At Altertable, we're building something simple but radical: data agents that work so you don't have to.
These agents don't just respond to queries. They monitor. They investigate. They remember. And when they find something that matters (an anomaly, a correlation, a segment worth testing, ...): they tell you. Otherwise, they stay quiet and keep watching. We're moving from passive dashboards to active dialogue.
This is analytics reimagined: not a dashboard waiting for a click, but an insight engine that thinks ahead.
An operating system for your data
Most agent systems depend on someone else's stack. We built our own, from ingestion to insight.
We run on a custom-built, in-house lakehouse architecture: engineered for fast, unified access across event data, product usage, and business metrics. Learn about our full technical stack. Our lakehouse isn't just storage; it's the foundation of a full-stack platform. From event collection to modeling to real-time analysis, we own the pipeline.
That's what enables our vision of a data operating system: a unified environment where agents, memory, computation, and interfaces all work together, seamlessly.
Tools matter. A lot.
Our agents aren't generating answers from thin air. They operate through our internal Model Context Protocol (MCP), a system that exposes core analytics primitives like:
- Schema introspection
- Semantic layer
- Event funnels & segmentation
- Time-based breakdowns
- Cohort definitions
- SQL access to raw data
- Chart rendering and dashboard manipulation
- and more...
Behind each agent action is a well-defined MCP call: transparent, auditable, and consistent. This is how we avoid hallucinations and deliver explainable, reproducible insights.
Memory: the key to continuous intelligence
Most analytics tools forget everything after a session ends. Ours don't.
Every Altertable agent operates with persistent memory: tracking past workflows, accepted/rejected hypotheses, user preferences, and prior anomalies. This shared memory is critical. It lets agents avoid repeating past failures, learn which patterns matter to your team, and adapt over time.
Without memory, AI agents are stateless scripts. With memory, they become intelligent systems that build on what they (and your team) already know.
What our agents actually do
Here's how a typical sequence unfolds:
- Coordinator agent receives a new task.
- Determines which agents to activate.
- Gathers memory and context from previous interactions.
- Injects company metadata, roadmap signals, recent changes.
- Preparer agent scopes and cleans the data.
- Joins events, filters sources, and ensures proper structure.
- Uses our internal MCP tools to create a reliable dataset for analysis.
- Investigator agent explores the data.
- Runs statistical tests, finds outliers, compares time ranges.
- Flags changes, anomalies, and hypotheses.
- Cross-checks memory to avoid redundant or rejected paths.
- Visualizer agent builds the story visually.
- Suggests or creates new visuals tied to the findings.
- Updates or improves dashboards accordingly.
- Synthesizer agent composes the output.
- Compiles agent contributions into a coherent narrative.
- Includes executive summaries, evidence, and historical context from prior runs.
Many of these agents run silently, in the background. They don't interrupt your day—they enhance it. If nothing notable happens, they log and wait. If something changes, they'll be the first to know, and the first to tell you!
Why it matters
The modern data stack is too manual. It's reactive, fragmented, and expensive. Teams are left babysitting pipelines and digging through dashboards, just to answer basic questions.
We're replacing that with a full-stack platform that thinks ahead. Our agents analyze, remember, and act because they're built on top of a data operating system designed for proactive, always-on insight.
If your stack feels more like a patchwork than a platform, let's talk.
