Synapse / v3.2 / 2025
A tracking platform built for ML teams who treat research as infrastructure. Log runs, compare checkpoints, and reproduce any result — without leaving the lab.
Platform / Capabilities
Six primitives. Together they form a system that captures every decision a research team makes — from the first hypothesis to the final checkpoint.
Drop in two lines of code. Metrics, hyperparameters, system stats, and git state are captured automatically — no manual annotation required.
Compare any two runs side by side. See exactly which hyperparameters changed, which metrics moved, and why a result differs.
Every checkpoint, dataset, and model binary is content-addressed and deduplicated. Reproduce any past result with a single command.
Trace any metric back through every parent run, dataset version, and code commit. The entire provenance tree, queryable.
Get notified when a run diverges, loss spikes, or a checkpoint beats the team best. Slack, email, or webhook — your choice.
Given a run ID, Synapse reconstructs the exact environment, code, data, and configuration. Reproducible by construction.
Research / Comparison
Most teams start with a shared doc. They end with lost runs, unreproducible results, and no memory of what worked. Synapse closes that gap.
"We ran 4,200 experiments last quarter. Before Synapse, reproducing any one of them took a day. Now it takes a command."
— Dr. Lena Voss, Head of ML / Helix AI
Metrics / Scale
Synapse is running in production at labs building foundation models, drug discovery pipelines, and autonomous systems.
Pricing / Tiers
Three tiers. No usage traps. The free plan is enough for a solo researcher; the lab plan is built for a hundred-person team.
Researcher
forever / solo
For individual researchers and students tracking their own experiments.
Team
per seat / month
For small ML teams who need collaboration and full lineage.
Lab
annual / enterprise
For research labs and foundation model teams at scale.