add_sequence_prospects
ChatGPTAdd prospects to a sequence. Each prospect is a Rox contact (rox_person_id), a Rox lead (rox_lead_id), or person data with at least an email or LinkedIn slug. Content is generated in the background after the call returns; poll GET /v0/sequences/{sequence_id}/prospects/{prospect_id} for status. A prospect already in the sequence is reported under duplicates.
cancel_prospecting_run
ChatGPTCancel a prospecting run. Stops a queued or running run. A run that already finished is left as it is. Partial results already written stay available on the poll.
create_sequence
ChatGPTCreate an empty sequence. Add steps with POST /v0/sequences/{sequence_id}/steps and prospects with POST /v0/sequences/{sequence_id}/prospects.
create_sequence_step
ChatGPTAdd one step to a sequence. Steps run in day order. If the sequence already has prospects, their content is regenerated to include the new step. Autosend is off for the new step; set it with PATCH /v0/sequences/{sequence_id}/autosend.
create_workflow
ChatGPTCreate a saved workflow, disabled by default. Webhook keys are shown once.
data_get_function_info
ChatGPTFetch parameter, return-type, and prompt info for one or more functions (POST /function_info). Each returned function-info object includes a prompt field — read it to learn what to do with the function's output (e.g. "Write the response to a file and dispatch a subagent to perform deep analysis on it, and call get_X_content to fetch their bodies"). The prompt encodes follow-up workflows that aren't obvious from the parameter or return-type signatures alone. Args: request.function_names: Names previously returned by list_functions. The response is a list of function-info objects in the same order. Batch related lookups in one call.
data_invoke
ChatGPTCall a Rox data_interface function (POST /invoke). PREREQUISITE: You MUST have called data_get_function_info for this function_name earlier in the conversation before calling data_invoke. Do not infer or guess the request DTO shape from the function name or from data_list_functions output alone — the parameter schema and the prompt field describing required follow-up behavior are only available via data_get_function_info. If you have not fetched it yet (or are not sure), call data_get_function_info first and then return here. Args: request.function_name: One of the names returned by list_functions. request.arguments: Must have the shape: {"request": { ... request DTO fields ... }} The request DTO is always wrapped under arguments.request, even when empty. Example: { "function_name": "get_companies_information", "arguments": { "request": { "rox_company_ids": ["2f5d2c0b-d949-4e50-8c4b-f76404b6022a"] } } }
data_list_functions
ChatGPTList all available functions on the Rox data_interface (GET /list_functions). Returns a list of {name, description} objects. Read each function's description to understand what it does and decide which one(s) to call — descriptions also include guidance on when the function is appropriate (e.g. "call this if the user asks about private company data"). IMPORTANT: After picking a function from this list, you MUST call data_get_function_info next — never call data_invoke directly. The info response carries the parameter shape, return type, and a prompt field describing required follow-up steps; invoking without it will almost always produce a malformed request or skip mandatory follow-ups.
enrich_company
ChatGPTEnrich one company using the authenticated organization's waterfall.
enrich_person
ChatGPTEnrich one person using the authenticated organization's waterfall.
get_customer_documents
ChatGPTList customer-specific document metadata with summaries for a company. Use this to browse customer-specific context documents for a company. Identify relevant documents from their metadata and summaries, then call get_document_with_chunks with their document IDs to retrieve full chunked content.
get_document_with_chunks
ChatGPTFetch chunked content for documents by ID. This function can return a large amount of content. Save the response to a file and dispatch a subagent to perform deep analysis on it, rather than reading the full content into your main context.
get_email_content
ChatGPTFetch full email content (body, recipients, metadata) by public ID. This function can return a large amount of content. Save the response to a file and dispatch a subagent to perform deep analysis on it, rather than reading the full content into your main context.
get_fiscal_quarter_end_dates
ChatGPTGet the org's configured fiscal quarter end dates (MM/DD).
get_note_content
ChatGPTFetch full content for a batch of notes by ID. This function can return a large amount of content. Save the response to a file and dispatch a subagent to perform deep analysis on it, rather than reading the full content into your main context.
get_org_chart_for_company
ChatGPTGet the org chart for a company. The org chart can return a large amount of data. Save the response to a JSON file and analyze the relationships by writing code to parse through the relationships encoded in the org chart JSON.
get_organization_documents
ChatGPTList organization-level document metadata with summaries. Use this to browse organization context documents that are not specific to a customer. Identify relevant documents from their metadata and summaries, then call get_document_with_chunks with their document IDs to retrieve full chunked content.
