Skip to content

VenusFactory2 v0.3.0

Latest

Choose a tag to compare

@github-actions github-actions released this 02 Nov 17:59
· 315 commits to main since this release

VenusFactory2 v0.3.0

VenusFactory2 brings protein questions, iterative model development and detailed protein research into one workspace.

Research workflows

  • Agent answers questions and finds enabled scientific tools through MCP.
  • Scientist plans, develops, evaluates and revises reusable models and tools, with a separate review before public registration.
  • Researcher combines literature, databases and available predictions into a cited protein report.
  • Shared navigation, session URLs, resumable progress, file previews, Model Plaza and saved workflows connect these modes.

Operations and configuration

  • Separate frontend, API, agent and MCP packages; durable task queues and independent workers.
  • Separate local/online state, session sandbox execution, administrator tool controls and usage accounting.
  • Each online browser visitor receives three new conversations per day across all modes. Shared-IP visitors retain separate balances; IP burst limits remain independent.
  • Optional runtime families cover upstream models, Factory checkpoints, reviewed RSI capabilities and multiple harnesses. Public queries are part of the platform; provider credentials stay local.
  • Supported official model APIs can be configured locally. Online routing remains limited to the operator's enabled free models.

Installation and upgrade

Use the README for quick start and installation guide for component selection and health checks. Linux is the supported environment for sandboxed execution.

Back up existing configuration and runtime data before upgrading. Keep .env, model-config.yaml, runtime installation profiles and runtime/ outside source control. Do not replace a configured deployment with example files. API, MCP and workers must use the same release when restarted.

Large environments, model weights and third-party repositories are installed separately. Historical Gradio/FRP interfaces, notebooks, private experiments and generated demonstration outputs are excluded. Optional scientific skills retain their upstream license and pinned Git submodule revision; initialize the submodule when using a Git checkout.

Existing Git clones

This release removes a previously exposed credential from Git history. Historical commit IDs and the v0.2.0/v0.2.1 tag objects have changed; the code at those historical release tips is unchanged.

For an existing installation, make a fresh clone and transfer only your private configuration and runtime data. Do not merge or push the old history back into the repository. Open pull requests based on the previous history need to be rebased or recreated against the updated main branch.

History cleanup does not revoke a credential or remove copies in external clones and GitHub caches. Any previously exposed credential must still be revoked by its owner.

Validation scope

The release includes source/layout checks, secret scanning, synchronized English/Chinese tutorials, frontend tests and a production frontend build. Backend regression was completed in an isolated checkout with no provider credentials; external-service tests are opt-in. Offline tests do not establish scientific performance, successful inference for every optional GPU model, or unrestricted public-server availability.

This release contains no trained model weights. Existing pretrained models keep their own licenses and installation requirements. A 10,000-model community library remains a research objective, not a claim about the number of models shipped here.

Source release archives omit optional Git submodule contents. python3 initialize.py install --stages skills also works from an unpacked release: it installs the exact upstream revision pinned in configs/bootstrap.json, preserving its license. In a Git checkout the same command initializes the pinned submodule. Existing modified reference libraries are not overwritten.