Zoho Put a Semantic Layer in the Box. Apache Ossie Is the Harder Part.
The ontology and semantic layer ships in preview, not general availability. And declaring compatibility with Ossie is not the same as round-tripping a model through it.
Zoho released a major rebuild of Zoho Analytics on 28 October, adding two context layers above the data: an ontology that maps the business’s entities and relationships, and a semantic layer that defines what its metrics mean. Ask Zia, the natural-language assistant Zoho has sold since 2018, is now a multi-agent system orchestrating six specialised sub-agents. There is a Model Context Protocol server with remote and local options, a command-line interface, a Claude plug-in, and headless business intelligence so agents outside Zoho can query the platform. Pricing is unchanged.
Two claims in the announcement need separating from the rest, because they carry the most weight and the least support. The first is that the semantic and ontology layers are available. The second is that the platform supports Apache Ossie.
What Zoho actually shipped
The architecture is the substance, and it is worth setting out before turning to the claims. The ontology layer maps organisational structure and the relationships between business entities. The semantic layer sits above it as a glossary: synonyms, descriptions, metric calculations, sample questions and the SQL that answers them. Both can be built by hand or with an AI assistant. Zoho’s argument is that this context is what an agent needs to answer a business question correctly rather than plausibly, and the argument is sound.
Around that sit the agentic pieces. Ask Zia orchestrates sub-agents for data analysis, modelling, visualisation, data preparation, coding and SQL generation. A custom agent builder lets enterprises ground their own agents on a chosen dataset or semantic model, with knowledge-base files and MCP connections. A workflow builder lets those agents trigger actions. Thirty new connectors take the total past 530, adding SAP, PayPal, Adobe Commerce, Twilio, Notion, Intercom, Supabase, Amazon DocumentDB and AlloyDB alongside five more Zoho products.
What Apache Ossie requires
The standard Zoho names has a longer history than the release suggests. It began as the Open Semantic Interchange in September 2025, launched by Snowflake, Salesforce, dbt Labs and Google among others, with a version 0.1 specification published in January 2026 under Apache 2.0. In July 2026 it entered the Apache Incubator and was renamed Ossie to avoid confusion with the ISO/OSI networking model. The coalition has grown from 17 launch partners to more than 50 organisations, with contributions from Snowflake, Salesforce, Databricks, dbt Labs, AtScale, RelationalAI, ThoughtSpot, GoodData and Dremio, and it runs in production at Snowflake, Salesforce, Dremio and dbt Labs.
What Ossie defines is a YAML format for exchanging semantic metadata across two layers: semantic models covering datasets, fields, relationships, metrics and AI context, and ontologies covering concepts, relationships and rules. Reference converters exist for dbt, GoodData, Snowflake, Databricks, Omni, Sigma, Apache Polaris and, in review, Apache Spark. The specification sits at a 0.2.0 draft and is not yet finalised.
Compatibility is not a runtime
This is where the claim needs care. Zoho’s material describes compatibility with the specification, which is a reasonable and useful position: definitions authored in Zoho can in principle be expressed in Ossie and read elsewhere. It is not the same as executing Ossie models natively or guaranteeing that a model survives a round trip.
Independent analysis of the Ossie ecosystem makes the distinction sharply. A reference converter in the project repository proves a conversion path exists; it does not prove that every measure, security rule, time grain or vendor extension survives the translation. The practical advice from that analysis is to test one representative metric end to end, including a join, a filter and a time dimension, before calling anything portable. Joining an interoperability initiative is useful context and is not by itself an import or export workflow.
Zoho has an advantage here it does not claim loudly enough. Because the semantic layer is native rather than bolted on, an organisation that adopts Zoho first and exports later avoids the usual trap of rebuilding definitions for each tool. Whether the export holds is the thing to test.
The preview problem
The second claim needs a date attached. Zoho says the latest version is available now, and cloud updates are indeed live. But reporting from the launch records that preview versions of pixel-perfect reporting, the ontology layer and the semantic model roll out globally over the next four to eight weeks, and that the two headline capabilities are not fully in production for every customer today. On-premises customers wait until the first quarter of 2027.
That is a normal cadence for a platform of this size and it is worth stating plainly, because the two features that justify the entire announcement are the two still arriving. An organisation planning a semantic model should plan against a preview rather than a general release, and should ask for the date in writing.
The governance section deserves its own reading, because it answers the objection that arrives with any agentic feature. The data catalogue gathers data, metrics, analytics assets and models in one place with lineage, impact analysis and access policy management. AI observability records which agents ran, who used them, what they accessed and how they answered. Administrators get usage and cost tracking per team, and Ask Zia offers a choice between Zoho’s own model and a customer’s frontier model, which matters where data residency is constrained. Sandbox and deployment management lets teams build and test in isolation, publish to production and roll back. That set is more complete than most agentic releases offer.
The question an analyst asked
The most useful critique on record comes from Jason Leone, quoted by TechTarget, who said what Zoho still has to show is how much work it takes for a customer to model their business in the ontology layer, because the agents’ answers are only as good as that model. That is the right question and it has no answer in the release.
An ontology is not a configuration file, and no generator produces one. Someone has to decide what a customer is, which system owns that definition, how two records in different systems refer to the same entity, and which rules constrain the relationship. Zoho offers an AI assistant to help build it, which is helpful and does not remove the judgement. For an enterprise with a data governance function this is a project. For the small businesses Constellation Research’s Michael Ni points to, who have neither a data warehouse nor a data team, it may be the first time anyone has asked them to define their business formally.
Why the in-the-box argument works
Ni’s assessment is that the update is significant mostly because the business context comes in the box. That is the commercially important point. The competing approach requires a warehouse with its own semantic layer, a separate modelling tool, and a governance process to keep them aligned. Zoho offers one product that holds the data, the definitions and the agents, on-premises if required, at a price that has not changed.
The release makes a related claim worth noting: the native layers eliminate the need for a separate cloud data warehouse. That holds for organisations without one, and less well for those with, since Zoho also supports a zero-copy mode that reads metadata from Snowflake or Databricks without moving the data. Both can be true depending on which customer you are.
Zoho’s deployment story is unusual in this category and worth stating. Zoho Analytics runs on Zoho Cloud, on-premises on a local server or in Docker, and on public cloud infrastructure across AWS, Azure and Google Cloud. Very few analytics platforms of this class offer an on-premises option at all, and for regulated industries in India, where data localisation expectations are explicit, that is a real differentiator. The cost is that on-premises customers wait until the first quarter of 2027 for the layers this release is about.
What to watch
Three markers. The first is whether the ontology and semantic layers reach general availability on the stated schedule, because a preview that slips is a different product. The second is whether Zoho publishes an Ossie round-trip test with a real model, since that would convert a compatibility claim into evidence. The third is the Zoho Data Platform, targeted for 2027, which the company describes as a common layer for structured and unstructured data beneath its applications. That would put Zoho where Snowflake, Databricks and Microsoft Fabric sit, and it is a far larger ambition than a business intelligence upgrade.
What it signals
The release marks a stage where business intelligence vendors stop selling charts and start selling the definitions the charts rest on. That shift makes the semantic layer the product and the dashboard the output, and it makes portability the question every buyer will eventually ask. That combination, native context, an on-premises option, unchanged pricing and an open export format, is a coherent pitch to a segment the large data platforms serve poorly: mid-sized organisations wanting governed analytics and agents without building a data platform first. Zoho’s answer, that the definitions live in an open format, is the right one. The work remaining is to demonstrate it rather than declare it, and the standard it chose gives it a way to prove the claim to anyone who asks, which is more than most vendors in this position can offer.