Skip to content

DeepScientist v1.6.0

Latest
Latest

Choose a tag to compare

View all tags
@ResearAI ResearAI released this 13 May 06:57
v1.6.0
3fa3485

DeepScientist v1.6.0 is a major capability release.

The release expands DeepScientist from a Codex-first local research workspace into a broader multi-runner, benchmark-aware, and natural-science-ready research operating system.

The biggest changes are:

  • first-class support for Claude Code, OpenCode, and Kimi Code runners
  • BenchStore and AISB benchmark catalog support
  • Science Evidence Graph and science-task discipline for computational science workflows
  • a large science package-routing skill with generated package cards for dry-lab scientific software workflows
  • stronger mobile/workspace surfaces, TUI utilities, diagnostics, install, and update behavior

Highlights

1. Claude Code, OpenCode, and Kimi Code support

DeepScientist now supports more than the Codex runner path.

This release adds and hardens runner support for:

  • Claude Code
  • OpenCode
  • Kimi Code
  • continued Codex compatibility improvements

The work includes runner settings surfaces, doctor/startup probes, provider failure diagnostics, MCP compatibility fixes, YAML/tool-call handling fixes, and docs for configuring non-default runners.

This is the most important runtime expansion in v1.6.0: users can now choose the agent backend that best fits their local setup and provider access.

2. BenchStore and AISB catalog

v1.6.0 introduces BenchStore as a benchmark discovery and installation path.

Key pieces include:

  • BenchStore catalog YAML support under AISB/catalog/
  • AISB B1-B12 benchmark catalog entries
  • 100+ AISB task catalog entries with bilingual metadata
  • homepage, discovery, display, and extended catalog fields
  • BenchStore YAML reference docs
  • GitHub Releases distribution spec for benchmark assets
  • one-click style prompts and install metadata for benchmark workflows

This makes DeepScientist much better suited for benchmark-driven research and AI scientist system evaluation.

3. Science Evidence Graph and dry-lab science workflows

This release adds DeepScientist-native support for natural-science and engineering workflows without importing a separate runner, HPC, CLI, or web stack.

The core design is:

  • real execution stays in bash_exec
  • durable scientific evidence is recorded through artifact.science(...)
  • Canvas can reconstruct science evidence nodes from durable artifacts
  • claims are typed as computed, parsed, digitized, or hypothesis

The new science skill adds package-routing and evidence discipline for dry-lab work across quantum chemistry, molecular simulation, bioinformatics, astronomy, materials, CFD, FEM, geoscience, nuclear, particle physics, and related computational domains.

The science skill includes 169 generated package cards inspired by FermiLink package-routing ideas, while keeping FermiLink out of the DeepScientist runtime.

4. Nature and paper-production companion skills

v1.6.0 also expands publication-facing workflows.

New or expanded skills include:

  • Nature-style figure workflow
  • Nature data availability support
  • Nature paper-to-PPT workflow
  • academic polishing
  • paper outline gates
  • paper plot templates
  • stronger paper review and write discipline

These are intended to make DeepScientist more useful after experiments finish, especially when turning evidence into figures, reports, slides, or manuscript-ready text.

5. Workspace, mobile, and TUI improvements

The workspace received a broad usability pass:

  • mobile quest workspace refinements
  • mobile navigation and onboarding E2E coverage
  • workspace debug inspector improvements
  • file links opening directly from chat
  • continuation notice improvements
  • TUI utility panels and debug tooling
  • workspace file-tree and explorer payload improvements
  • responsive settings and onboarding fixes

The practical result is a workspace that behaves better outside the ideal desktop-only path.

6. Install, doctor, update, and runtime hardening

v1.6.0 includes many reliability fixes around local installation and diagnosis:

  • deterministic npm ci install behavior
  • runtime Python environment invalidation on install
  • better Codex startup probe explanations
  • clearer provider failure diagnostics
  • Windows / WSL2 setup guidance
  • Windows Codex GBK handling fixes
  • bounded bash logs and oversized tool-result sidecars
  • update and launch path hardening

One practical note: users on 1.5.17 may need to update manually with:

npm install -g @researai/deepscientist@latest

The old 1.5.17 self-update path had a launcher-side npmBinary bug. The 1.6.0 launcher has that path fixed.

Detailed Change Areas

Runner support

  • added Kimi runner and settings/docs surfaces
  • added experimental Claude runner support and later hardening
  • added OpenCode-related setup and runner support
  • improved runner startup checks and provider diagnostics
  • fixed Claude prompt/tool argument handling
  • improved MCP and long-running tool compatibility

BenchStore / AISB

  • added BenchStore catalog support and YAML reference docs
  • added AISB benchmark catalog entries and images
  • added extended catalog fields for homepage, discovery, and display
  • documented GitHub Releases packaging for benchmark assets
  • improved BenchStore tutorial and startup flow

Science

  • added artifact.science(...)
  • added science node types for package checks, computational runs, dataset analyses, parameter sweeps, validation results, and claims
  • added Science Evidence Graph UI integration
  • added the science skill with package index, domain index, package-check playbook, HPC-through-shell discipline, and claim-type discipline
  • added FermiLink attribution/provenance without making it a runtime dependency

Research workflow

  • added paper outline gates and academic outline workflow
  • improved paper write/artifact deltas
  • improved publishability stop-loss guidance
  • added Nature companion skills
  • strengthened future-facing limitations and decision discipline

UI / UX

  • improved mobile and responsive workspace behavior
  • improved onboarding and runner switching
  • improved lab canvas layout and trace surfaces
  • improved admin/debug surfaces
  • replaced the hero orb asset with the DeepScientist logo

Reliability and docs

  • improved doctor output for Codex startup probe failures
  • improved install reproducibility
  • improved Weixin long-poll and proxy guidance
  • updated Quick Start, settings, provider, WSL2, BenchStore, and acknowledgement docs
  • refreshed npm package metadata to 1.6.0

Upgrade

npm install -g @researai/deepscientist@1.6.0

or:

npm install -g @researai/deepscientist@latest

Verify with:

ds update --check
ds doctor

Version

  • npm package: @researai/deepscientist@1.6.0
  • npm release: published
  • git tag: v1.6.0
Assets 2
Loading