The Product

An AI lab in a box.

A lab needs a place for code and data, reproducible experiments, team chat, and compute. TensorSpace sets them up together on one box, so you can get to work instead of assembling the lab yourself.

Book a Demo See What's Inside

Early access · open-source core

What's Inside

Four systems. One box.

Everything a working lab needs. Each part is open source, and everything is set up to work together.

Rust · MIT

tsp

Reproducible pipelines on plain git.

Define stages in tsp.yaml and tsp uses git object IDs to tell what is stale. tsp repro reruns only what changed. Each experiment is a real commit you can diff, share, and reproduce. LLM pipelines work the same way as any other pipeline.

Go · MIT

Git + Data tab

Your repos grow a lab UI.

A full git host where every repo has a Data tab: pipelines, metrics, plots, and experiments rendered right next to the code that produced them. Review a model the way you review a pull request.

Rust · MIT

tensorchat

Team chat that lives in the lab.

Channels, DMs, threads, reactions, search, and bots in one binary, one SQLite file, and a 32 kB web client. Your conversations stay on your box beside the work they are about.

Rust · MIT

gal

Documents that edit themselves live.

Real-time collaborative documents on a real operational-transform engine. It supports Wave-style, character-by-character co-editing from a single binary and a single database file.

Reproducible by Default

Experiments you can replay.

Declare your stages once. tsp repro reruns only what changed and writes the result into git. The repo's Data tab then shows the pipeline, metrics, and plots beside the code that produced them.

  • Every experiment is a commit you can diff and share
  • No daemon or separate server. Git carries the lab
  • Review a model the way you review a pull request
$ tsp repro
reading tsp.yaml · resolving against tsp.lock

3 stages stale · prepare, train, evaluate
├─ prepare   data/raw.csv → data/train.parquet ✓
├─ train     model.pkl rebuilt in 41s ✓
└─ evaluate  metrics.json written ✓

$ tsp metrics diff
accuracy   0.912 → 0.937   (+0.025)
loss       0.341 → 0.288   (−0.053)

lockfile updated · commit to publish the experiment

How It Works

The box is the unit.

One box is one lab: your git host, your chat, and your docs come up together under your own name, with all of your data in one place.

  • Each box belongs to one tenant, with no shared data
  • Every component is open source; the box is just the easy way to run them
  • Take your data with you at any time. It is all git, SQLite, and files

Early Access

Pick your orbit.

Every tier includes the same lab. The difference is how much hardware and support you need.

Solo

One researcher, one box.

  • All four components, single user
  • Shared infrastructure
  • Bring your own model keys
  • Community support
Get early access
Most wanted

Team

The whole startup in one place.

  • All four components
  • Metered AI gateway included
  • Example pipelines seeded on day one
  • Priority support
Get early access

Lab

Research groups with real compute needs.

  • Dedicated hardware
  • GPU attach for training runs
  • Longer retention
  • Onboarding with your data
Get early access

Enterprise

Identity, compliance, and scale.

  • Single-tenant everything
  • SSO and identity integration
  • Region choice and retention policy
  • Support with an SLA
Get early access

Mission Control

Bring up your lab.

Fifteen minutes, your use case, and a straight answer on whether the box fits it.

Book a Demo