Research moves fast. Cly keeps every claim connected.

The local-first system of record for computational research—linking sources, methods, code, experiments, results, and decisions in one auditable graph.

Interactive browser demo

Interactive Cly workspace using a dummy project

Explore four curated product views with built-in dummy data. The rest of the workspace remains visible but locked in this public preview.

Neural surrogate reliabilityDummy projectCurated Cly preview
Local
Project overview

Neural surrogate reliability

A computational study of uncertainty-aware neural surrogates across out-of-distribution physical regimes.

  1. Question
  2. Sources
  3. Method
  4. Experiment
  5. Evidence
  6. Claim
Research phaseEvidence consolidationUpdated today
Active claims42 need attention
Experiments41 running
Reproducibility76%Mostly reproducible
Evidence graph1412 relationships

Research direction

The question and hypothesis currently organizing this project

Research questionWhen can neural surrogate models replace expensive physical simulations without hiding failure modes?
Current hypothesisCalibration-aware ensembles preserve decision quality while reducing simulation cost by at least 20×.

Claims and integrity

What is strongest, what is weakest, and what could block publication

Literature, with memory

Search papers without losing why each one mattered.

Cly ranks open literature locally, preserves the ranking rationale, and connects saved methods, limitations, and claims back to the project graph.

Literature WorkspaceNeural surrogate reliability · dummy projectInteractive product surface
Cly Research · Evidence

Compare methods, results, limitations, and claims.

Ranked resultsQuery: “calibrated uncertainty for neural surrogates under compound shift3 papers · rationale preserved
Compatibility, stated plainly

Designed around the tools research teams already use.

  • Local projectsLocal-first
  • GitHubRepository adapter
  • VS Code + CursorEditor-independent
  • Jupyter notebooksScan and register
  • Claude + CodexHarness support
  • Local LLMsOllama-compatible

Cly keeps the research record provider-neutral: run Claude Code, Codex, Cursor or OpenCode harnesses, or keep private work on Ollama-compatible local LLMs. External connection setup continues to harden while the research core remains editor-independent.

The research claim is the central object.

Cly preserves why the work exists—and whether the conclusion still holds.

A notebook can show what ran. A repository can show what changed. Cly keeps those facts connected to the question, evidence, method, decision, and claim they serve.

Watch the graph assemble
Scroll experiment · Research graph assembly

Watch the record assemble around the claim.

Separate research objects move onto one shared graph. Relationships appear only when the record can name them.

ObjectiveDecision quality under shiftObjective 03 · human authored
SourceCompute-matched benchmarkPaper 19 · verified passage
CodeCalibration evaluationsrc/eval/coverage.py · commit 6f2a
ExperimentCompound-shift evaluationRun 024 · 24 seeds · complete
DecisionUse calibration as co-primaryDecision 08 · accepted
ArtifactCost vs calibrationFigure 07 · reproducible
Assembling claimClaim 04 · result claimCalibration-aware ensembles preserve decision quality at 20× lower simulation cost.
6 named relationships84% confidence
  1. 01Objects
  2. 02Relationships
  3. 03Claim
Move to shift perspectiveScroll to assemble
Research workflow

From question to claim, without losing the thread.

Cly keeps the research record beside the work as it changes. Earlier decisions remain visible; new evidence changes the state instead of erasing the history.

01
Plan

Name what the work needs to establish.

Keep the question, hypothesis, metric, constraints, and decision criteria in view before an agent or notebook begins.

02
Build

Carry research context into computation.

Cly Dev gives sessions, code, diffs, terminals, tests, branches, and approvals a place inside the larger research record.

03
Audit

Let changed evidence change the status.

Stale outputs, irreproducible runs, incomplete support, and disagreement remain visible instead of resolving into false confidence.

04
Coordinate

Hand off a record, not a pile of files.

Collaborators and reviewers can see the decisions, open questions, object history, and exact evidence that still needs judgment.

Claim and provenance

Every result has a lineage.

Trace the dummy project from input to claim. Every object below is selectable, and every state represents a different integrity story.

Figure & Table Provenanceneural-surrogate / chain
Selected recordFigure 2 · Cost vs calibration Ready
Artifact objectfigure-2.svg

Feeds Claim 04 through an explicit project relationship.

Cly Dev · Agent session

Delegate the work. Keep its purpose attached.

Start a preplanned dummy session and watch Cly Dev return execution evidence to the research record—not just a chat transcript.

Figure 2 lineage reviewagent/figure-2-lineageLocal

Ready for a research-scoped task.

The session already knows the project, selected objective, branch, and approval boundary.

One connected system

Research defines the record. Dev performs the work. Core keeps the lineage intact.

Primary surface

Cly Research

Questions, evidence, experiments, decisions, claims, and review.

SourcesExperimentsClaimsAudits
Execution surface

Cly Dev

Computational work recorded against the research objective it serves.

Agent sessionsFiles + diffsTerminal + testsBranches + approvals
Shared foundation

Cly Core

Project-scoped state, relationships, provenance, identity, and permissions.

Research graphObject historyProvenancePermissions + sync
Local-first and permissioned

Your research stays yours.

Cly is built around project-scoped local state, visible transfer boundaries, and explicit human decisions. Provider-specific controls continue to harden as external adapters are implemented.

Neural surrogate reliabilityLocal project
Research recordProject-scoped SQLite and referenced local artifactsLocal
Provider accessContext should be previewed and approved before transmissionBoundary
Device identityFingerprint verification, key rotation, and revocationImplemented
Audit historyObserved, approved, and changed state remains attributableVisible
Local by default

Research metadata is project-scoped; large artifacts stay where your work already lives.

Approval before effects

Human judgment remains the boundary for context transmission and consequential actions.

Conflicts stay explicit

Concurrent or contradictory state requires resolution instead of silent replacement.

Direct founder channel

Start a conversation.

founders@cly.one