Credible vs. Omni

Omni serves agents well, and it does so as a product: chat, dashboards and the modeling IDE are the center of gravity, and the semantic layer is what makes them trustworthy — which is also what makes the analyst experience as good as it is. Credible is arranged the other way round. The model is the product, and Credible Workspaces, notebooks, reports, dashboards, data apps, agents and applications are all consumers of it. You get a BI surface either way; the difference is whether the model exists to serve the product or the surfaces exist to serve the model.

Omni’s agent is Omni’s: a coordinator that picks the next tool, adapts mid-flight, retries when something breaks and stops when it has something worth showing — tuned by the people who built the product, which is why it behaves well from the first question. We did not build an agent, because you already own one. Credible plugs into the harness your team already runs — Claude, ChatGPT, Gemini, the coding agent in your editor, the agents you write yourself — and hands it what it needs to get better at your data: skills and MCP tools that are open source and yours to change, and telemetry from every question and answer, so improving how your agent works is something you do rather than something you request.

Omni’s modeling format is its own, in the LookML tradition, and the team behind it knows that tradition better than almost anyone. Its Agent Skills are the same shape of idea as ours — repeatable instructions an agent follows — and anyone with chat access can write one, which is a genuinely good way to spread expertise inside a company. Ours are open source in Malloy Publisher, alongside the retrieval tools they call, so they run in whatever agent your team already uses with no Credible account, and you can read what a skill tells the model to do before you trust it. Malloy is an open-source language, Publisher is the open-source server we build and maintain, and Credible is in the Apache Ossie ecosystem. It is a slower way to build a product and we think it is the right one, because a model written in the open outlives whoever is serving it.

Omni’s modeling agent suggests metrics and join paths to a person working in its UI, and its workbook-first order — model as you analyze, promote what proves useful to the shared model — means the model grows out of real questions rather than a design phase. Credible’s coding agent builds the model itself with open MCP tools and skills, drawing on context already in your stack: warehouse metadata, existing SQL and dbt projects, docs, dashboards. Because the output is Malloy in your repository, it arrives as a diff you review like any other code, and you can run that agent in the editor your team already uses or in Credible Workspaces, which packages the same skills in a familiar app. Same goal, different assumption about who is at the keyboard.

Omni’s Dashboard Builder plans the queries, picks the charts and lays out a dashboard using the logic and permissions already in the semantic model, saved as Omni content and edited where it was made. Credible’s agent writes the artifact as source: HTML and queries against the governed model, served from a Malloy package, which makes a dashboard or a data app something you diff, review, test and ship with the rest of your software. Both are agent-made; the difference is whether what comes out is content inside a product or code in your repository.

Omni is more than an IDE with history: model YAML syncs to your git provider, branches are required before changes, and you can insist every change go through a pull request. The direction is worth being precise about — the CLI fetches the YAML down from Omni, your saves push back to an Omni branch, and git receives the result on merge, so the working copy is Omni’s and git holds the record. Credible starts from the other end: Malloy text is in your repository and Publisher serves it from there, so a change is an ordinary pull request in the repo that is already the source of truth. Both give a team a real review step; the difference is which side owns the file.

Governance as annotations in the model, globally distributed high-availability serving, and materialization you declare with one annotation and we hold hot in memory, rather than a cache you operate. Omni describes query-result caching with schedule-based refresh and aggregate awareness over your warehouse, and leaves the heavier end to the warehouse or to dbt, which keeps its own product focused. We took it on because serving context to production agent traffic is the job we picked — and because it means the warehouse does not have to be the thing that makes a query fast. It also means a model can start before the warehouse does: Omni’s connectors are warehouses, and Credible brings its own query engine and storage when there isn’t one yet.

Omni inherited more from Looker than the modeling tradition — the way the product is sold and packaged came along too: a vendor-owned format, a multi-tenant cloud with no self-hosted option, and a sales-led platform fee with role-based seats. That is the standard shape for BI, and it prices what a person does at a desk, which is what BI was built for. It describes the workload less well once the consumer is an agent, because agents hold no seats and their traffic is machine-paced rather than per-person. Credible is metered on the other axis: usage per organization, unlimited users on every plan, MCP access charged nothing, and an open-source tier you run yourself on your own compute — so adding people or agents changes what you run rather than what you are licensed for.

Omni ships a LookML converter that maps Explores to its Topics and a modeling agent that drafts fields and joins, and its own migration guide is honest that the semantic work — descriptions, synonyms, business definitions — and the dashboards remain human work. Our adapter is the same shape of tool with one difference: `malloy-lookml-review` is MIT-licensed in Malloy Publisher, so the rules it applies to your project are a markdown file you can read before you run it, and change if you disagree.