When you add content to a Safe, Along does more than store the raw text. It reads every document, meeting transcript, and email, then extracts the meaningful things inside — people, companies, topics, deals, contracts — and records how those things relate to one another. The result is a knowledge graph: a structured network of nodes and typed edges that your AI assistant can traverse when answering questions.
The knowledge graph is maintained automatically as content is ingested. You never need to build or edit the graph manually — Along handles extraction and linking every time new content arrives in a Safe.
What the knowledge graph contains
Every piece of ingested content becomes a Document node in the graph. Along then reads that document and extracts entities — named things — and the relationships between them.
Entity types
The exact set of entity types depends on the preset and any custom configuration for your Safe (see Presets below), but the foundation includes:
| Entity type | What it represents |
|---|
Person | Anyone mentioned in a document: meeting attendees, email senders, transcript speakers |
Organization | Companies, firms, institutions, and other organizations |
Document | A document, transcript, email, or uploaded file |
Event | A meeting, call, or scheduled interaction |
Topic | A specific subject or theme discussed across documents |
Location | A geographic place or office location |
Deal, Product, Competitor, Objection, and ActionItem. A legal Safe adds Contract, Case, Clause, and Jurisdiction.
Relationships
Entities are connected by typed relationships that capture how they interact. A few core relationships present in every Safe:
| Relationship | Meaning |
|---|
MENTIONED_IN | A person, organization, or topic appears in a document |
WORKS_AT | A person is employed by an organization |
ATTENDED | A person participated in a meeting or event |
DISCUSSES | A document covers a particular topic |
RELATED_TO | One topic is related to another |
AUTHORED | A person sent or created a document |
ABOUT | A document involves an organization |
KNOWS | Two people appeared together in the same document or meeting |
Conceptual structure
Document <──(MENTIONED_IN)── Person ──(WORKS_AT)──> Organization
MENTIONED_IN. Those people are connected to their organizations via WORKS_AT. When a query touches one node, the graph can follow these edges to find related content several hops away.
How ingestion works
When you add a document to a Safe, Along processes it in two stages:
-
Entity extraction — Along reads the document and identifies entities according to the Safe’s configured types. A meeting transcript might yield several
Person nodes (speakers and attendees), an Organization (the company on the call), a Deal, and several Topic nodes for the subjects discussed.
-
Relationship linking — Along connects extracted entities to one another and to the
Document node. If a person is already in the graph from an earlier document, Along merges the new information into the existing node rather than creating a duplicate.
The graph grows incrementally. Each new document adds nodes and edges, and the connections across documents accumulate over time. A question about a person can eventually surface every meeting they attended, every company they work with, and every topic they’ve discussed — drawn from dozens of separate documents.
Use the ingestionInstructions field on your Safe to guide what the graph captures. You can tell Along which entity types matter most for your use case, what to prioritize or ignore, and how to interpret ambiguous content. For example: “Focus on deal stages and competitor mentions. Ignore internal HR content.”
How retrieval works
When your AI assistant queries a Safe, Along uses a two-phase retrieval strategy:
-
Vector search — Along finds the documents most semantically similar to the query. This returns a focused set of seed documents.
-
Graph expansion — Along takes the entities found in those seed documents and traverses the graph outward — up to 2–3 relationship hops — to find additional documents connected to those entities. This surfaces related meetings, emails, and records that vector search alone would miss.
For example, if a query returns a call transcript mentioning a particular deal, graph expansion can pull in the follow-up emails assigned to that deal, earlier discovery calls with the same company, and action items linked to the deal — all because they share entities in the graph.
The instructions field on your Safe shapes how these retrieved results are assembled into an answer: what to emphasize, what context to include, and how to present the response.
Presets
Along ships with six presets that configure the graph schema for common use cases. Each preset extends the universal base with domain-specific entity types and relationship patterns.
Universal
Included in every Safe. Provides the foundational schema: Person, Organization, Document, Event, Topic, and Location, connected by relationships like WORKS_AT, MENTIONED_IN, ATTENDED, DISCUSSES, and KNOWS. This preset is suitable for general-purpose knowledge bases and team wikis.
Sales
Adds entities for tracking revenue-focused activity: Deal, Product, Competitor, Objection, and ActionItem. Relationships include SELLING, EVALUATING, CHAMPION, DEAL_WITH, COMPETES_WITH, RAISED, and ASSIGNED_TO. Retrieval patterns prioritize finding all documents related to a deal, surfacing competitor mentions across accounts, and tracking open objections and next steps.
Finance
Adds Account, Transaction, Portfolio, and Instrument. Relationships include HOLDS, TRADED, ADVISES, and COUNTERPARTY_TO. Retrieval patterns connect clients to their portfolios and surface transaction history tied to accounts and advisors.
Legal
Adds Contract, Case, Clause, Court, and Jurisdiction. Relationships include PARTY_TO, GOVERNED_BY, CONTAINS_CLAUSE, FILED_IN, and PRECEDENT_FOR. Retrieval patterns connect case filings to courts, link contracts to their parties and governing jurisdictions, and traverse clause-level detail within agreements.
Healthcare
Adds Patient, Provider, Condition, Treatment, and Facility. Relationships include TREATS, DIAGNOSED_WITH, PRESCRIBED, TREATED_WITH, REFERRED_TO, and AFFILIATED_WITH. Retrieval patterns surface care histories, provider-patient relationships, and condition-treatment linkages.
Support
Adds Ticket, Bug, Feature, SupportProduct, and Customer. Relationships include REPORTED, AFFECTS, RELATES_TO_BUG, RESOLVED_BY, ESCALATED_TO, and FEATURE_OF. Retrieval patterns find all tickets reported by a customer, surface bugs affecting a given feature, and trace escalation paths.
Entity relationship traversal
The along tool exposes an explore_connections capability that lets your AI assistant traverse the graph directly. Given a starting entity — for example, a person identified by email, or an organization by name — explore_connections follows relationship edges up to a configurable depth (default 2, max 3) and returns the connected nodes and the relationship types linking them.
This is useful when you want to understand the broader context around a specific entity: who a person works with, what deals are linked to a company, what topics appear across a set of meetings, or which documents are related through shared entities. Your AI assistant can call explore_connections proactively as part of answering a question that requires mapping relationships rather than finding a specific document.