Getting Started: Domain Scientists¶

ADEPT provides a natural language interface for scientific workflows. Biologists, chemists, physicists, and data scientists can interact with 28+ specialized tools through conversational chat — no programming required.

Prerequisites¶

A running ADEPT instance (local Docker Compose or cloud deployment)
Valid user credentials (provided by your system administrator)
A web browser

Connecting to ADEPT¶

ADEPT exposes two primary interfaces for domain scientists:

Interface	Best For
OpenWebUI	General chat, file uploads, multi-turn conversations
JupyterLab	Notebook-style workflows, inline visualizations

Your administrator will provide the URL and login credentials. Authentication uses single sign-on via Keycloak.

Example Workflows¶

Protein Sequence Search (BLAST)¶

User: Run a BLAST search for this sequence against the nr database:
      MKTAYIAKQRQISFVKSHFSRQLEERLGLIEVQAPILSRVGDGTQDNLSGAEKAVQVKVK

Agent: I'll run a BLAST search against the NCBI nr database...
       [Results: top 10 hits with E-values, identity percentages, and organism names]

Literature RAG Query¶

Upload a collection of PDFs, then query them conversationally:

User: Upload these 5 papers on CRISPR-Cas9 delivery mechanisms.
Agent: Files uploaded and indexed. Ready for queries.

User: What delivery vectors show the highest in vivo efficiency across these papers?
Agent: Based on the uploaded literature, lipid nanoparticles (LNPs) show the
       highest in vivo efficiency at 70-90% editing rates...

Sandboxed Code Execution¶

User: Calculate the binding free energy using the MM-PBSA method for this
      protein-ligand complex. Here's the trajectory data.

Agent: I'll write and execute a Python script using MDAnalysis...
       [Executes code in isolated sandbox, returns results and plots]

Available Tools¶

Category	Tools	Description
Bioinformatics	BLAST, UniProt, AlphaFold	Sequence search, protein data, structure prediction
Chemistry	PubChem	Compound properties, similarity search
Literature	ChromaDB RAG	Upload and query PDFs, CSVs, documents
Code Execution	Sandbox	Run Python/R in isolated containers
Web	Web Search	Search and summarize web content
Data	CSV Processing, File Management	Upload, transform, and analyze tabular data
HPC	Nextflow Pipelines	Submit and monitor HPC workflows

Tool Discovery

Ask the agent "What tools do you have available?" to see the full list of tools accessible in your current session.

Multi-Agent Teams¶

For complex problems, ADEPT can assemble teams of specialized agents:

User: Create a team with a biologist and a data scientist to analyze this
      proteomics dataset and identify differentially expressed proteins.

Agent: Creating multi-agent session with roles: biologist, data_scientist...
       [Biologist agent interprets biological significance]
       [Data scientist agent handles statistical analysis]
       [Results synthesized into a unified response]

Each agent receives domain-specific instructions and can use the full tool suite.

File Upload and Processing¶

ADEPT supports uploading and processing multiple file types:

CSV/TSV: Automatic schema detection, RAG indexing, SQL querying
PDF: Text extraction, chunking, vector embedding for RAG
Code files: Syntax-aware processing
Images: Inline display in responses

Batch Processing

Upload multiple files at once for batch indexing. Use "Create a RAG index from my uploaded files" to build a searchable knowledge base.

Session Continuity¶

ADEPT maintains conversation context across long sessions:

Auto-compaction: Long conversations are automatically summarized to stay within context limits
Thread persistence: Return to previous conversations by thread ID
Session isolation: Your files and data are isolated from other users

Tips¶

Be specific about databases and parameters when requesting searches (e.g., "BLAST against SwissProt" vs "BLAST against nr")
Upload reference documents before asking questions about them
For reproducibility, ask the agent to show its code before executing
Use multi-agent teams for interdisciplinary problems that benefit from multiple perspectives
Large file uploads are processed asynchronously; the agent will notify you when indexing completes