Constructive has unveiled agentic-db, an innovative open-source Postgres database designed to enhance AI agents with capabilities that include persistent memory, structured knowledge, and hybrid retrieval. This release is a significant milestone for the company, which has amassed over 100 million open-source downloads, and it reflects a shift in how agent frameworks manage memory.
Transforming Memory Management for AI Agents
The traditional method of storing agent knowledge in markdown files has been criticized for its inefficiency. As noted by AI researcher Andrej Karpathy, relying on markdown is akin to navigating a library without a catalog—time-consuming and cumbersome. In contrast, agentic-db streamlines this process by allowing databases to search, filter, and rank information at the query level, eliminating context window bloat and token burn. This means that agents can retrieve precisely what they need without unnecessary overhead.
Kristopher Floyd, Managing Partner at Frontier Syndicate, collaborated with Constructive to refine agentic-db based on his experiences with inefficient CRM systems. “My agent’s only option was to dump everything it knew into the prompt and hope for the best,” he explained. With agentic-db, agents can utilize vector search, full-text search, and auto-embeddings, making it possible to efficiently search their knowledge base rather than relying on brute-force methods.
Dan Lynch, Founder and CEO of Constructive, emphasized the importance of a robust database for agent-first infrastructure. “Agents need to remember, search, and reason over structured knowledge—not re-read a pile of markdown files every time they wake up,” he stated. By making Postgres the foundation of this infrastructure and open-sourcing it, Constructive aims to equip developers with a powerful tool for building intelligent agents.
Features of agentic-db
agentic-db is delivered as a single installable Postgres schema, deployable in one command via pgpm, the PostgreSQL Package Manager. This schema encompasses a comprehensive set of features essential for agent functionality:
- Long-term memory: Episodic memories with vector, BM25, and spatial search capabilities.
- Conversations and tool tracking: A replayable event log of all agent interactions.
- Skills, tools, and prompts: A versioned registry for semantic intent matching.
- Rules and behavioral policies: Declarative trigger/action pairs for governance.
- Task orchestration: A priority queue with tracking for assignments and results.
- Runtime observability: Structured logs and metrics for asynchronous workflows.
- World model: A comprehensive personal CRM and life-OS.
Every text-bearing table in agentic-db is automatically indexed for five retrieval modes, including semantic vector search and fuzzy matching, ensuring efficient data access. An auto-embedding pipeline keeps embeddings current for optimal semantic search performance.
Agent Skills and CLI Integration
Constructive’s platform generates Agent Skills and CLIs directly from the database schema, ensuring that every database produced is agent-first. These skills are structured instruction files that facilitate seamless integration into an agent’s workspace, compatible with a variety of AI assistants. Additionally, a comprehensive CLI supports CRUD operations, search, and administrative commands, all derived from the same schema.
Michael Shawn Lockwood, a seasoned database architect, expressed enthusiasm for agentic-db’s potential. “What looks like ‘just a database’ is actually a full foundation layer for agentic systems,” he remarked, highlighting its capabilities in long-term memory and task orchestration. He anticipates that agentic-db will become a foundational layer for real-world systems in the coming years.
agentic-db is currently available under the MIT license for local use, providing a robust solution for individual developers. For teams requiring secure Postgres solutions at scale, Constructive is developing a cloud offering that will extend the same schema with enhanced security and provisioning features.
Developers can access agentic-db on npm and GitHub, with further details available at pgpm.io.
