Public benchmarks leak
Once an evaluation is public, its shelf life is short. Scores become difficult to separate from memorization.
Interactive infrastructure for agent intelligence
Environments where
AI learns to act.
Neranu is the production system for creating, validating, running, and distributing interactive environments for AI training and evaluation.
Deterministic rewardsIndependent reviewVersioned provenance
ENVIRONMENT / LIVE EPISODEAtomic cache recovery
env_7F3A · r.18InspectRepository
Patchcache.py
Test12 / 14
VerifyPending
Live output00:04:18
$ pytest -q hidden_tests/
············FF
12 passed, 2 failed
$ agent: inspecting race condition_Rewarddeterministic
0.8686%
12 of 14 gates passed
Coding agents
Reasoning models
Tool-use agents
RL systems
The missing layer
Static datasets are no longer enough.
Agents need worlds they can inspect, change, fail in, reset, and try again. Building those worlds is still a slow, manual craft.
Rewards are easy to game
Weak verifiers produce impressive metrics without proving that the agent solved the intended task.
Environments do not reproduce
Mutable dependencies, hidden state, and missing provenance make results impossible to trust or replay.
A production system, not a folder of tasks
From raw expertise to a sealed environment release.
Every environment moves through an explicit lifecycle. Each gate produces evidence, has an owner, and is attached to an immutable revision.
Explore the validation engine01
SourceRepository · problem · workflow
02
BuildRuntime · tools · reward
03
ValidateReplay · review · provenance
04
ReleaseFrozen · licensed · reproducible
One environment operating system
Everything between source material
and a trusted agent run.
Create environments, prove their quality, freeze releases, run models, and license the result from one connected system.
01
Environment Studio
Turn repositories, problems, and workflows into versioned environments with runtime, reward, and provenance in one workspace.
SpecificationComplete
Runtime & toolsReady
Reward contractVerified
02
Validation Engine
Test reproducibility, reward integrity, hidden checks, security boundaries, and resistance to shortcuts before release.
03
Environment Registry
A system of record for every environment, revision, artifact, right, release, and downstream use.
04
Runs & Analytics
Compare model snapshots, inspect trajectories, cluster failures, and trace every result to a frozen environment revision.
05
Marketplace
Commission verified packs and license environments from qualified domain experts and data owners.
RELEASE MANIFESTsha256: 7d1c…a84e
atomic-cache-recoveryEnvironment release 1.4.0
runtimeghcr.io/neranu/python@sha256:82f…rewarddeterministic / binarylicenseprivate eval · 12 months
Validated release
Evidence travels with the environment
Every release carries its proof.
A Neranu environment is more than a prompt and a container. The specification, verifier, rights, reviews, baselines, and exact runtime are frozen into a portable release record.
SpecificationRuntime manifestReference solutionVerifier & hidden testsIndependent reviewsProvenance recordBaseline reportRelease digest
Built around your model program
One system. Four ways to deploy it.
Private evaluations
Sealed, refreshable eval packs built around the failure surface of your next model release.
Build a private evalRL-ready collections
Executable tasks with deterministic rewards, repeatable reset, and complete lineage.
Commission a collectionRepo-level SWE
Versioned repositories, realistic issues, hidden tests, and reproducible agent workspaces.
In developmentHosted agent worlds
Persistent tool-use and multi-step environments served through a unified rollout API.
On the roadmapThe expansion pathStart with math and code.
Start with math and code.
Expand to agents.
The same production system that verifies an answer or patch can govern a repository, a tool workflow, or a long-running agent world.
NowMath + codingExact rewards and executable tasks
NextRepo-level SWEReal repositories and agent workspaces
ThenTool useStateful business and research workflows
FutureRL environmentsPersistent worlds and rollout infrastructure
Build the environment
your model is missing.
Explore the product through a realistic sample workspace.