| name | understand-platform |
| compatibility | Requires Altertable MCP server |
| description | Explains Altertable platform concepts and architecture. Use when asking what Altertable is or how agents, discoveries, memories, insights, and dashboards fit together. |
| metadata | {"author":"altertable-ai","requires":"altertable-mcp"} |
Understanding Platform
Quick Start
Use this framing when someone asks what Altertable is:
- Altertable is the data runtime for the AI era: one governed lakehouse for humans, applications, analytics, and agents.
- Its foundation combines DuckDB compute, Parquet storage, managed catalogs, external catalogs, and semantic models.
- Agents and MCP clients use the same governed context to inspect schemas, run SQL, build insights, monitor dashboards, and use knowledge.
- The platform's core operating loop is:
Data -> Insights/Dashboards -> Tasks/Agents -> Notifications/Discoveries -> Review/Feedback -> Memories/Knowledge -> Better Future Analysis.
When to Use This Skill
- User asks "what is Altertable?" or "how does the platform work?"
- User wants the difference between agents, discoveries, and memories
- User asks how insights and dashboards connect to monitoring
- User asks how Altertable differs from traditional warehouse-first stacks
- User needs a conceptual architecture explanation before implementation details
Core Platform Narrative
Most data stacks were optimized for batch pipelines, dashboards, and occasional human queries. Altertable is optimized for always-on analysis where agents continuously consume data.
Use these points in order:
- Foundation: modern lakehouse architecture with DuckDB workers over Parquet and object storage.
- Access: managed catalogs, external catalogs, HTTP APIs, SQL adapters, MCP tools, and Ask Agent all query the same data layer.
- Context layer: semantic models and knowledge entries give agents shared business definitions.
- Operational output: insights, dashboards, tasks, notifications, and discoveries turn analysis into ongoing workflows.
- Learning loop: memories retain feedback and improve future agent behavior.
Concept Map
Agents
Autonomous data collaborators that execute both repetitive and higher-level analytics work.
- Synchronize sources and maintain data readiness
- Inspect catalogs, run SQL, and use semantic models
- Build or update queries and visual outputs
- Monitor insights and dashboards continuously
- Generate notifications or discoveries when something noteworthy happens
- Learn from feedback through memories
Discoveries
Reviewable findings generated by agents.
- Include context, rationale, and suggested actions
- Require human approval or rejection
- Can represent anomalies, trend changes, segment shifts, schema/model changes, and event readiness
- Become a primary collaboration interface between agents and teams
Memories
Persistent knowledge accumulated by agents across runs.
- Episodic: what happened
- Semantic: what it means
- Procedural: how to handle it next time
- Reinforced or weakened by discovery review outcomes and repeated use
Knowledge Entries
Durable reference material that the team writes down or syncs from repositories.
- Metric definitions, operating procedures, product terminology, and source caveats
- Used by Ask Agent, MCP clients, and tasks while answering questions
- Complements memories, which are learned from agent runs and feedback
Insights
Persistent analyses and visualizations over lakehouse data.
- Funnel, segmentation, semantic, and SQL insights cover different analysis needs
- Serve as reusable analytical building blocks
- Can be monitored directly by agents
Dashboards
Collections of insights organized for KPI tracking and shared monitoring.
- Aggregate related metrics and context in one place
- Support shared variables for coordinated filtering
- Can have attached agents that watch for anomalies and trend shifts
How the Concepts Work Together
Data ingestion -> Catalogs + lakehouse storage/query engine
-> Semantic models + knowledge
-> Insights and Dashboards
-> Tasks and Agent Monitoring
-> Notifications and Discoveries
-> Human Review and feedback
-> Memories and knowledge updated
-> Better future monitoring and analysis
Communication Guidelines
When explaining the platform:
- Start with outcomes (continuous analysis, faster decisions, lower marginal cost at scale)
- Then map to concepts (agents, discoveries, memories, insights, dashboards)
- Emphasize human-in-the-loop review for quality and trust
- Distinguish "analysis artifacts" (insights/dashboards) from "agent outputs" (discoveries)
- Describe memories as adaptive context, not static storage
Common Pitfalls
- Presenting Altertable as only a BI/dashboard tool
- Describing agents as one-shot assistants instead of continuous collaborators
- Skipping the human review stage in the discovery lifecycle
- Treating discoveries as equivalent to insights (they are not)
- Omitting memory feedback loops when explaining how agent quality improves over time
- Leading with implementation internals before clarifying conceptual flow
Reference Files
