openlabCB Research Workspace
Define a high-level research question and measurable virtual objective.
Research control room
Track objective, evidence, simulations, decisions, and follow-up work.
—
No simulation run yet
— · refreshes while active
Guardrails
Built for conceptual research, not executable wet-lab operations.
Run trajectory
Best and mean virtual objective score across iterations.
Top candidates
Latest abstract ranking.
Priority tasks
What the research workspace says to do next.
Project health
Lightweight completeness indicators.
AI Copilot
Research framing, data interpretation, virtual study design, and uncertainty analysis.
Optimize
Run a reproducible abstract search over generic scientific descriptors.
Search settings
These parameters control the virtual search only.
Best result
No run yet.
Iteration history
Convergence, uncertainty, and top candidates.
Candidate comparison
Top descriptors from the latest run.
| Candidate | Score | Uncertainty | Stability | Selectivity | Solubility |
|---|
Autopilot trace
AI strategy decisions across cycles.
Chemistry
Work with molecular identities, formula descriptors, reaction-study records, analytical observations, materials, electrochemistry, and formulation datasets.
Molecular identity
Formula analysis is deterministic and transparent. SMILES/InChI are stored as identifiers unless an external cheminformatics engine is connected.
Molecule library
Portable project records; no claims of experimental validation.
Cheminformatics comparison
Real RDKit fingerprint and Tanimoto similarity utilities. Similarity is an informatics metric, not a biological or safety prediction.
Reaction balance checker
Checks conservation of elements in a written equation; it does not prescribe reaction conditions.
Chemistry study records
Capture conceptual reaction, spectroscopy, materials, electrochemistry, and formulation records.
Recent chemistry records
High-level metadata and observations only.
Biology
Sequence and protein analysis, assay metadata, omics datasets, pathways, cell-model records, and bioinformatics adapter hooks.
Sequence / protein analysis
Built-in checks cover length, alphabet, nucleotide GC%, and a transparent protein mass estimate.
Sequence library
Sequence strings are project data. Validation is informational, not a design or wet-lab protocol.
Sequence / variant QC
Optional Python-backed QC for FASTQ read quality and VCF variant metadata.
Sequence utilities
Reverse-complement and translation are informational sequence operations.
Biology records
Track assays, omics, pathways, cell models, and protein metadata.
Recent biology records
Metadata and analysis-ready context.
Virtual Organism Studio
Create professional in-silico bacteria, virus, or parasite research records for comparison, provenance, and safe simulation. A digital organism is a computational object, not a physical organism or engineering recipe.
Saved virtual organisms
Neutral lineage model
The lineage model tracks abstract neutral mutation counts only. It does not generate redesigned genomes or optimize virulence, host range, immune evasion, transmission, or antimicrobial resistance.
Microbiology, virology & parasitology workbench
Analyze bacterial, viral, or parasite nucleotide data without turning computational screening into a clinical diagnosis. Built-in tools are descriptive, screening, or model-only and preserve provenance.
Reference comparison & phylogeny
Paste one item per line as label|SEQUENCE. Reference comparison is a fast 5-mer screening method; tree inference is a screening UPGMA view, not publication-grade phylogenetics.
User-supplied marker screen
OpenLab does not bundle pathogen/virulence/resistance marker sequence databases here. Supply your own validated marker panel as label|SEQUENCE.
Safe microorganism simulation models
Bounded mathematical models for neutral mutation counts, generic population growth, and generic epidemiological spread. They do not model pathogen optimization or laboratory conditions.
Neutral evolution
Population dynamics
Generic spread (SIR)
Microbiology analysis history
Saved results include hashes, method, model/calculation status, version and scientific-validity boundary.
Genome feature map
Parse GFF3 feature coordinates and visualize a lightweight genome track.
Genome browser
Client-side overview of parsed feature intervals.
Engines
OpenLab starts with its built-in science stack and automatically bootstraps compatible specialist engines on the host. Extra licensed, cluster, WSL, or remote engines can still be registered below.
Chemistry coverage
Each capability is labelled in plain language so users know whether it is usable right now.
Biology coverage
The same transparent boundary applies to molecular, omics, structural, and systems-biology engines.
