Meetings and memory
Meeting prep, the knowledge base Zero and the agents answer from, and what ZeroReach remembers about each lead and company.
Meetings
Meetings are booked by the voice agent on a call (inside the workspace's business hours) and appear on the agenda. GET /api/v1/agenda lists booked meetings and scheduled AI callbacks between two times. The default range is now to 7 days ahead, and the longest is 92 days.
ZeroReach has no calendar integration for meeting prep. It works from what ZeroReach already knows.
Meeting prep
The Meeting copilot writes a brief for a booked meeting from our own records: the meeting, the lead's emails, replies, calls, notes, research and memory, plus the workspace's own prep instructions. The brief covers:
- who they are and why they replied;
- prior emails, calls and notes;
- open questions;
- a suggested agenda, talking points and risks.
| Endpoint | Who |
|---|---|
POST /api/v1/ai/meeting-prep | Writes or refreshes the prep. Nothing is sent to anyone. |
GET /api/v1/meetings/{meeting_id}/prep | Any member: the meeting and its prep (prep: null until one is written). |
GET / PUT /api/v1/meeting-prep/settings | The workspace's prep instructions. Anyone can read them; admins set them (at most 2000 characters, null or blank clears). |
Writing a prep is one AI call within the free-tier budget. It answers 503 when AI isn't available, and 409 agent_paused while the Meeting copilot is paused or off.
Knowledge base
The knowledge base holds what the agents may say about your company: your offer, FAQs, pricing and answers to objections. The Writer uses it in emails, and the voice agent searches it for approved answers on calls. When nothing matches, the voice agent is told not to guess and to promise a follow-up.
| Endpoint | Adds |
|---|---|
POST /api/v1/kb/documents/text | Pasted text, such as an offer description or an FAQ. |
POST /api/v1/kb/documents/upload | A text, Markdown, HTML or PDF file. |
POST /api/v1/kb/documents/url | A website crawl: same site only, robots.txt honoured, limited pages. It runs in the background; poll GET /kb/documents/{document_id} until it is ready or failed. |
Documents are split into chunks and embedded. GET /api/v1/kb/search runs a semantic search, best match first. With an API key, use kb:read and kb:write.
Memory
ZeroReach keeps a knowledge graph per workspace: people, companies and concepts, with dated, cited facts.
- From your data, without AI: lead changes and timeline events become people and company nodes,
works_atlinks, sequence enrolments and touches. - From text, with AI: call transcripts, classified replies and research runs are queued for fact extraction. When the AI budget is used up or AI isn't configured, extraction stops, and the sources wait for a later run.
- Signals add facts such as
hiring_for,uses_toolandfunding_event.
A newer value of a single-valued fact replaces the current one, and the old value keeps its validity dates.
Read memory with:
GET /api/v1/leads/{lead_id}/memoryGET /api/v1/companies/{company_id}/memory
Each answer has current facts (with confidence), open objections, relationships and recent interactions. Every item cites its source: a timeline event, a call, a classified reply, a research run or the lead record. The text field holds the same memory as a compact block, which is exactly what the AI writer and the voice agent see.
Zero and MCP clients read the same memory with the get_memory tool.