We build the missing stack around AI agents.
One operating stack, five experiments, and selective pilots for teams building agent-native workflows.
One operating stack, five experiments
Agent request enters
Nexus Prime memory Recalls what matters. shipped tetris.codes compression Fits the window. alpha Grain protocol Carries it cleanly. research NXL runtime Runs it cheap. research Phantom missions Directs the work. research Result exits
Result loops back into memory. The stack learns each cycle.
memory
Nexus Prime
Recalls what matters.
shipped
install
npm i -g nexus-prime - npm package nexus-prime@7.9.27
Recover execution context across agent handoffs in a founder-led engineering workflow.
compression
tetris.codes
Fits the window.
alpha
install
curl -fsSL https://get.tetris.codes | sh Compress long coding-agent sessions before context windows become the bottleneck.
protocol
Grain
Carries it cleanly.
research
Carries it cleanly.
runtime
NXL
Runs it cheap.
research
Runs it cheap.
missions
Phantom
Directs the work.
research
Directs the work.
the lab
AI experiments that become infrastructure.
Nexus-PRIME Labs builds the missing stack around agents in public-facing slices and private pilot rooms.
We start with a sharp workflow, prove the smallest durable layer, and keep only the pieces that make agents easier to trust, route, compress, run, or direct.
The work is founder-to-founder: narrow, technical, and honest about maturity.
selective pilots
For teams already feeling the edge of agent workflows.
- FitFounder-led teams with a real agent-native workflow, not a slide-deck curiosity.
- ProcessBring one painful loop. We map the missing layer and choose the smallest useful experiment.
- OutputA working artifact, a sharper operating model, and a decision on whether the layer deserves to harden.