The bigger picture

Where pgvis is going — openly, one milestone at a time

pgvis is an I/O-free Rust engine that turns Postgres (and SQLite) into MCP, REST, and OpenAPI. The roadmap below is concrete and code-driven. Past 1.0, we want to go further — GPU-accelerated vector search and a managed, Firebase-style pgvis DBaaS — all of it built in the open, under the same Apache-2.0 license as the engine.

Roadmap

Milestones to 1.0

pgvis is in beta — no git tags or release notes have been cut yet, so what's "shipped" below means "lives on `main` and is exercised by the test suite." Each row is a concrete piece of work, not a marketing line.

  1. Today Shipped

    REST, OpenAPI, MCP on Postgres & SQLite

    Working from main: REST + OpenAPI on both Postgres and SQLite, MCP tools generated from the same introspected schema, the I/O-free engine, and the pgvis-lib Builder facade.

  2. Hardening In progress

    MCP stdio polish

    Tighten the MCP stdio surface: error handling, tool descriptions, schema discovery. Works with any MCP client.

  3. Next Planned

    MCP over SSE

    Server-Sent Events transport for MCP, so agents can stream alongside the existing stdio and Streamable-HTTP options.

  4. 1.0 Planned

    Stable embed API & first tagged release

    Lock the pgvis-lib Builder and core surface so downstream Rust apps can depend on it without churn. First git tag and published release notes will land here.

Source: commits on `main`

Beyond 1.0

The longer horizon

What we want to build on top of pgvis once the engine is stable. These are forward-looking and explicitly not part of the v0.x roadmap above — but they will ship in the open, just like the engine.

On the horizon

GPU-accelerated pgvector

AI agents need low-latency embedding search next to the relational and MCP surfaces pgvis already provides. We're exploring GPU kernels on top of Postgres' pgvector so similarity queries scale alongside the rest of your schema — not in a separate vector database.

Early R&D · not yet on the OSS roadmap
In early planning

Managed pgvis — a Firebase-style DBaaS

A hosted product where the same OSS engine you run locally backs the managed offering. Bring-your-own Postgres, or run on managed infrastructure with MCP, REST, and OpenAPI surfaces switched on at the click of a button. Same engine, same code, just deployed and operated for you.

Planning stage · design discussions happen on GitHub

All of this is being built openly. The engine, the GPU vector layer, and the managed DBaaS all stay Apache-2.0 — no rug-pulls, no source-available licenses, no commercial fork. If you want to shape the direction, the design discussions live on GitHub.

Join the discussion on GitHub

Want to shape what 1.0 looks like?

Issues, design feedback, and PRs all land on GitHub. The early roadmap moves fastest when the people using pgvis are the ones telling us where it's tight.