add_approved_answer
ChatGPTSave a finished question-and-answer pair to the organisation's Approved Answer Library, so Iris reuses it when answering future questionnaires. Requires the 'write' permission — if the connection was authorized before write access existed, this call fails with instructions to reconnect — and the user's own permission to edit the library. The answer is placed the way the organisation's library is configured: it goes live only if the library skips the review inbox, and otherwise lands in the inbox for review. Pass status 'draft' to save it unpublished instead. Only call this for an answer the user has reviewed and asked to save — confirm the exact question and answer wording with them first. If the library already has an answer to the same or a similar question the call saves nothing and returns those answers instead: show them to the user and call again with allow_duplicate=true only if they still want a new one. Use this for 'save this answer to the library', 'add this as an approved answer tagged Security', or similar requests.
add_file_to_knowledge_map
ChatGPTAdd a file or raw text document to the organisation's Knowledge Map. Requires the 'write' permission — if the connection was authorized before write access existed, this call fails with instructions to reconnect. Provide exactly one of file_content_base64 (real file bytes) or text_content (raw text/markdown, which Iris converts to a .md file automatically). The document is indexed asynchronously; this tool returns immediately with a processing status. Use this for 'add this PDF to our knowledge base', 'save this text as a document tagged Pricing', or similar requests to add file/text content.
add_url_to_knowledge_map
ChatGPTScrape a web page — or crawl a small site starting from it — and add the result to the organisation's Knowledge Map. Requires the 'write' permission — if the connection was authorized before write access existed, this call fails with instructions to reconnect. By default only the given URL is scraped (single page). Pass depth_level > 0 to crawl linked pages too, up to 3 hops and 50 pages — a deliberately small cap so this tool can't be used to kick off a large, unsupervised crawl; ask the user before crawling and confirm the page/depth limits with them. The content is scraped and indexed asynchronously; this tool returns immediately with a processing status. Use this for 'add this page to our knowledge base', 'scrape https://example.com and tag it Security', or 'crawl our docs site and add it', or similar requests to add web content.
approve_project_answer
ChatGPTRecord the user's own approval of one answer in an Iris project, exactly as approving it in Iris does. Requires the 'projects:write' permission. Only call this when the user asks to approve. It approves for the user only, never for someone else, and only an answer they are an assigned reviewer on — it never reassigns the answer to them. When a workflow governs the answer, the approval moves it to the next step (status Tentatively Approved) and it is Approved only after the final step. The result says which step the answer is on, whether it is now fully approved, and whether it was added to the Approved Answer Library (live, in the review inbox, as a draft, or not added) — tell the user. Use this for 'approve this answer' or 'send it to the next step'.
ask_iris
ChatGPTAsk Iris AI a natural language question about your organisation's knowledge base. Searches indexed documents, approved answers, and previous Ask Iris history, then returns an AI-generated answer with cited sources and a confidence score.
export_approved_answers
ChatGPTExport the approved answers library — all of it by default, or the subset the same filters as search_approved_answers select — as a file, and return a download link. The server builds the file; you do not assemble it from rows. With no tags argument this exports the caller's entire visible library, including answers that carry no tags. It does NOT fall back to the organisation's default tags the way search does: someone asking for the library means all of it. Use this whenever someone asks for the library as a spreadsheet, a CSV, or a file of any kind, and whenever a request covers more answers than are worth reading into the conversation. Do not page through search_approved_answers to build a file yourself — that is what this tool is for, and it has no size limit you need to reason about. Most exports come back ready with a download_url in one call. Give the user that link exactly as returned and stop — it is already the final download link and stays valid for 24 hours. Do not call this tool again to confirm it. A very large one returns status 'queued' with an export_id: the link is emailed when it finishes, and calling this tool again with that export_id fetches the link — that is the only reason to pass export_id. Only one export runs per organisation at a time; status 'busy' means try again shortly.
get_crm_opportunity
ChatGPTCheck whether a project is linked to a Salesforce or Microsoft Dynamics opportunity. Queries both CRM integrations in parallel and returns the link status for each, including the opportunity ID if linked and whether the integration is active. Use this for questions like 'is there a Salesforce opportunity connected to project 94?', 'what CRM deal is linked to this project?', or 'check if project X has any CRM opportunity attached.'
get_custom_instructions
ChatGPTList all custom instructions configured for the organisation. Custom instructions define how Iris AI should format and style its responses — including tone, length, structure, and perspective preferences. Use this to understand the organisation's preferred response style before calling ask_iris, or to answer 'what custom instructions do we have?', 'what is our default response style?'
get_executive_summary
ChatGPTGet an executive-level reporting summary for the organisation. Returns total projects, total deal value in the pipeline, average answer revisions per question, average time to complete a project, and average AI accuracy. Requires reporting access permission. Use this for 'give me an overview of how we are performing', 'what is our total pipeline value?', or 'how accurate is Iris on our projects?'
get_knowledge_base_health
ChatGPTGet health statistics for the organisation's knowledge base (approved answers library). Returns total approved answers, how many are pending review, how many are in the inbox awaiting processing, active review cycles, and overdue review cycles. Requires approved_answers_full_access permission. Use this for 'how healthy is our knowledge base?', 'how many answers need review?', or 'are there any overdue review cycles?'
