SuperMe

Ask a question and get an AI-generated answer. Say what kind of target you mean: `person` (individuals, drawn from their indexed content), `workgroup` (a saved group you own, expanded to its members), or `community` (a community or company, synthesized across members). They can be combined, and participants are asked in parallel: one answer for a single participant, one perspective each for several. At most 5 profiles are asked per question, counting each workgroup member individually. Everyone passed is looked up; people and the community you name are always asked, and the workgroup members whose profiles best fit the question fill the remaining places. Re-asking the same question reaches the same profiles; only when the user asks for new perspectives, pass the earlier answers' target_user_id values in `exclude` to reach the next ones. Each parameter searches only its own kind. When one matches nothing, nothing is sent to anyone — not even the targets that did match — and the reply is `person_not_found` / `workgroup_not_found` / `community_not_found` naming only the one that missed. Fix that one and retry the whole call. After a `workgroup_not_found`, do NOT retry the same string as `person`: a group name passed there will match some unrelated individual and they will be sent the question. Show the user the listed workgroups and let them choose. Make one ask call listing all candidates (the server picks who is asked) and ask one broad neutral question each can answer from their own view; do not split by person or stance or make extra calls to reach more people, and compare and synthesize after results return. If names must be resolved first, batch them in one user_profile_search list call. Pass conversation_id only for one single-participant thread. When the answer takes longer than the tool's wait budget, the call returns status "in_progress" instead of the answer; the work continues in the background. Read the outcome with conversation_result, which holds until the SuperMe conversation has finished or its wait budget runs out. Calling ask again with the same conversation_id only fetches the pending result — the question is not re-sent, so checking can never create a duplicate message.

ask

Create a new workgroup owned by the authenticated user. A workgroup is a saved set of SuperMe users that you can reference later (e.g. by passing its handle to `ask` as `workgroup`). The handle must be unique within your account — if a workgroup with this handle already exists, the call returns an error. Members can be added at creation time or later via `workgroup_edit`.

workgroup_create

Find people who know about a given topic in the SuperMe network. Returns a ranked list of people with relevant expertise, including a short explanation of why each person was selected. Use this when you need to identify who knows about a subject — e.g. before deciding who to ask a detailed question to with `ask`. For one user question, make one search with the full question, even if it has multiple angles or tradeoffs; do not split into separate searches per angle. If asking several returned people for takes, call `ask` once with all of them in `person` and one broad neutral question each can answer from their own view. Do not make separate ask calls just to tailor wording to subsets; compare and synthesize after results return. Pass `community` (a community handle, name, or id) to restrict the search to the members of that community.

user_expert_search

List the files this user has uploaded to SuperMe through the upload panel, with file_id, name, type and size, newest first. Use the file_id values with ask_my_agent(file_ids=[...]) to attach files to a SuperMe agent turn. If the list is empty, call file_manager to let the user upload one.

list_files

List the authenticated user's workgroups. A workgroup is a saved set of SuperMe users (e.g. a project team or advisory board) that the owner has assembled for repeated reference. Returns all workgroups owned by the caller, sorted by most recently used first. If the user asks to ask a workgroup that is not present, do not substitute a similar group or call `ask`; show the available closest workgroups and ask for confirmation first.

workgroup_list

Read one user's full public profile by user_id or username. Use this for meeting prep, profile summaries, and comparing what named people are known for. Returns identity plus the un-truncated summary, structured work experience, education, and skills — the depth a search card's short bio doesn't carry. Resolve a name to a user_id or username with `user_profile_search` first. Do not call `ask` for this kind of profile-context task unless the user explicitly asks for the person's take/advice/perspective. Returns an error if the profile is missing or not visible to you (private / outside your network).

user_details_read

Read a pending MCP conversation turn without posting or charging quota. Returns the status (in_progress, success or failed), seconds elapsed, the conversation_url and the answer once it exists. Only conversations started by this caller through ask or ask_my_agent are readable. To collect an answer in the same turn, call it and repeat while the status is in_progress; each call returns as soon as the conversation has finished.