Installed & detected engines
OpenLab checks its private engine pack and the computer's normal PATH. Use Scan computer to also inspect Conda environments and WSL.
python start_openlab_full.py can install compatible open-source engines into .openlab-engines, but OpenLab also detects engines the user installed independently through system PATH. Scan computer additionally checks Conda environments and WSL. From localhost, Select executable path can pin a known engine executable that is not on PATH. WSL-only engines remain a setup state until connected through a WSL compute bridge or endpoint.Register an engine
Connect a validated external tool or local service without putting it inside OpenLab.
Test endpoint sends a GET request and expects a successful response. Engine execution sends a POST JSON request using openlab-engine-v1 with task, input, and resources. Local/private HTTP(S) endpoints are allowed by default; public remote endpoints require HTTPS and the server administrator to set OPENLAB_ALLOW_REMOTE_ENGINE_ENDPOINTS=1.Registered engines
Endpoints are metadata until you explicitly test and validate them. Registration does not claim scientific validation.
Scientific Tools
Unit-aware calculations and transparent statistics that can be reproduced from the project file.
Unit converter
Supports common scientific dimensions without hidden assumptions.
Statistics
Descriptive statistics or simple linear regression. Results are analytical aids, not clinical evidence.
Calculation history
Recent built-in calculations are preserved in the project.
Experiment design
Generate generic full-factorial designs without encoding operational lab procedures.
Uncertainty & planning
Independent uncertainty combination and normal-approximation power planning.
Signal peak detector
Detects simple local maxima in an x/y series.
Dataset quality control
Missingness, duplicate rows, numeric summaries, and simple outlier flags.
Space-filling DOE
Latin-hypercube design for generic parameter studies.
PCA
Project numeric matrices into principal components using the optional Python scientific runtime.
Clustering
K-means exploratory grouping for numeric observations.
Network analysis
Graph degree, connectivity, and density for pathway-style or interaction-style edge lists.
Advanced statistics
Screening tools with explicit approximation notes.
Response-surface DOE
Central-composite-like planning aid for bounded factors.
Data Formats
Parse common chemistry and biology text formats and keep the import provenance with the project.
Parse scientific text
Built-in: CSV, TSV, FASTA, FASTQ, VCF, BED, GFF/GTF, SMILES, MOL, SDF, JCAMP-DX metadata, mzML metadata, JSON. GenBank/PDB/mmCIF use the optional scientific Python engine.
Project scientific files
Small text metadata can be embedded in the portable project; larger files can be uploaded as resumable file-backed artifacts or converted to Parquet.
Domain parser / QC
Use domain-aware engines on scientific text without replacing the original provenance record.
Scientific export
Export project records without modifying the project.
Binary scientific data
Profile larger HDF5, Parquet, NetCDF, or mzML files without importing them into the project.
Large-file guidance
Large raw files stay outside the portable project; only hashes/metadata belong in the project bundle.
Jobs
Longer analyses can run as named jobs with status, progress, results, and cancellation.
Submit job
Job inputs are JSON so the same job can be reproduced from the project file.
Queue
File-backed job state with restart recovery; export the project for portable ownership.
Compute & Engines
Keep the OpenLab server as the coordinator and run scientific workloads on your own computer, an intentionally shared lab GPU worker, or a registered remote GPU service.
Choose where your compute runs
Personal workers are bound to your anonymous OpenLab session. Shared workers are explicitly administered. Remote services use the standard engine endpoint protocol.
Use the CPU/GPU in your own Windows, Linux, WSL, or workstation machine. Jobs from other OpenLab sessions cannot use a personal worker.
Use an administrator-provided GPU server or cluster that is intentionally available to multiple OpenLab sessions.
Connect a private/server/cloud scientific API using openlab-engine-v1. Useful for managed GPU services and specialist engines.
Use the GPU already exposed by a supported browser for OpenLab's client-side kernels. No worker installation and raw data can stay in browser memory.
Browser GPU · zero-install WebGPU
Run OpenLab-owned GPU kernels directly in this browser. This is ideal for local matrix/data acceleration; native programs such as GROMACS, ORCA, xTB and ColabFold still use a paired worker or endpoint.
Remote accelerator service runner
Run a registered GPU/accelerator API without leaving Compute & Engines. Choose a service and task; advanced input stays optional for custom services.
Advanced service input
Only needed when your custom endpoint expects structured input. Normal service-specific adapters can keep this as {}.
Engine discovery
OpenLab detects supported executables on the server. When started with the full launcher, installed local engines are exposed automatically; remote/WSL/GPU engines can still be added with another bridge.
