More than storing memory,
make it trustworthy and reusable.
Manage one organization’s knowledge in each installation. Follow membership approval, Memory, documents, Graph, and agent connection from start to finish.
01 · Get started
Start with a membership request and approval.
Signing up requests membership in this installation. An operator must approve it before you can read or save organization knowledge. There is no separate organization to create or choose.
Refresh the approval screen to check your status. The installation administrator configures the first operator. Blocked or removed accounts need an operator to restore access.
Choose a default sharing scope when saving.
Read · organization members
Manage · admin · owner
Read · team members and organization administrators
Manage · team manager or organization admin/owner
Read · you
Manage · you
Organization-wide writes and management require admin or owner access. Team members and organization administrators can access team content; personal content defaults to its owner. Explicit Memory ACL grants can also permit named users or teams to perform specific actions.
02 · Context search
Choose the search surface that fits your question.
Choose All knowledge to search Memory, document chunks and Knowledge nodes together, or narrow the result type. Select a result to read its full content and sources without losing the list.
Search diagnostics are optional details. Relative relevance compares results within this search; it is not a confidence score. Archived, expired and future Memory and unprocessed documents are excluded.
When configured, reranking applies to All knowledge and MCP recall. Memory, Documents and Graph searches use lexical and optional semantic scores. If reranking fails, results use hybrid ranking.
03 · Manage Memory
Keep the current decision and why it changed together.
Open Memory in the sidebar and choose New Memory to record a decision, rule or useful fact with a sharing scope. Select an item to open Content and source. Users with the required permission can edit it and inspect Version history.
A change reason is optional. Editing preserves the previous version. If another person saves first, reload the current version before retrying. Archive creates a new version and removes the Memory from ordinary search.
04 · Ingest documents
Preserve originals and search their chunks.
Upload UTF-8 text, Markdown, CSV, JSON, or XML from Documents. Choose a sharing scope and inspect processing status and source text. Ready means document search is available; AI knowledge extraction and verification continue separately.
pendingprocessingreadyPending and processing documents refresh automatically while the page is visible, for up to two minutes. Use Refresh after polling pauses. For a failed document, inspect the error and choose Retry processing when you have write permission. Archiving requires manage permission.
05 · Explore Graph
Follow relationships without losing the evidence.
Open Knowledge Graph from the sidebar or explore relationships from a search result. Search for a node, select it in the map or node list, then inspect connected relationships and original sources. Use the selected node as a new exploration center.
Drag nodes to pin their positions, drag the background to pan, and scroll or pinch to zoom. Use fit, reset layout, or full screen; press Escape to leave full screen. Keyboard node navigation and readable source evidence remain available.
Graph progress tracks extraction, verification, and automatic processing, not completion of human review. Search prioritization targets unfinished candidates whose extracted names or aliases match. Refresh relationships to load new connections.
06 · Review AI candidates
Automatically apply qualified knowledge and review uncertain items.
With a knowledge extraction model and document worker configured, a separate AI verification checks source quotations, usefulness, conflicts, and ontology rules. Repeated entities and relationships in the same scope are grouped with their sources. AI judgments and quotation checks do not guarantee factual accuracy.
Automatic acceptance
Apply explicit, useful items that pass source evidence and policy checks to the Graph, retaining their sources.
Needs review
Compare reasons and source text for uncertain, conflicting, or ontology-constrained items, then select which sources to apply.
Automatic exclusion
Leave unsupported or incidental items and relationships that do not identify a specific fact out of the Graph.
Accept or reject knowledge from selected sources. Accepting a relationship includes its required endpoints; rejecting an entity also rejects its unreviewed relationships. Other items remain pending, and rejection does not delete previously accepted Graph data.
Run AI curation queues existing unverified candidates and unfinished automatic decisions. Use chunk review for remaining extraction batches and curation history for the latest 50 assessments and automatic or human decisions. Empty extractions stay out of the default review list.
07 · Connect Agents
Connect Agents to Context through MCP.
Services connect to /api/mcp to remember, recall, and forget long-term memories. Agent Memory also manages RAG documents and Knowledge Graph. Use context_search to search all three together. Agent connection provides the endpoint, credentials, and an optional Agent Studio registration template.
/api/mcp
An organization admin or owner can manage an Agent token. Without user delegation, it permits organization scope only. A trusted service can send X-User-Email to act with an active member’s permissions. In Studio, bind the server to a version and enable memoryRecall with recall allowed for automatic Memory recall.
context_searchrecallrememberforgetdocument_searchknowledge_searchknowledge_neighborhoodExplore real context now.
Start by creating a Memory or uploading a document. Search it, check the original evidence, and connect an Agent when the knowledge is ready.