更新日志

Vecura 的新功能

新模型、新集成和平台改进 — 持续交付

May 20, 2026

Release·v1.4

video 占位符
Subagents fanning out across a 500-target kinome panel

Parallel subagents for kinome-scale selectivity sweeps

Long-running selectivity sweeps now fan out across independent subagents — each scoring a different target family in parallel.

Up to 8× faster on full-kinome panels

A 500-target panel that used to take an overnight run now finishes inside a normal working session. Subagents divide the panel across nodes and stream results back as each completes.

Per-subagent model selection

Pick the best-suited model for each target family — Vecura schedules them onto separate subagents so a slow ESMFold run never blocks a fast Boltz one.

  • ▸GPCRs: Boltz-2 (default)
  • ▸Kinases: OpenFold for ATP-competitive, Boltz for allosteric
  • ▸Membrane proteins: ESMFold with custom restraints

Unified progress view

A single dashboard shows per-target progress, GPU minutes consumed, and any failed runs ready to retry.

image 占位符
Per-target progress dashboard with retry actions

May 19, 2026

Model

image 占位符
New endpoint coverage matrix

ADMETlab 3 endpoints expanded to 220

We added 20 new ADMET endpoints, including expanded CYP isoform coverage and improved hERG channel modeling.

May 18, 2026

Integration

Benchling LIMS sync (beta)

Vecura now reads from and writes to Benchling registries. Hits from a screening run can be promoted to a registered compound in one click.

Two-way sync

Pull existing libraries into a Vecura screen, then push hits back as registered compounds with full provenance attached — run ID, model versions, parameters, and scoring metadata.

Configurable field mappings

Map Vecura fields (SMILES, score, ADMET flags, target) to your Benchling schema once; reuse across projects without re-configuration.

image 占位符
Field mapping editor

May 13, 2026

Release·v1.3

Snapshot rollback across multi-agent runs

Every step in a multi-agent run is now versioned. Roll back to any earlier state — tool versions, parameters, intermediate artifacts.

  • ▸Time-travel debugger UI
  • ▸Cryptographic provenance on every artifact
  • ▸One-click reproduce-on-different-machine

May 6, 2026

Model

Boltz-2 now available

Boltz-2 brings substantially improved accuracy on protein–ligand co-folding for fragments under 500 Da.

Apr 29, 2026

Improvement

@-mention context for SMILES, PDB, sequences, and papers

Point Vecura at exact artifacts inside any prompt — PDB IDs, SMILES strings, FASTA sequences, DOI papers, or internal docs.

Apr 22, 2026

Release·v1.2

Team rules and lab conventions

Teach Vecura your lab's preferred libraries, mandatory safety panels, and IP exclusions once — applied to every future run.

Apr 15, 2026

Integration

Slack & Microsoft Teams notifications

Subscribe a channel to a run and get notified when results are ready, with hit count and safety summary inline.

Apr 8, 2026

Fix

Improved covalent docking pose convergence

Resolved a corner case where covalent docking would not converge for warheads with strained geometries.

Apr 1, 2026

Release·v1.1

Python SDK and Jupyter kernel

Drive Vecura from any notebook, script, or production pipeline. Outputs are returned as Pandas DataFrames, RDKit molecules, or PDB files.

更新日志结束 · 订阅获取更新 →