griasa
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
Griasa holds the user's recorded meetings, the promises made in them, and what is known about each colleague. For everything about one person, project or topic in one call, use get_context. Start with list_commitments or get_next_meeting_brief for "what is open" and "what is next"; list_decisions for what was decided and why; search_meetings and get_meeting for what a meeting said; ask for get_transcript only when a summary is not enough, since it returns everything that was said. These tools only read. If the user has allowed it, add_conversation offers a chat for their review and propose_close_commitment suggests that a promise is done; neither reads, changes or deletes anything, and the user decides.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_commitmentsA | Open promises from recorded meetings, newest first, in two lists: 'mine' (what the user owes) and 'waiting_on_others' (what others owe the user); finished ones are left out. Use for "what do I owe", "what is overdue", "what is Anna still waiting on". For one colleague's whole picture use get_person; to get a name's exact spelling first, list_people. |
| list_decisionsA | Decisions taken in recorded meetings, newest first: what was decided, why, who made the call, the options not chosen, and the exact words it was decided in. Use for "what did we decide about X", "why did we choose Y", "did we ever consider Z". Reversed decisions are included and marked; leave them out with standing_only. |
| get_personA | Everything known about one colleague: notes, email addresses, their open promises, and the last recorded meeting with them. Accepts a name or email in any spelling; if nobody matches, returns the known names. |
| search_meetingsA | Find recorded meetings by words in the title or text, by who took part (one person or several), or both. Words match in any order and in any grammatical form, and, when the user has turned on search by meaning, by meaning across languages; best matches first, otherwise newest first. Returns id, title, date, participants, summary and, when text matched, a short 'matched' excerpt — never the transcript. Use the id with get_meeting or get_transcript. |
| get_meetingA | One meeting by id: summary, participants and the promises made in it. Does not include the transcript; use get_transcript for what was said. |
| get_transcriptA | The full transcript of one meeting: everything said, including remarks in passing. Long and sensitive; call it only when get_meeting's summary is not enough. |
| get_next_meeting_briefA | Prep for the user's next calendar meeting: attendees, what each side promised the other, and what was discussed last time. Built from the calendar and recorded meetings; returns available=false with a reason when no meeting is coming up. Use before a call; for a meeting that already happened use search_meetings and get_meeting. |
| list_upcoming_meetingsA | The user's calendar meetings in the next days, soonest first, each with its attendees and the open promises between the user and them: what the user promised them and what they promised the user. Use for "what's my week", "what do I owe the people I'm meeting". Only events with other people or a call link; all-day events are left out. Needs calendar access; returns available=false with a reason without it. For the very next call in detail, get_next_meeting_brief. |
| draft_followupA | A follow-up email for one meeting, written from Griasa's notes: the summary and who promised what for all participants, or, with person, only what that person and the user promised. Returns subject, body, the addresses Griasa knows and a mailto link. Griasa sends nothing: hand the link to the user or send the draft with the user's own mail tool after they confirm. Meeting id from search_meetings. |
| get_contextA | Everything Griasa knows about one person, project or topic, in one call and within a size budget: open promises in both directions (overdue first), recent meetings with their summaries, and for a person the next meeting with them. Use it before answering a question about someone or something; follow up with get_meeting or get_transcript only for detail it leaves out. When the budget is tight, the oldest meetings are left out first and |
| list_peopleA | Everyone Griasa knows by name, with how many of their promises are still open. Use to find the exact spelling before get_person or list_commitments. |
| list_projectsA | List every project the user files recordings and notes under, with id and name, in the order the user keeps them in Settings; unfiled items sit in an Inbox that is not listed. Use to answer what the user is working on or to match a meeting to a project; pass an id as project to search_meetings or list_commitments. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 12 tools
Most tools occupy clearly distinct slots: search_meetings finds ids, get_meeting summarizes, get_transcript expands, list_* enumerate different resource types. The one real overlap is get_person versus get_context, which both promise 'everything about a person'; the descriptions disambiguate (person-only whole picture vs. person/project/topic within a size budget), and list_commitments also partly overlaps get_person but is explicitly scoped to the two-direction open-promise view. A single call could still be misselected in a hurry, but boundaries are documented.
Every tool follows a verb_noun snake_case pattern (get_person, list_people, search_meetings, list_commitments, draft_followup). The only outlier is the longer get_next_meeting_brief, which still fits the get_ prefix convention. Verb choice also tracks intent consistently: get_ for single-entity retrieval, list_ for enumeration, search_ for matching, draft_ for generation.
Twelve tools sit comfortably in the well-scoped range, and each earns its place: retrieval tiers (search/get_meeting/get_transcript), two calendar tools, people/projects/commitments/decisions listings, context aggregation, and one drafting tool. There is no obvious filler or duplicate entry, and the layering of summary-to-detail tools is deliberate rather than redundant.
For a recall-and-prep assistant the read surface is thorough: search, detail, transcript, promises, decisions, calendar prep, projects, cross-entity context, and follow-up drafting. The gap is on the write side — there is no way to mark a commitment done, record a note, or attach a meeting to a project, so the agent can only advise the user to act elsewhere. That is a workable limitation given the product appears to ingest recordings automatically, but it leaves a lifecycle dead end.