get_people_search
ChatGPTRead the current status and matched people for a Rox people search.
get_prospecting_run
ChatGPTPoll a prospecting run. Returns the run's status, prompt, limit, accounts, and one page of its matched and rejected prospects. Pass next_cursor back as after. A non-empty page always includes that cursor, for its last row. has_more is true when another page is ready now. When has_more is false, keep the cursor and poll again later for new rows. If a poll returns no rows, next_cursor is null: keep the cursor you sent. limit defaults to 200. Each prospect includes reasoning and the full profile. Only matched and rejected people are returned. counts covers the whole run. While the run is still going, a saved change can take up to a minute to appear. counts can change before those rows are on a page. Keep polling until the run is completed, failed, or cancelled. The same person can show up again; upsert by subject.linkedin_url. subject on each prospect is what you copy into POST /v0/enrich/person. Polling is safe within the rate limit and is not billed.
get_sequence
ChatGPTGet one sequence.
get_sequence_prospect
ChatGPTGet one prospect in a sequence.
get_transcripts
ChatGPTFetch full transcript content by transcript ID. Returns meeting/call transcripts with speaker attribution. You should call this function if the user asks anything about the private data of a company. This function can return a large amount of content — a single multi-hour meeting transcript can easily exceed 50k tokens. Save the response to a file and dispatch a subagent to perform deep analysis on it, rather than reading the full content into your main context.
get_workflow
ChatGPTGet saved configuration and trigger input requirements, not unpublished drafts.
get_workflow_run
ChatGPTGet metadata for your LIVE run while you retain access to its workflow. Does not return outputs or traces. Workflow owners cannot read other users' runs through this endpoint.
list_contacts_for_company
ChatGPTList CRM contacts for a customer. This can return a large number of contacts. Write the response to a file and parse it as JSON
list_customers
ChatGPTList all internal customers (existing accounts). This can return a large number of customers. Write the response to a file and parse it as JSON.
list_deal_stages
ChatGPTList the org's deal pipeline stages. Pair with list_deals to interpret deal flow.
list_deals
ChatGPTList deals visible to the current user. Call this when the user asks about deals in general — with no filter, it returns every deal the user can see across all companies. Pass rox_company_id to scope to a specific company, and start_date/end_date to keep only deals whose close date falls within that window. This function can return a large amount of content. Save the response to a file and write code to parse the JSON based on what the user is asking about — e.g., filter by stage, sort by amount, group by owner — rather than reading the full content into your main context.
list_emails_for_company
ChatGPTList email metadata (headers, recipients, timestamps) for a company. You should call this function if the user asks anything about the private data of a company. After calling this function, you should identify emails that are relevant to the user's question and call the get_email_content function to get the full body of those emails.
list_emails_for_user
ChatGPTList email metadata (headers, recipients, timestamps) for the current user. After calling this function, you should identify emails that are relevant to the user's question and call the get_email_content function to get the full body of those emails.
list_events_for_company
ChatGPTList calendar events for a company (includes transcripts when available). After calling this function, you should identify events that are relevant to the user's question. If transcripts are available, call the get_transcripts function with their transcript IDs to fetch full content. This function can return a large amount of content when transcripts are attached. Save the response to a file and dispatch a subagent to perform deep analysis on it, rather than reading the full content into your main context.
list_events_for_user
ChatGPTList calendar events for the user (includes transcripts when available). After calling this function, you should identify events that are relevant to the user's question. If transcripts are available, call the get_transcripts function with their transcript IDs to fetch full content. This function can return a large amount of content when transcripts are attached. Save the response to a file and dispatch a subagent to perform deep analysis on it, rather than reading the full content into your main context.
list_notes_for_company
ChatGPTList note metadata for a company. You should call this function if the user asks anything about the private data of a company. After calling this function, you should identify notes that are relevant to the user's question and call the get_note_content function to get the content of those notes.
list_notes_for_user
ChatGPTList note metadata for the current user. After calling this function, you should identify notes that are relevant to the user's question and call the get_note_content function to get the content of those notes.
list_sequence_prospect_tasks
ChatGPTList a prospect's tasks, one per step, in send order.
list_sequence_prospects
ChatGPTList the prospects in a sequence, oldest first.
list_sequence_steps
ChatGPTList a sequence's steps in the order they run.
