The Open Semantic Interchange (OSI) is a vendor-neutral, open-source specification for portable semantic model definitions. It gives BI tools, AI agents, and warehouses a shared YAML/JSON format for datasets, relationships, fields, and metrics, so the same definitions can travel across tools without manual reconciliation. Honeydew is one of the founding partners of OSI, alongside Snowflake, Salesforce, and dbt. Portable semantics are core to how Honeydew approaches a semantic layer: the same metric or attribute definition should mean the same thing in every tool that consumes it. See the announcement post for context on why Honeydew joined.Documentation Index
Fetch the complete documentation index at: https://honeydew.ai/docs/llms.txt
Use this file to discover all available pages before exploring further.
OSI is an early specification under active development (current draft
0.2.0.dev0). Concepts, field names, and mappings may change before the
specification stabilizes.Concept Mapping
OSI and Honeydew model the same problem space, but the granularity differs. The table below maps each OSI concept to its Honeydew equivalent.| OSI Concept | Honeydew Equivalent |
|---|---|
semantic_model | A Honeydew workspace |
dataset | An Entity with its Source Data |
dataset.source | Entity Source Data |
dataset.primary_key, unique_keys | Entity granularity and unique keys |
relationship | A Relation between entities |
field (column reference) | A Source Attribute |
field (computed expression) | A Calculated Attribute |
field.dimension.is_time | A time attribute, often backed by a time spine |
metric | A Metric |
expression.dialects | Compiled to the target warehouse dialect (Snowflake, Databricks, BigQuery) |
ai_context | AI Metadata on the object |
custom_extensions | Honeydew-specific properties in the YAML schema |
Concepts Honeydew Adds Beyond OSI
OSI defines the portable metadata layer. Honeydew implements that layer and adds the operational machinery a semantic layer needs in production:- Query compiler and BI integration — metrics and attributes compile to warehouse SQL exposed through SQL, XMLA, GraphQL, and MCP interfaces, with native integrations on top of those interfaces into Power BI, Tableau, Excel, and other BI tools.
- AI Context Layer — OSI carries
ai_contextmetadata on individual objects; Honeydew adds a full context layer above the semantic layer (instructions, analytical skills, external business knowledge, and historical memory) and an agentic AI analysis engine that combines context retrieval with semantic compilation — deep analysis, agents, Slack, and Teams apps. - Domains — hierarchical projections of the same semantic model for different audiences (finance vs. marketing, regional sales, an AI agent’s scope). A domain selects entities and fields, applies mandatory filters, and overrides parameters; domains extend and compose other domains, so one model serves many perspectives without duplication.
- Aggregate awareness — semantic-aware query rewriting to pre-aggregated tables, alongside incremental aggregate updates and entity caching for further performance.
- Security and access control — row-level security and access control, enforced from the query interface down to the warehouse.
- Workflow, environments, and deployment — the semantic model is YAML and inherently git-native; Honeydew adds the workflow around it: workspaces, environments, and promotion paths for dev/staging/prod.
Converting To and From OSI
Honeydew provides a bidirectional converter between OSI and the Honeydew workspace format:- OSI → Honeydew — import an OSI semantic model and produce a Honeydew workspace (entities, relations, attributes, metrics, AI metadata). See Import Tools.
- Honeydew → OSI — export a Honeydew workspace as an OSI semantic model for consumption by other tools that support OSI.