conversation_result

Search for SuperMe users and workgroups by name or identifier. **Single identifier** returns an object with two keys: - users: up to limit matches, each with user_id, name, title, location, avatar_image, and in_network. Users you follow are ranked first. - workgroups: matches from your own workgroups, each with handle, member_count, and members (list of user_id + name). A workgroup hit means the query refers to a saved group, not an individual. **List of identifiers** returns results (one entry per identifier with its matched users), resolved_user_ids, and unresolved. Use one list call when resolving multiple named people for the same request; do not split them into separate search calls. Routing: a users hit — pass the user_id to ask as `person`. A workgroups hit — pass the handle to ask as `workgroup`, never as `person`; it expands to the members (a single answer for a one-member group, one parallel perspective per member for more).

user_profile_search

Internal sink for the SuperMe upload panel — the panel's own Upload button calls this with the dropped files. Do not call it yourself: to let the user upload a file, call file_manager and let them drop it in. It takes raw base64 payloads, so calling it directly is never the right way to get a user's file into SuperMe.

981f60a7a215_store_files

Use the user's SuperMe My Agent for personal intelligence, network intelligence, and actions. My Agent works on the user's behalf using their private knowledge, professional network, SuperMe profile and library, and connected tools. It can: Personal Intelligence - Get personal takes, advice, and answers grounded in the user's profile, private knowledge, notes, saved insights, work history, meeting context, prior conversations, articles, and podcast or video takeaways. - Search and synthesize across connected sources such as Notion, Drive, Dropbox, and Gmail so the answer reflects the user's own context rather than only public information. - Personal Intelligence is fast and is appropriate for quick, context-aware takes as well as deeper synthesis across the user's own knowledge. Network Intelligence - Discover relevant people and experts, search the user's professional network, inspect or compare profiles, and gather perspectives for research or decisions. - Run multi-turn AI group chats with individual people or saved workgroups, synthesize the participants' perspectives, and create, read, or update reusable workgroups such as advisory boards, hiring panels, or research groups. - IMPORTANT: Unlike fast Personal Intelligence requests, Advanced Network Intelligence usually takes longer than a regular ask because it can orchestrate planning, multi-step research, tool calls, and group chats. For a quick expert perspective, use `user_expert_search` to find relevant experts, then use `ask` to ask them. If the person, community, or workgroup is already known, use `ask` directly. Profile Management - Read and update the user's own SuperMe profile, including their name, professional headline, About summary, education, and work experience, or have SuperMe regenerate their profile summary from eligible profile intelligence. Library Management - Search, read, list, edit, or delete saved insights; filter the library by source, date, type, privacy, or title; change insight titles, content, or privacy; and fetch an insight's original connected-source content when available. - Create, read, update, organize, or delete named notes in the user's library. Research, Communications, and Ongoing Work - Browse and read the web and GitHub, inspect uploaded files, manage the user's conversation inbox, reply to conversations, and delegate longer research or actions to background subagents. - When the relevant connection, permission, and feature are available, work with connected apps and remote MCP tools, create Drive content, send Gmail or WhatsApp messages, create or manage recurring routines, and schedule a check-back. Send the user's request in question and relay My Agent's response. To resume across multiple turns, the calling assistant or agent must pass the same returned conversation_id on every follow-up. When the user has uploaded files through the SuperMe file-upload panel, attach them with file_ids rather than reading them and pasting their contents into question — My Agent reads the file itself, so images and PDFs stay intact and large files never enter this conversation. When a turn outlives this tool's wait, the call returns status "in_progress" instead of the answer; the work continues in the background. While that answer is pending, calling this tool again with the same conversation_id only waits for and fetches the result — the message content of such calls is not posted, so checking can never create a duplicate message. To fetch the outcome, wait a minute or two, then call this tool again with the same conversation_id. Send new instructions only after the pending answer has arrived.

ask_my_agent