db9 (db9.ai) is a new serverless PostgreSQL platform built for AI agent workloads, combining a full relational database with a cloud-native filesystem in a single workspace. Its most distinctive feature ships before an agent writes a single query: db9 publishes a machine-readable instruction file at https://db9.ai/skill.md that any agent runtime can fetch and follow to autonomously install, authenticate, and operate the platform — making the database self-describing to agents without a human in the loop.
The core premise is that agents operate across two fundamentally different data types: structured state — session metadata, task records, embeddings, run history — that benefits from SQL's relational power, and unstructured raw context — transcripts, session snapshots, reports, artifacts — better handled as files. Rather than stitching together a vector database, an embedding pipeline, blob storage, and a job scheduler, db9 collapses all of these into one PostgreSQL-compatible backend accessible via both SQL and file operations. Built-in capabilities include auto-embeddings callable in SQL via an embedding() function, vector similarity search, outbound HTTP calls from SQL via an HTTP extension, filesystem queries via the proprietary fs9 extension, distributed cron scheduling via pg_cron, and full-text search — all requiring zero external configuration or API key management in application code.
Environment branching rounds out the core feature set: a single command clones an entire agent environment including tables, rows, files, cron jobs, and user permissions, enabling isolated testing against production-like conditions. An optional my-claw-dash plugin extends this into the observability layer, integrating with the <a href="/news/2026-03-14-nanoclaw-docker-partnership-six-weeks-after-open-source-launch">OpenClaw agent gateway</a> to stream runtime events into db9 as both structured audit tables and append-only JSONL log files — creating an immutable audit trail for multi-agent systems. The platform lists integrations with Claude Code, OpenAI Codex, Cursor, Cline, VS Code, OpenCode, OpenClaw, and Vercel.
db9 enters a crowded serverless Postgres market where Neon and Supabase are the leading incumbents. Neon, which raised a $46 million Series B in 2023, shares the branching metaphor but its model is developer-centric copy-on-write snapshots optimized for CI/CD workflows rather than agent environment cloning, and it bundles no co-located filesystem, no HTTP-from-SQL capability, and no built-in embedding pipeline. Supabase, valued at over $2 billion after its 2024 Series C, offers a broader platform with pgvector support, but its storage is a separate S3-compatible layer rather than a SQL-queryable filesystem, and its design philosophy remains Backend-as-a-Service for human developers. Purpose-built vector databases — Pinecone, Weaviate, Qdrant, Chroma, and Milvus — cover similarity search but lack relational semantics, transactional guarantees, scheduling, filesystem layers, or HTTP egress. That gap is what db9 is targeting.
The competitive risk is real: Neon and Supabase have substantially larger ecosystems, enterprise distribution, and funding, and could add agent-native features as bolt-ons. db9's window depends on how quickly it can build developer and agent runtime adoption before better-resourced competitors close the feature gap. The skill.md self-onboarding pattern has no direct equivalent among incumbents — that, combined with the unified SQL-plus-filesystem architecture, is db9's clearest bet on where agent infrastructure is heading. Whether it's enough of a head start is the open question.