TigerFS Mounts PostgreSQL Databases as a Filesystem for Developers and AI Agents

Host A: Welcome to DevTools Radio, I'm your host, and today we're talking about something that genuinely made me do a double-take — a new open source project called TigerFS that lets you mount a PostgreSQL database as a filesystem.

Host B: Okay, so like... you're literally just browsing your database the way you'd browse a folder on your desktop?

Host A: Exactly. You can run ls, cat, grep, find — all your standard Unix tools — directly against a live PostgreSQL database, no SQL required, no API calls, nothing.

Host B: That's wild. So who built this and what's the actual use case they had in mind?

Host A: It comes from Michael Freedman, the CTO and co-founder of TigerData, and his pitch is pretty compelling — he built it primarily for AI agent workflows. His argument is that agents don't need fancy APIs, they already understand filesystems intuitively.

Host B: That actually makes a lot of sense when you think about it. Tools like Claude Code or Cursor are already navigating file trees constantly, so giving them a filesystem that secretly has full ACID transactional guarantees underneath is kind of genius.

Host A: Right, and there are two main ways to use it. File-first, where you organise things like Markdown docs in directories with automatic versioning and atomic writes. And data-first, where you mount an existing PostgreSQL database and just start poking around its data.

Host B: The data-first one is really interesting to me — so you can filter and sort data just through the filesystem path itself, and that translates into actual database queries behind the scenes?

Host A: That's the idea. You're essentially querying without writing a single line of SQL. Each file maps to a real PostgreSQL row, and multiple agents or humans can read and write concurrently without stepping on each other.

Host B: Okay but I have to ask the obvious question — performance. Because this sounds great for small workloads, but what happens when you've got a massive dataset?

Host A: You're not alone in wondering that. Over on Hacker News, that's basically the top concern. One user pointed out it'll probably shine for config data, context files, smaller datasets — things that fit comfortably in memory — but might struggle beyond that. It is described as experimental, to be fair.

Host B: And it's worth noting there's some historical precedent here too, right? Someone actually compared it to Oracle's old Internet Filesystem from the early 2000s.

Host A: Yeah, Franck Pachot from MongoDB brought that up — he mentioned Oracle got so excited about mounting databases as filesystems back in the Y2K era that they literally rebranded with a lowercase "i" in Oracle 8i. So this idea has been floating around for a while, TigerFS is just bringing it into the age of AI agents.

Host B: Practical setup question — how do you actually get this running?

Host A: On Linux it mounts via FUSE, on macOS via NFS, and it connects to any existing PostgreSQL instance or a managed cloud service. MIT licensed, no external dependencies. So the barrier to trying it is pretty low.

Host B: Honestly I think the coolest part is how it reframes the filesystem as the API itself — no abstraction layer, no SDK to learn, just the tools developers already know.

Host A: Freedman literally put it that way: "The filesystem is the API." And he said he's curious what else people might use it for beyond agent workflows, which suggests there's a lot of unexplored territory here.

Host B: I'd love to see this used for developer tooling around migrations or data exploration — just grep-ing through your production schema feels both terrifying and incredibly convenient.

Host A: Ha, that's the DevTools spirit — equal parts exciting and slightly dangerous. Alright, TigerFS is worth keeping an eye on if you're building agent workflows or just love creative PostgreSQL experiments. Links in the show notes as always.

Host B: Thanks for listening to DevTools Radio, everyone. Stay curious, stay building — we'll catch you next time.