Polynya: Turn your Postgres into workspaces for AI

The increasing complexity of AI agents demands reliable, isolated data contexts. Traditional data warehousing solutions often fail this test because they require significant infrastructure overhead (e.g., managing Kafka pipelines, maintaining CDC connectors like Debezium, or orchestrating complex Airflow workflows). Polynya targets this precise pain point, presenting itself as a secure, high-fidelity analytical proxy between operational PostgreSQL databases and AI workloads. At its technical core, Polynya streams data from Postgres into the Iceberg format. This isn't just a simple dump; it establishes an active data lineage, allowing AI agents to query the data through a disposable, ephemeral ClickHouse instance. This architectural choice is critical: the agent's queries never hit production, mitigating risks associated with running complex, ad-hoc AI queries against live operational tables. Furthermore, the real-time sync capability—every 30 seconds, guaranteed across all tiers—is a significant differentiator, offering 'always fresh' data without the need for complex pipeline management that plagues competing CDC solutions. From a workflow perspective, the ability to create and persist 'workspaces' is the system's key innovation. It moves the paradigm beyond simple reporting. When an agent queries the data, it doesn't just get a single number; it can 'Save this as a view.' This capability allows the AI to build an analytical layer—a view defining 'revenue by payment method' or 'inactive signups'—which then persists within the workspace. This compound effect means the analytical context improves with every interaction, turning raw data into a continually refining, agent-specific data model. While the initial focus is on ease of use (demonstrated by the simple `npx polynya create --all` CLI command), the architecture is designed for scalability and decoupling. By adhering to the standard Apache Iceberg format, Polynya guarantees zero vendor lock-in, allowing users to graduate to various query engines when their needs outgrow the included ClickHouse. In essence, Polynya provides an entire data lifecycle wrapper: ingestion (Postgres to Iceberg), access (ephemeral ClickHouse), and context building (persistent workspaces/views), all wrapped in a developer-friendly, low-friction layer.