Who this is for
- Lab analysts reviewing assay data, sequences, and QC checkpoints
- Research teams that want AI assistance without sending everything to a public chat UI
- Organizations that need human approval before agents touch files or run tools
- Teams already using Neural Junkie for software who also run biology workflows
Official Life sciences pack
Install from Domain packs (⌘⇧K) → Store → Life sciences. Or pick Life sciences & lab work in the first-run wizard.
| Component | What it does |
|---|---|
| BiologyExpert | Preset agent with bio MCP tools and OpenBio-tuned chat |
| OpenBioLLM 8B | koesn/llama3-openbiollm-8b:latest — recommended Ollama pull |
| Tool runner | qwen2.5:7b routes MCP tools when the chat model has no native tools API |
| analyze_sequence | DNA/RNA/protein checks, length, reverse complement |
| fold_protein | ESMFold via Hugging Face Inference → PDB artifacts under hub data |
| Sequence review runbook | Importable template for structured review workflows |
Typical workflow
- Download Neural Junkie and complete the wizard with local models or cloud HF token for folding.
- Enable Life sciences pack; pull OpenBioLLM + qwen via Model library (⇧⌘M).
- DM BiologyExpert or @mention in a channel for sequence questions.
- Use
/collaboratefor multi-step analysis write-ups with review and approval phases. - Configure HF token under Settings → AI Providers when using ESMFold or GGUF import.
Customer & instrument packs
The public Life sciences pack covers general biology tooling. Organizations with proprietary formats (e.g. plate readers, multiplex QC, TIM imports, custom scan viewers) can build a customer pack with Pack Dev Studio and install via sideload zip — same hub, private specialists and viewers, no public store listing. See Domain packs and PACKS_CUSTOM.md.
Privacy & data locality
With local Ollama, chat and tool routing stay on your workstation. ESMFold and Hugging Face downloads require network access and your HF token. Slack, Jira, and cloud LLM providers are optional. Read Security & privacy before deploying on shared lab machines.
Hardware notes
OpenBio 8B + qwen2.5:7b fit comfortably on 16 GB RAM machines; see Hardware requirements. Multi-agent chat shares one Ollama backend — you do not load six separate 14B models.