list_sequences
ChatGPTList the sequences you own or collaborate on.
list_slack_messages_for_company
ChatGPTList recent Slack messages across all channels linked to a company. This function can return a large amount of content when multiple channels and threads are active. Save the response to a file and dispatch a subagent to perform deep analysis on it, rather than reading the full content into your main context.
list_workflows
ChatGPTList owned and shared workflows, including disabled ones, newest first.
open_rox
ChatGPTOpen Rox to search for people or enrich one person or company. Opening the app does not start a lookup or spend credits.
rox_fastapi_fetch
ChatGPTsearch_contacts
ChatGPTSearch internal CRM contacts by name or email across all companies.
search_customers
ChatGPTSearch internal customers (existing accounts) by name or domain.
search_people
ChatGPTFind people in Rox from a natural-language description and show them in an interactive table. Use this when the user wants to find, search for, or prospect people (e.g. "VPs of Engineering at Series B fintechs in New York"). It starts a Rox people search and the results load into the widget as they are found, with a link to continue in Rox. Pass the user's request as prompt.
search_prospects
ChatGPTSearch for prospect candidates. Send prompt, a natural-language description of who to find. Optional rox_company_ids limits the search to those Rox accounts and overrides any company named in the prompt. The response is a candidate pool, not a vetted list. completion is complete when the pool was filled; a narrow constraint such as a city may have been relaxed to fill it. incomplete_agent_stopped means the prompt was held and fewer were found. Send the rows as returned to POST /v0/prospects/validate with your fit criteria as prompt, then send each matched subject to POST /v0/enrich/person. credits_used is 1 per 10 candidates returned, rounded up; 0 on 422 or 503.
show_sequence
ChatGPTShow a Rox sequence's steps, prospects, and autosend in an interactive widget with Open in Rox. Call this after creating or changing a sequence, or when the user asks to see one.
start_prospecting_run
ChatGPTStart a prospecting run. Send prompt describing who to find. Scope it with rox_company_ids and/or domains. Domains resolve to accounts that already exist in the organization. Omit both for an open search. A domain that matches nothing is listed in unresolved_domains when the run still has another account. If every domain fails and no account id was sent, the request is 400. A LinkedIn company page or numeric company id is not an input. The account's page is resolved for the run. Unknown fields are rejected. discovery_mode is standard (the default) or premium. A standard run stops at target_limit (1 to 1000, 100 when omitted) and searches 5 times that limit. A premium run is the deep search for one account: it judges until 2,000 matches or its pool runs out, with no limit on rejections, from up to 20,000 people found by broad searches plus everyone whose title is the requested role. Premium needs exactly one account and no target_limit (400 otherwise). Both modes are echoed as discovery_mode on the start and poll responses. The response is 202. Location is the poll URL. start_status is started, pending when the run was accepted and has not started yet, or failed. Poll GET /v0/prospects/runs/{run_id} until status is completed, failed, or cancelled. There is no completion notification.
start_workflow_run
ChatGPTStart one LIVE run, including for a disabled workflow, as yourself. Uses the saved trigger and published version without changing enablement or schedules. Returns after bounded run registration, not execution completion; the reported status may already be terminal. HTTP requests are free; LIVE execution is billed normally. No idempotency guarantee: every retry can duplicate execution, including after a startup timeout or error. A startup error may mean the workflow has already started. No final output guarantee.
update_sequence
ChatGPTUpdate a sequence's name, tags or send priority. Does not regenerate content. send_priority is your own setting, so a collaborator can change it; name and tags are the owner's. Set autosend with PATCH /v0/sequences/{sequence_id}/autosend.
update_sequence_autosend
ChatGPTSet whether a sequence and its automated steps send automatically. Autosend is your own setting, so a collaborator can set theirs. Only what you send changes; the response has your full autosend state.
validate_prospects
ChatGPTValidate prospect candidates. Send prospects copied from POST /v0/prospects/search as returned. Each prospect's current employer is verified first. Always send prompt with your fit criteria: it decides matched or rejected on role, seniority, geography and anything else you state, with a one-sentence reasoning, and rejects people who have left. Without prompt only current employment is checked and everyone still employed is matched. Send rox_company_ids from search to require current employment at one of those accounts. Send each matched subject to POST /v0/enrich/person. credits_used is 25 per matched and 3 per rejected; failed prospects are not billed. 402 if the batch cannot be covered at 25 per prospect; 503 if every prospect failed.