IX Foundry · Coming soon

Frontier AI, on compute people own.

IX Foundry is the full-stack lab to train, evaluate, deploy and continuously improve custom models and agents — running on IX's tokenized GPU clusters. Builders get frontier infrastructure. The people who own it earn from every run.

Coming soonOwn the compute
11
GPUs live on testnet
9
clusters tokenized
100%
settled on-chain
THE FLYWHEEL

One loop. Two sides.

Foundry is the demand engine of IX: AI workloads generate the revenue that tokenized infrastructure pays out as on-chain yield.

01

Own

Investors buy fractional shares of GPU clusters on the IX marketplace.

02

Power

Those clusters become Foundry capacity for training and inference.

03

Build

Teams train, evaluate and deploy custom agents on owner-backed compute.

04

Earn

Every metered run settles on-chain and distributes yield to owners.

01RL Environments

RL Environments

Turn any task or workflow into a reinforcement-learning environment. Scaffold, iterate, evaluate and publish from a single CLI loop.

  • Verifiable, reward-driven tasks
  • One loop: init → dev → eval → push
  • Community environment hub
Create environmentsSoon
FIG.1
FOUNDRY CLI
$ foundry env init browser-qa
created ./browser-qa/env.py
$ foundry env dev
rollout 24/24 · sandbox ok
$ foundry env eval --model ix-7b
reward 0.84 ▲ · passed 21/24
$ foundry env push
02Evaluations

Evaluations

Hosted evaluations to benchmark every checkpoint against open models — no infrastructure, no setup.

  • Benchmark vs. 100+ open models
  • Zero infra to manage
  • Public leaderboards at launch
Run your first evalSoon
FIG.2
ModelScore
ix-foundry-32b84.2
open-70b71.8
open-8b63.4
baseline-7b51.9
benchmark: agentic-tasks-v2
03Hosted Training

Hosted Training

Fine-tune and post-train large models on IX's tokenized Orbit Clusters — managed end-to-end, from supervised runs to large-scale RL.

  • SFT, LoRA & large-scale RL
  • Runs on tokenized Orbit Clusters
  • Applied research support
Start trainingSoon
FIG.3
04Inference

Inference

Ship any checkpoint to dedicated or serverless endpoints with native LoRA support — every call metered for on-chain settlement.

  • One-click deployment
  • Native LoRA adapters
  • Metered, on-chain billing
Deploy a modelSoon
FIG.4
05Proof-of-Compute

Proof-of-Compute

Every training run and inference call produces a signed usage receipt that settles on-chain — revenue builders can verify, and yield owners can trust.

  • Signed usage telemetry
  • On-chain settlement
  • Public proof-of-revenue
See the ledgerSoon
FIG.5
ENVIRONMENT HUB

A hub of reusable environments.

Community-built environments for training and evaluating agents on real, verifiable work.

ExploreStarredMy EnvironmentsPreview
ENV.01412
terminal-agent
agentscli
foundry-labs2d agov0.4.2
ENV.02356
code-review
codeqa
foundry-labs5d agov0.3.8
ENV.03289
browser-research
webagents
community1w agov0.2.1
ENV.04247
math-verify
mathrl
community3d agov0.5.0
ENV.05198
sql-analyst
datasql
community6d agov0.1.9
ENV.06164
doc-extract
docsqa
community2w agov0.2.4
OPEN TOOLING

Open-source at the core.

foundry-envs

Author verifiable RL environments in Python.

import foundry_envs as fe
 
env = fe.load("terminal-agent")
score = env.eval(model)
foundry-rl

Distributed post-training, from SFT to large-scale RL.

$ foundry rl train \
--env terminal-agent \
--model ix-7b \
--nodes 4
foundry-receipts

Verifiable usage receipts, settled on-chain.

$ foundry receipts ls
run-8842 train 4.2h settled
run-8843 infer 1.1m settled
epoch-42 yield → owners

Own the compute. Train the future.

IX Foundry is coming soon — the lab where AI workloads become real, verifiable yield.

Own the compute