Connected bridges
Trusted bridge pairings persist across OpenLab restarts until explicitly revoked. Offline workers can reconnect with the same token.
Guided engine task
Run common quantum/semiempirical jobs without writing JSON. OpenLab chooses a compatible local or connected worker.
Stage project artifact to a worker
Attach a file already stored in OpenLab to a compute job. The bridge verifies its SHA-256, stages it inside its sandboxed workspace, and can return output artifacts automatically.
Run specialist scientific job
Choose the scientific engine/task and complete the guided fields. Raw JSON is optional expert mode only.
Advanced / expert input
Normal scientists do not need this. Enable it only when you intentionally want to edit the exact allowlisted input object.
External compute queue
Multiple bridges can consume jobs concurrently; job metadata is stored in this anonymous project.
Pair a personal GPU computer
Use three simple steps: 1) download the standalone worker on the GPU computer, 2) click Pair my computer to copy a private pairing command, and 3) run that command with Node.js 20+ on the GPU computer. The worker connects outbound to OpenLab, so no inbound GPU port is required.
Click “Pair my computer” to generate a private worker command.
localhost before pairing.Cluster / GPU status
Live bridge capabilities, GPU availability, queue depth, and worker health.
OpenMM simulation adapter
Run a bounded computational PDB simulation when OpenMM is installed on the OpenLab science runtime.
Visualize
Build simple dataset plots and preserve the chart specification with your project.
Canvas
Client-side rendering from the selected dataset snapshot.
Saved plot specs
Reusable chart definitions.
Trajectory topology summary
Read-only topology metadata from a PDB structure using MDTraj when available.
Protein / structure viewer
Lightweight canvas viewer for PDB Cα coordinates. Drag to rotate.
Spectrum / chromatogram
Plot x,y pairs from exported instrument data.
Sequence map
Client-side visual overview of sequence composition.
Correlation matrix
Computed from the selected numeric dataset columns.
Reaction canvas
Lightweight visual layout for reaction SMILES components; verify chemistry with RDKit before relying on it.
Figure export
Export the currently rendered plot/figure as a high-resolution PNG.
Literature
Capture DOI-linked references and make source provenance explicit before using AI-generated interpretations.
Search literature
Searches Crossref metadata; verify the source before treating it as evidence.
Resolve DOI
Creates a normalized DOI link; title/authors should be verified before publication.
Evidence checklist
Use this before treating literature-derived content as authoritative.
Evidence claim
Link a claim to a source and record whether the source supports it.
Evidence ledger
Project-level claim/source relationships.
PubMed text metadata
Retrieve publisher/NCBI XML metadata for a PMID. Availability depends on the source.
Evidence comparison
Screen two excerpts for shared terminology and possible contrast cues. This is not a contradiction proof.
Experiments
Plan, annotate, and archive conceptual experiments without storing executable protocols.
| Title | Status | Hypothesis | Objective | Updated |
|---|
Conceptual experiment
Datasets
Import datasets for analysis. Projects can persist to ordinary files, and structured tables can be compacted/exported as Apache Parquet without introducing a database.
Import CSV
Quick analysis import keeps a bounded project preview; use Resumable scientific upload for large files.
Project datasets
Preview, analyze, version, and delete datasets associated with this project.
Resumable scientific upload
Chunk large CSV/TSV and other scientific files directly to the file store. Uploads resume after a server restart; CSV/TSV can stream into Apache Parquet without a database.
Selected dataset analysis
Descriptive statistics and pairwise numeric correlations.
Analytics
Project health, run comparisons, completion signals, and dataset diagnostics.
Run comparison
Recent virtual runs ranked by best observed proxy score.
| Objective | Best score | Uncertainty | Iterations | Population | Created |
|---|
Completeness
Useful indicators for a reproducible project package.
Dataset diagnostics
Numeric columns, missingness, uniqueness, and correlations.
Notebook
Keep reasoning, decisions, references, and follow-up tasks beside the project.
Research notebook
Plain text / Markdown is stored in the project file.
Tasks
Small project actions without account infrastructure.
References
Save citation metadata or paper links for project context.
Saved references
Included in search, AI context, and exported reports.
Collaboration
Work together without accounts or a database. Access is capability-based and rooms are ephemeral; each browser keeps its own durable project file.
Your anonymous identity
No email, password, username, or login is required. This name is only a local display label.
Join a project
Use an owner-provided editor or viewer invitation.
