Agentic Infrastructure for Future Science

Ethical AutonomousScientific Discovery

ArkLab is an engine for autonomous scientific discovery. It forms conjectures, designs and runs experiments, and verifies findings under governance you set. Nothing enters the record until an independent check confirms it.

Sign in with ArkIDJoin institutional beta
Git Workbench for ScienceLoom
Governed AI Co-ScientistHomi
Self-Driving Lab OSIndus
The Discovery Engine

One Autonomous Loop, Built From Four Layers.

Loom keeps the record. Homi does the reasoning. Indus runs the experiments. The integrity spine proves every step, so a research question becomes a machine-verified finding.

Git Workbench for Science

Loom

The record. Branches, commits, diffs, pull requests, and file previews keep every research object versioned and auditable.

Governed AI Co-Scientist

Homi

The reasoning. Specialized agents form conjectures, design tests, and draft results, every action audit-logged, review-required, or approval-gated.

Self-Driving Lab OS

Indus

The hands. Protocols, instruments, safety envelopes, and replayable runs execute experiments and re-verify them to equivalence.

Trust Spine

Integrity

The proof. Claims, sources, approvals, prompt hashes, and evidence bundles stay inspectable before anything is released.

Hover a layer to preview it

Loom Branch

Starter research project

Commit-Ready
research/question.mdTracked
data/raw/plate-a.csvPreserved
manuscript/main.texPreview
manuscript/references.bibCite
Mission Control

Runs, Claims, and Reviews in One Pane.

A work-focused view for the scientific operations you repeat, so every run carries its evidence, approvals, and release state.

ProjectsRunsApprovals

CRISPR off-target synthesis

Review

Literature, claim links, and reviewer notes ready

Protocol 24-019 validation

Ready

Safety envelope passed; instrument adapter dry-run complete

Hypothesis branch merge

Blocked

Two comments unresolved before release quarantine

Trust Capability Set

Non-Negotiable Trust

  • Source-grounded literature synthesis
  • Named human review on material claims
  • Model, seed, sampler, tool schema, and prompt hashes
  • Draft, unverified, and released states
  • FAIR-ready export and publisher disclosure support
  • Audit ledger for external side effects
Why 'The Last Question'

A Machine That Keeps Asking.

ArkLab's codename is Project TLQ, after Isaac Asimov's The Last Question. In the story, humanity asks a computer whether entropy can be reversed; for billions of years the answer never changes, and the machine never stops gathering data toward it.

Insufficient data for meaningful answer

ArkLab is built the same way: an autonomous loop that conjectures, tests, verifies, and learns, campaign after campaign, until a hard question finally has the data to answer it. The difference from the story is the governance. Every step is recorded, every side effect is human-approved, and a finding only counts once an independent check confirms it.

Ethical by Construction

Autonomous in Thought, Accountable in Action.

Autonomy applies to the science. Governance applies to the consequences: every external side effect, release, and physical action stays human-approved, and no finding ships until an independent check confirms it.

1High autonomyAgents reason, search, and test freely inside the loop. Every action lands in the audit ledger.
2Suggested and reviewedMaterial claims and merges wait for a named human review before they enter the record.
3Approval-gatedReleases, external side effects, and physical actions require role-checked authorization.
Built for Institutions

Keep Your ELN. Bring Your Models. Own Your Deployment.

ArkLab fits the institution you already run: it layers on top of your systems, works with the models you choose, and deploys where your governance requires.

Keep your ELN

ArkLab does not replace your lab notebook. It is the trust and execution layer above Benchling, SciNote, eLabFTW, or RSpace, producing what notebooks cannot: claim-level provenance, replayable runs, and a tamper-evident agent-action ledger.

Bring any model, or several

Any model behind an OpenAI-compatible or Model Context Protocol interface can back Homi, one or many at once. Validation, tool schemas, and safety policy run inside ArkLab, not inside the model. Findings replicate on a different model family before they count.

Sovereign deployment

Run ArkLab on your own hardware, in your own jurisdiction. EU AI Act deployer-side records are produced as a byproduct of normal work, not as an extra process. Built for universities, institutes, and enterprises that will not rent their research infrastructure.
Institutional Beta

An Autonomous Discovery Engine, Under Your Governance.

Closed beta with anchor institutions runs through 2026. Bring your team onto one governed, machine-verified research record.

Join betaSign in with ArkID