About
We're building the layer
every team will need.
Plynf started from a single observation: every team running agents in production hits the same wall, and most of them try to build their own substrate before realizing how much non-differentiating plumbing that involves.
The bet
In two years, every team running production agents will have some form of this layer. We're building the one they should buy.
Three forces converge in the next eighteen months. Every serious team running agents in production hits the quadratic-cost wall, the resilience wall, the integration wall. The standard for tool exposure (MCP) has stabilized, meaning an independent platform can sit cleanly above the protocol layer rather than trying to define one. And as model prices fall while agents take on bigger jobs, total token spend per workload climbs, making efficiency gains the difference between unit economics that work and ones that don't.
The strategic question is not whether teams need this layer. They will. The question is who provides it. Three plausible answers: each team builds it themselves (expensive, half-finished, non-differentiating), a hyperscaler ships it tied to their own model (vendor-locked, slow to evolve), or an independent platform wins (model-agnostic, focused, the obvious choice).
Plynf is built for the third outcome.
Principles
What we won't compromise on.
Build for agents, not around them
Every API in Plynf is shaped by the question: would an agent use this? Not 'is this fine for a human user?' That single rule produces dramatically different ergonomics.
Measure everything we claim
71% token reduction is a measurement, not a marketing line. Run the benchmark yourself from a clean clone. It's in the repo. Every other claim on this site is the same way.
Backwards compatibility is non-negotiable
Every demo from v0.1 still runs unchanged. API v1 is locked: additive-only for at least 12 months. Breaking changes get a /v2/ namespace and a migration guide.
Open source as a fact, not a marketing tool
Apache 2.0. The cloud product runs identical code. No proprietary forks, no 'community edition' with the good parts ripped out.
By the numbers
Status, today.
- Version
- v1.5
- Tests
- 2,867
- SDKs
- 5
- License
- Apache 2.0
Ready when you are
Build the agent.
We'll handle the substrate.
Subscribe, paste your key into your agent, done — nothing to install. Or self-host with one command. Either way: production-ready when you are.
Prefer self-hosting?
$ curl -fsSL https://plynf.com/install.sh | sh