Mumpix – Local-first AI infrastructure and $1B developer grant

Hi HN,

I’m working on Mumpix, a local-first infrastructure stack for building AI systems that don’t depend on centralized cloud platforms.

The goal is to make AI memory, state, and infrastructure run anywhere: phones, browsers, desktops, servers, or edge devices.

The stack currently has four layers:

MumpixDB Structured hierarchical database designed for AI memory and application state.

Examples of keys:

memory^assistant^context memory^assistant^recent workflow^task^42^status

Features: • hierarchical key model • prefix scans • watchers for state changes • deterministic snapshots • link resolution • generation-based sync

⸻

MumpixFS

A file substrate designed to integrate with the memory layer. Handles files, aliases, and version references.

Example:

files^alias^report files^versions^report^latest

⸻

MumpixSL

The system-level runtime and daemon (mumpixd) that exposes the stack as shared infrastructure through transports like: • Unix sockets • REST • WebSocket • D-Bus • Binder (Android)

This lets multiple applications share a single memory system.

⸻

MumpixFE

Frontend runtime that runs MumpixDB directly in the browser, allowing the same memory model to work client-side.

⸻

Why this exists

Right now most AI systems rely heavily on centralized APIs and infrastructure.

That works well for model inference, but memory and application state don’t necessarily need to live in the cloud.

A lot of agent systems today store state in: • prompts • vector databases • ad-hoc JSON blobs

MumpixDB is an attempt to build a deterministic structured memory layer that applications and agents can depend on.

⸻

$1B infrastructure grant

To encourage adoption we’re also launching a Mumpix AI Infrastructure Grant.

The idea is to allocate $1B worth of infrastructure and developer resources over time to projects building: • AI agents • local-first AI applications • edge AI systems • privacy-preserving AI tools

The base platform will remain free to build on.

⸻

Feedback welcome

This is still early and I’d love feedback from the HN community on: • the hierarchical key model • daemon vs in-process usage • browser runtime (MumpixFE) • integration patterns for AI agents

Happy to answer questions.