keyed to (a wrong single id fails as a 403 'Invalid domain or callids', which is not a permission error). A personal account pinned to one extension can read transcripts of calls it took part in; any other call is refused. Returns the AI summary, topics and sentiment by default; pass detail:"full" only when the word-by-word transcript is genuinely needed — it is roughly five times the size, so reading a day of calls at 'full' costs hundreds of thousands of tokens and will exhaust the context long before the day is covered.
voice_get_user
ChatGPTGet a specific user/extension in a domain, including status and configuration.
voice_list_abandoned_calls
ChatGPTCalls that hung up while waiting in queue during a datetime range, with caller, dialed number, start time (UTC and local) and wait time — the callback list. Optionally limit to one queue. Where the platform has no abandoned-call report (an unknown route on some versions), the server derives the same list from the call records itself — queue legs that no agent answered, one entry per call — and says so (derivedFromCallRecords, with the method and how many records it read). A very long range is cut off and marked truncated; narrow it to a day or a week.
voice_list_active_calls
ChatGPTList calls currently in progress in a domain (live snapshot).
voice_list_answer_rules
ChatGPTList a user's answer rules in priority order — per time frame: forwarding, simultaneous ring, do-not-disturb, voicemail behavior. Pass a timeframe name (use "*" for Default) to read that single rule. Strictly single-user: there is no bulk/domain-wide mode, so call it once per extension.
voice_list_auto_attendants
ChatGPTList auto attendants (IVR menus) in a domain — the entry points of most inbound call flows: each attendant's name, extension ("user"), starting prompt and time frame. To answer "what does pressing 1 do", pass attendant (the row's user) and prompt (its starting-prompt) to read that attendant's key map: each option-N's destination, the no-key-press and unassigned-key-press handling, dial-by-extension settings and the greeting's text-to-speech script. An option can lead to another prompt (a sub-menu); read it the same way.
voice_list_blocked_numbers
ChatGPTList blocked/filtered numbers for a domain (or one user) — calls matching these never reach the normal call flow. May time out on first call for larger/heavier domains; safe to retry once before treating as a failure — this is a read-only call and cannot have partially applied anything.
voice_list_call_queues
ChatGPTList call queues in a domain with their configuration and live stats (agents available, calls waiting). Pass a queue to read just that one. May time out on first call for larger/heavier domains; safe to retry once before treating as a failure — this is a read-only call and cannot have partially applied anything.
voice_list_dial_rules
ChatGPTList the dial rules of a domain's dial plan — how dialed digits map to destinations. The dial plan usually shares the domain's name; pass dialplan to read another. Warning: the full rule set comes back in one unpaginated response, and domains with ~100+ rules can exceed the tool-result size limit — expect a possible oversize failure on rule-heavy domains rather than treating it as an outage.
voice_list_domains
ChatGPTList the phone system domains (tenants) this account can read, with description, reseller, user count, and time zone. Paginated. detail:"full" returns the complete upstream configuration per domain (unscoped accounts only; anyone else reads one domain's configuration with voice_get_domain). The user count is the phone system's own figure and disagrees between endpoints — the response says how to get one you can explain.
voice_list_message_sessions
ChatGPTList SMS/MMS message sessions (conversations) for a domain or one user: participants, timestamps, message counts. Each row carries "messagesession-kind": "sms" for an SMS/MMS thread with an outside number, "chat" for an internal chat. The same session id can name both kinds on different users; the response flags that, and voice_read_messages then needs the matching user. Message text is stripped unless the domain's recording policy mode is 'full' (under 'redacted', read a conversation's masked text with voice_read_messages).
voice_list_phone_numbers
ChatGPTList phone numbers (DIDs) in a domain and where each one routes.
voice_list_queue_agents
ChatGPTList agents with availability/login state — one queue's agents when a queue is given, otherwise every agent in the domain across all queues. sub_portal_status carries the agent's free-text status message (busy reason). A personal account sees only its own agent entries; pair with voice_set_queue_agent_status to log in or out or go busy with a reason.
voice_list_queued_calls
ChatGPTLive snapshot of the calls currently waiting in a call queue, with caller and wait state. Warning: the payload scales with the queue's configured agent/ring-group roster, not with the number of waiting callers — a queue with a single waiting caller has returned a response large enough to exceed the tool output limit. For a staffing/wait snapshot prefer voice_list_call_queues' per-queue counts; reserve this tool for when per-caller wait-time detail is specifically needed.
voice_list_sites
ChatGPTList sites (office locations) in a domain. Site names can then filter CDR reads for per-location reporting. May time out on first call for larger/heavier domains; safe to retry once before treating as a failure — this is a read-only call and cannot have partially applied anything.
voice_list_sms_numbers
ChatGPTList the phone numbers enabled for SMS/MMS in a domain (or assigned to one user) and where each delivers. Not paginated: the upstream returns a single response with no paging parameters or truncation marker, so on a number-heavy domain the list may be silently incomplete — cross-check counts against voice_list_phone_numbers before treating it as exhaustive. Where the phone system refuses the domain-wide list for this account's login, the server reads each user's numbers instead and says so (derivedPerUser).
