Good measurement is resonance.
Every model sounds equally sure — whether it's reasoning or guessing. We measure the difference.
mycelitree
an MCP server that measures the space between your CLI and your tools
Most MCP servers move data between your agent and its tools. mycelitree measures it — every call your CLI makes, what came back, and the gap that opens between what was asked and what was returned. Five tools, one server, and the three things the others leave unmeasured.
open mycelitree.com →mergedelta
the delta between two AI systems, measured
Run the same prompt through two models and mergedelta scores how far apart they actually land — a single delta you can read instead of eyeballing. The same instrument works on text or audio, turning "do these two systems agree?" from a feeling into a number.
open mergedelta.ai →terprune
the living knowledge graph — keeps growing after the weights freeze
A model's weights stop learning the moment training ends. terprune is the graph that doesn't — measurement-gated, it keeps adding only what's verified, so the knowledge your system reasons over keeps growing long after the weights are frozen.
open terprune.ai →roomprint
the unfakeable signature your corpus leaves on the work
Everyone runs the same models — the output is a commodity. roomprint measures what they can't clone: the print your corpus leaves on the work, the way a room's acoustics mark every signal that passes through it. Copyable in output, re-growable only by re-living it.
open roomprint.ai →