db9 — Postgres but for agents

The database world just got a new player, and it's not targeting traditional applications—it's built for AI agents. db9 positions itself as "Postgres but for agents," offering a serverless PostgreSQL database with a twist: it's designed to work seamlessly with the way AI agents think, operate, and store information.

The Agent-First Philosophy

What makes db9 different from standard PostgreSQL? It's the unified approach to data and files. Traditional databases separate structured data (SQL tables) from unstructured data (files, documents, transcripts). db9 merges these worlds, allowing agents to store structured state in Postgres while keeping raw context, transcripts, and session snapshots as files—all in one workspace.

This dual-interface approach means agents can use SQL when they need power and file operations when they need simplicity. Need to query customer preferences? Use SQL. Want to store a conversation transcript? Drop it in the filesystem. Both live together, accessible through the same CLI.

Built for the Agent Workflow

Agents have unique needs that traditional databases don't address well. They need to remember things, retrieve context, run tasks, and maintain state across sessions. db9 tackles these challenges head-on:

Memory and Context: Agents can store personal preferences, session data, and task information in tables while keeping raw documents and transcripts as files. The example workspace shows an agent with pricing discussions stored as both structured data (topic: 'pricing') and raw markdown files.

Knowledge Retrieval: For research and documentation agents, db9 keeps source documents in the filesystem while storing chunks, metadata, and vectors in Postgres. This allows agents to retrieve grounded context from a single backend—no more juggling between a vector database and a traditional database.

Run History and Outputs: Automation agents can store reports and artifacts as files while keeping run history, status, and metadata in Postgres. Each agent run gets its own workspace, making it easy to track what happened and why.

AI Built In, Not Bolted On

Here's where db9 gets really interesting: AI capabilities aren't an afterthought—they're built into the database itself. The platform includes auto-embeddings, vector search, and HTTP extensions natively in SQL.

Want to generate embeddings inline? Just call embedding() in a query. Need similarity search? Use the <-> operator with built-in embeddings—no external pipeline, no API keys in application code. Need to call external APIs? The HTTP extension lets you make requests directly from SQL.

This approach eliminates the complexity of managing separate vector databases, embedding pipelines, and HTTP glue code. It's all there, ready to use.

Branching Everything

One of db9's standout features is environment branching. Not just tables—everything. One command creates an isolated copy of your entire environment: data, files, cron jobs, and user permissions. This means you can test against real conditions without affecting production, then delete the branch when you're done.

For teams working with agents, this is huge. You can spin up a staging environment that's a perfect copy of production, run your agent through its paces, and tear it down—all without complex infrastructure management.

The Practical Bits

Underneath the agent-focused features, db9 is still PostgreSQL. That means you get all the reliability, ACID transactions, and ecosystem support you expect. The platform includes built-in file storage (no S3 buckets to configure), distributed cron job scheduling that never sleeps, and full-text search.

Type generation is automatic—one command produces TypeScript or Python types from your schema. The CLI makes everything accessible, from creating databases to running queries to managing files.

Integration with the Agent Ecosystem

db9 plays well with the tools agents already use: Claude Code, OpenAI Codex, Cursor, Cline, VS Code, and others. It also offers plugins like my-claw-dash for deep insights and enterprise-grade observability, streaming events into db9 as immutable JSONL audit logs.

The Bottom Line

Is db9 revolutionary? Maybe not. But it's thoughtful. It recognizes that AI agents have different needs than traditional applications and builds a database that serves those needs directly. By unifying SQL and file operations, building AI capabilities in natively, and making branching trivial, db9 removes a lot of the friction agents face when working with data.

For developers building agent-powered applications, db9 offers a compelling alternative to the traditional "database plus vector store plus file storage" stack. Whether it becomes the go-to choice for agent development remains to be seen, but it's certainly addressing real pain points in the space.

One thing's clear: as AI agents become more prevalent, the tools they use will need to evolve. db9 is betting that the future of database design looks quite different from the past—and that future is agent-first.