voice_list_time_frames
ChatGPTList time frames (business hours, holidays, after-hours windows). Domain-shared frames by default; pass a user for that user's personal frames. Time frames drive answer rules — read both to explain a user's or office's call handling. Note: entries come in at least four distinct shapes (days-of-week, specific-dates, holiday, and "always"-type with no schedule array) with no discriminator field, and old/stale frames (sometimes years old) sit alongside live ones with no staleness or expiry flag — cross-reference voice_list_answer_rules to see which frames are actually wired into live call routing.
voice_list_user_devices
ChatGPTList SIP devices registered to a user, including registration state — useful for troubleshooting a phone that isn't working. Strictly single-user: there is no bulk/domain-wide mode, so call it once per extension.
voice_list_users
ChatGPTList users/extensions in a domain — the company directory: extension, name, department, site, email, presence. Optionally filtered by a search string (name, extension, email). detail:"full" adds the complete per-user configuration; for one user's configuration prefer voice_get_user. Real domains interleave system users — auto attendants, call queues, time-of-day routes, ring groups — among the people. "service-code" tells them apart: empty for a person, and "system-aa", "system-queue", "system-tod", "system-ringgroup" and the like for a system user. Pass type "people" or "system" to filter on it; the response counts both on the rows it read. Test/demo accounts still look like people, so a person row is not proof of a staff member. Requires an office-manager or higher account (a personal single-extension seat is refused). Covers the company's own extensions only, never outside callers — a caller's name is on the call record (call-orig-from-name). May time out on first call for larger/heavier domains; safe to retry once before treating as a failure — this is a read-only call and cannot have partially applied anything.
voice_list_voicemails
ChatGPTList voicemail messages for a user (new or saved folder). In 'deny' or 'metadata' policy mode, audio links and transcript text are stripped from the listing; in 'redacted' mode, transcript text is included with identifiers masked while audio links stay stripped. In 'full' mode each message's audio comes as playback-url, a short-lived link (until playback-url-expires-at) that plays it in a browser; every play is recorded in the audit trail, and an expired link is replaced by listing again. Scoped to one user/extension — no domain-wide aggregate exists; to report a company-wide voicemail count, enumerate extensions via voice_list_users and call this once per extension.
voice_read_cdrs
ChatGPTRead call detail records for a domain (or a single user) within a datetime range. Returns caller/callee, direction, duration, and disposition. When call recording with transcription is enabled, rows also carry a "call-intelligence-job-id" usable with voice_get_transcription. Sentiment/topic analytics are included only when the domain's recording policy mode is 'full'. Filterable by call type, site, department, and partial caller/dialed number match. Paginated. A 'recordedContent' summary reports how many of the returned calls actually have a transcript, and when none do, whether that is the recording policy or the domain never capturing — read it before concluding transcripts are unavailable. Columns empty across every record are dropped and listed in omittedFields; on wide pages pass 'fields', which matters far more than the drop (a full CDR row is ~1KB). Pages are capped at 100 rows because larger ones come back without the call-id columns that recordings and transcripts are looked up by — walk longer spans with start/nextStart rather than a bigger limit.
voice_read_messages
ChatGPTRead the messages of one SMS/MMS session, including message text. Message content may contain sensitive personal information and requires recording policy mode 'full' or 'redacted' for the domain; under 'redacted', message text is returned with personal identifiers automatically masked and MMS media links are withheld. 'user' is the DOMAIN-side participant from the session's participants list: on-domain extensions appear as short user ids, external parties as phone numbers (pinned personal accounts can pass "~").
voice_set_queue_agent_status
ChatGPTACTION (changes upstream state): Set a call-center agent's status: log into or out of a call queue, or mark the agent busy with a reason code — action is "login", "logout", or a busy status ("break", "lunch", "meeting", "web", "other", or "cust1"–"cust8" for the domain's custom statuses as labelled in the portal). Applies to every queue entry the agent owns — a desk phone and a mobile app are separate entries — and changes nothing else: not queue membership, not any configuration. A busy status is agent-wide across all queues; time in each status is what the manager reports show (voice_get_agent_log returns per-status durations). The optional free-text reason is saved as the agent's status message, visible on the live roster (voice_list_queue_agents, sub_portal_status) and in the portal; it is NOT cleared by logging back in — pass an empty reason to clear it. IMPORTANT: login state and availability are separate settings on this platform, and the change is applied asynchronously, so a logout can be accepted while the agent still shows available for dispatch and keeps receiving calls. The response reports the roster afterwards plus receivingQueueCalls — report what it actually says rather than assuming the action took effect. A personal account can set only its own status; an office manager (or higher) can name any agent in the company. Confirm the agent and action with the user before calling. voice_list_queue_agents shows the roster and current state.
voice_update_user
ChatGPTACTION (changes upstream state): Update an existing user/extension's profile in a domain: first name, last name, and/or email. Nothing else — passwords, extensions, permissions, and billing settings cannot be changed here. Requires an office manager (or higher) account. Confirm the extension and the exact new values with the user before calling. At least one field is required; the response reports what changed, from and to.