get_my_assignments
ChatGPTGet questions that are currently assigned to the authenticated user across all projects. Returns the total count and a breakdown by project along with details of each assigned question (text, status, project name, and all current answers including text and dropdown selections). Use this to answer 'how many questions am I assigned to?', 'what do I need to work on?', or 'which projects have open tasks for me?'
get_project_detail
ChatGPTGet detailed information about a specific project, including its status, owner, reviewer, due date, and a full breakdown of question counts by status (approved, needs approval, answer pending, etc.). Use this to answer questions like 'how many questions are in project X?', 'who owns the ACME project?', or 'what is the status of project 123?'. Either project_id or project_name must be provided; if neither is given, an error is returned.
get_project_questions
ChatGPTList or search the questions in one Iris project, together with their current answers. Returns each question's ID, text, section, status, confidence, assignees and answers (long text answers are trimmed to a preview — use get_question_answers for the full text of one question). Either project_id or project_name must be provided; use list_projects to resolve a name to an ID. Results are paginated. Use this for 'what questions are in the ACME project?', 'show me the unanswered questions in project 123', or 'find the questions about encryption in the Globex RFP'.
get_project_stats
ChatGPTGet high-level statistics about projects in Iris. Returns counts of projects grouped by status (e.g. In Progress, Review, Submitted), plus project hub aggregate stats (total workspaces, overdue, due today). Use this to answer questions like 'how many projects are in review?', 'how many projects do I have?', or 'are there any overdue projects?'
get_question_answers
ChatGPTGet one question in an Iris project with its answers in full — never trimmed, including dropdown selections, per-answer status, confidence, cited source file names and assignees — plus the question's section, tags, custom instructions and context. At most 20 answers are returned; total_answer_count gives the true number and answers_truncated is true when the question has more than were returned. Requires both project_id and question_id; use get_project_questions to find question IDs. Use this for 'what is the answer to question 456 in project 123?', 'show me the full response we gave for the encryption question', or 'which sources were cited for that answer?'.
get_rfp_pipeline
ChatGPTGet aggregate RFP/project pipeline statistics. Returns total number of RFPs, how many have requirements extracted, total requirement count, and how many requirements are evaluated (met, partial, not met, pending, unknown). Requires projects_feature permission. Use this for 'how many RFPs are in the pipeline?', 'how many requirements have been evaluated?', or 'what is our evaluation breakdown?'
get_win_loss_stats
ChatGPTGet win/loss deal statistics for the organisation. Returns total deals, won and lost counts, win rate percentage, total won deal value, and lost deal value. Optionally filter by deal status or date range. Requires reporting access permission. Use this for 'what is our win rate?', 'how many deals did we win?', or 'what is our total won deal value this year?'
list_projects
ChatGPTList or search projects in Iris. Returns each project's name, ID, status, due date, deal owner, reviewer, client name, and question counts. Use this to find projects by name, filter by status, or see what projects exist. Also useful for resolving a project name to its ID.
list_tags
ChatGPTList all knowledge-base tags available in the organisation. Returns each tag's ID, name, description, and whether it is a default or restricted tag. Use this to discover available tags before calling ask_iris with tag filters, or to answer 'what tags do we have?', 'what categories exist in our knowledge base?'
search_approved_answers
ChatGPTSearch and read the organisation's approved answers library — the reusable question/answer pairs Iris draws on. Returns matching answers with their tags, status, owner, source project and dates. Pass tags when the user names a subject area, or when a first search comes back mixing unrelated domains — the library spans everything the organisation has ever answered on, so a question about encryption can pull sales and legal answers alongside the security ones. Narrowing by tag is the fix, and list_tags shows what is available. Tags are not required: with none supplied the organisation's default tags apply automatically, so do not stop to ask before a straightforward search. Answer text is previewed to 1500 characters by default. Pass answer_max_chars=0 for the full text — with uuids when you want a few specific answers whole, or on its own when you are collecting everything. Results are paged: when has_more is true, call again with cursor set to next_cursor. A page can be shorter than page_size when the answers are long — that is normal, and next_cursor is still correct. Use this for 'what do we say about SOC 2?', 'show me our approved answers tagged Security', or 'find the answer we gave about data residency'. If the user wants a file — a spreadsheet, a CSV, a download — use export_approved_answers instead. It builds the file server-side and returns a link, with no size limit to reason about. Do not page through this tool to assemble one.
update_project_answer
ChatGPTReplace the text of one answer in an Iris project with an improved version. Requires the 'projects:write' permission — if the connection was authorized without it, this call fails with instructions to reconnect. Before calling, show the user the exact final text and get their explicit confirmation; never save text they have not seen. Saving does not approve: never call approve_project_answer in the same step unless the user asks. answer_text is plain text or Markdown (headings, lists, bold, links, tables), not HTML. Pass the answer's edit_token from get_question_answers; if the answer's text changed since (an edit or a regeneration), nothing is saved — read it again, show the user what changed, and confirm before retrying. Only text answers can be edited, and only answers get_question_answers marks editable (not an answer that already has approvals, nor one assigned to someone else when the user is a Collaborator). When get_question_answers reports has_rich_content, the new text drops the answer's images, tables and links (the assistant only sees their text, not the URLs): tell the user and set replace_rich_content only once they agree. Use this for 'save this improved answer back to the project' or similar.