Sundial vs. Sigma Computing
Everyone is shipping an analytics agent. Almost nobody is making sure it's set up to win.
We're in a magical age where agents can do analysis, humans can build the loops to power it, and no one spends hours writing SQL or Python. Business users fire queries and get back answers. But do they trust those answers? And if they don't, how are the tools helping them get better?
Honestly, the "no one has to write code" narrative is getting old. The bar needs to move up — to "no one needs to double-check whether what the agent did was the right way to do it." Sounds tough? It is. And that's exactly the craft Sundial has cracked.
Everyone is building the AI-on-BI. Everyone is building a context layer. But who is setting it up?
That was our top takeaway from the Snowflake Conference. In the age of AI, building a barebones analysis agent is cheap; making it usable is expensive. That's where Sundial resonated with the people we met — the trustworthy analyst that runs on encoded analytical frameworks and has context set up from day one.
Pick Sigma if you want an exploratory tool that can help answer your questions.
Pick Sundial if you want the answers.
How do you choose?
Both run on top of your warehouse. Both can be used by business users, not just data scientists. The difference is what each one thinks its job is: to deliver an answer, or to deliver the thing that can potentially deliver the answer.
Every analytics tool now has an AI agent sitting on a warehouse with a context layer underneath to hold the semantics. In a demo, they all look the same. So the question that actually matters isn't "do you have an agent?" It's who made sure that agent is set up to succeed? Agents are a dime a dozen. The skills and the setup around them are what separate a useful analyst from a confident-but-wrong one — and that setup takes specialised work.
Most tools quietly hand that specialised work back to you.
What is Sigma Computing?
Sigma (formally Sigma Computing) is a warehouse-native platform for building dashboards, AI apps, and agents yourself — now positioned as "the AI runtime for business." It's a strong product: spreadsheet UX, SQL, Python, write-back, and a surface for building on governed data. If you know exactly what you want and have the people to build and maintain it, Sigma gives you the room to do it.
Where it's strong, it's the breadth of build-it-yourself tools and warehouse-native scale. The catch is that its AI is a general-purpose model wired into a spreadsheet — LLM functions in cells, plus passthrough to warehouse models like Cortex. You get all the bells and whistles to do your job, but the analytical judgement around it is your team's to build and keep current.
What is Sundial?
Sundial is an opinionated AI analyst that does real analysis on your warehouse data, and gives your data team the accuracy and observability to trust every answer. We've been insights-first for five years, pre-dating the LLM era, and have spent that time encoding analytical frameworks so anyone can analyse the way an expert analyst would.
It doesn't need you to set it up perfectly. It ships with a point of view on which metrics matter, how they connect, and what good analysis looks like — useful on day one, not day ninety. A team of former data leaders plus a set of agents do the setup with you in the first 30 days; it's not a chatbot to figure out alone. Sundial has dashboards and data apps too, but they're opinionated, not a blank canvas. It can fire virtually any query against your warehouse; the semantic layer, Playbooks, and trust system vet the work, so you always start from a vetted layer.
Feature comparison
Sundial is built for trust and speed-to-insight; Sigma for build-it-yourself breadth. Here's where each leads, capability by capability.
| Capability | Sundial | Sigma Computing |
|---|---|---|
| Setup & time-to-value | ||
| Useful on day one, set up for you~30 days, by experts + agents | Available | Not offeredyou build & own it |
| Pre-built analytical frameworksRCA, retention, anomalies | Availableexpert Playbooks | Not offeredbuild your own |
| The intelligence | ||
| Has a point of view on every questionPlaybooks + opinionated defaults | Available | Not offeredopen canvas |
| Proactive insightssurfaces what you didn't think to ask | Available | Available with caveatsrule/threshold alerts you configure |
| Context Engineorg, connectors, semantics | Availableowned & curated | Available with caveatsvia data models / dbt |
| Real analysis, not just metric lookupswhy did retention drop? | Available | Available with caveatsif you guide the agent |
| Trust & rigor | ||
| Confidence signal on each answerguaranteed vs. directional | Available | Not offered |
| Observability + evals so rigor holds | Available | Available with caveatsaudit/lineage only |
| Self-serve & build surface | ||
| Dashboards + data apps on your warehouse | Availableopinionated, ready-made | Availableblank-canvas DIY |
| Spreadsheet UX + Python notebooks | Not offered | Available400+ functions |
| Foundation | ||
| Meet you where you workSlack, Teams, MCP | Availablenative chat | Available with caveatsoutbound alerts + MCP |
| Self-hostable / private cloud · BYO LLM | Available | Not offered |
Anyone can put an agent on a warehouse. The work is making sure it's set up to win. Teams like OpenAI and Gamma already run their analysis this way — see Sundial answer a real question on real-looking data, or try Sundial for free.