Create a collaboration room
Creates a file-backed relay room from your current project while keeping the project portable and user-owned.
Room status
Real-time relay with no SQL/database; project and room state can be file-backed locally.
Anonymous members
Presence is activity-based; no account is created.
Peer-to-peer link
Optional WebRTC data channel for direct browser-to-browser project operation transfer. The relay only carries short-lived signaling messages.
Room controls
Owner-only safety and recovery controls.
Room version history
Recent history is kept in the file-backed room snapshot and mirrored locally for recovery.
Team comments
Short discussion threads that stay inside the ephemeral room.
Live activity
Recent room events.
Agents
Use bounded scientific agents for planning, analysis, statistics, chemistry, biology, and peer review. Agents operate through explicit project-safe tools.
Run an agent
Plans are recorded in the project for reproducibility.
Agent catalog
Each agent advertises the tools it may use.
Parallel agent panel
Run independent reviewers in parallel; combine their outputs in your project.
Background agent jobs
Queue an agent task without keeping the browser request open. State remains ephemeral and is exported with the project snapshot.
Agent memory
Project-scoped notes can be reused by future agent runs. Stored locally with the project.
Recent agent runs
Ephemeral session history is included in the project file.
Integrations
Register local services, file-drop connectors, or instrument inboxes without accounts or a central database.
Register connector
Metadata only until you send an inbound file or connect a validated local service.
Inbound file
Store a scientific file in the current project and retain source metadata.
Configured integrations
Connectors are project metadata and remain portable in your .openlab file.
Local instrument / LIMS request
Queue an allowlisted request for a local/private-network instrument bridge. No cloud account is required.
Integration policy
REST connections are restricted by the local bridge and require explicit instrument/network opt-in.
Record instrument run
Track vendor, sample, run, calibration, and linked artifact metadata without storing vendor credentials.
Recorded runs
Portable run metadata linked to your project.
Plugins
Register portable plugin manifests without installing arbitrary code into the server.
Register manifest
Untrusted manifests remain metadata only until a human validates the external engine.
Installed manifests
No plugin code is executed by OpenLab core.
Workflows
Chain safe built-in scientific tools into a recorded, repeatable project workflow. No arbitrary code execution.
Build workflow
Create an allowlisted scientific workflow from guided web controls. Expert JSON remains available but is not required.
Advanced / expert workflow JSON
Workflow history
Results are stored in the project and travel with its local file.
Artifact integrity
Create a SHA-256 manifest entry for a project artifact.
Reproducibility notes
Hashes prove content identity; workflow records preserve operation inputs, outputs, and timing.
System Readiness
A user-facing view of what works now, what needs installation, and what only needs setup.
What an ordinary user needs
Scientific software normally stays on the OpenLab server or paired workers.
What the server administrator handles
Install or connect specialist capability once for the whole deployment.
Python & scientific modules
These are the Python-backed tools OpenLab can call on this computer right now.
Scientific capability modes
Ready means executable now; Limited means OpenLab has a usable workspace but an advanced external engine may still be required.
External scientific engines
Supported engines are shown separately from engines actually installed on this computer.
AI readiness
Shows configured cloud providers and private AI services detected on this OpenLab host.
Architecture constraints
Deliberate consequences of no database and no login.
External deployment completeness
OpenLab probes specialist engines, host hardware, sandbox backends, instrument replay, network traversal, and scientific evidence readiness without pretending external dependencies are installed.
Identity
Your anonymous identity can be moved between computers with an encrypted recovery file. No account is created.
Current identity
Stored only in this browser.
Recovery guidance
No account means you control the identity file.
Project
Your project is anonymous. Export it whenever you want durable ownership.
Ownership & portability
Your project can live entirely on your own PC.
File-backed storage
No database. OpenLab can persist projects, artifacts, CRDT room state, and analytics tables as ordinary files; Parquet compaction is optional for large structured datasets.
Local project library
Private browser snapshots stored on this PC.
Workspace history
Recent optimization runs and packaged events.
AI Providers
Use cloud AI or private local models. OpenLab supports native OpenAI, Anthropic and Gemini APIs, Ollama, plus OpenAI-compatible servers such as LM Studio, LocalAI, vLLM and llama.cpp.
Local AI readiness
Checks common loopback endpoints only. Nothing is sent to the internet by this discovery check.
Audit
Transparent record of important workspace actions during this session.
| Time | Action | Details |
|---|