FAQ

Deployment & environment

Docker deployment fails?

1. Confirm Docker ≥ 20.10 and Docker Compose ≥ 2.0: docker --version && docker compose version
2. Check that .env.runtime.secrets is configured correctly (DB_PASSWORD, JWT_SECRET_KEY, at least one LLM API key)
3. Inspect container logs: docker compose logs --tail=100
4. On port conflicts (30080 / 38000), edit the port mappings in docker-compose.yml

Can I use it without a GPU?

Yes. The bioinformatics tools don’t depend on a GPU; LLM calls go through cloud APIs (Zhipu / DeepSeek), so no local GPU is needed.

Which operating systems are supported?

Docker deployment supports Linux, macOS, and Windows (WSL2 recommended). Native Windows (non-WSL) is untested.

Login & accounts

What is the default account on first login?

The system creates an administrator account automatically during deployment. Your first visit to the Web UI (http://localhost:30080) guides you through setting the password. See the deployment log for the initial account details.

Forgot your password?

Reset it on the server:

docker compose exec app python manage.py reset-password <username>

Network & proxy

The main platform is slow to access from China?

Use the China mainland accelerator: https://www-ai.u1289916.nyat.app:26572 (via NATFRP tunnel). Both entries point to the same platform.

Not reachable from the public internet after deployment?

1. Make sure the firewall allows ports 30080 and 38000
2. On cloud servers, check the security group rules
3. For HTTPS, put Cloudflare in front or configure your own Caddy/Nginx reverse proxy

Usage paths & entries

How do Manual Submission and Agent-Assisted differ?

In Manual Submission you pick a tool from the catalog, fill the form, and submit the task. In Agent-Assisted you describe your need in natural language in the Agent Workbench, and the Agent selects tools and assembles parameters through MCP under confirmation gates. Both call the same tool and task system. See Usage paths.

Can GWAS be submitted directly from a form like RNA-seq?

Yes. GWAS is now a one-page form in the tool catalog: pick the genotype (VCF or PLINK BED) and phenotype table, choose the trait column from a dropdown and the reference version (B73 v4 / v5 / Others), pass the pre-submission intent review, and submit once — the five models (GEMMA LMM/LM + rMVP GLM/MLM/FarmCPU) all return results in one run, with SnpEff annotation, publication-grade plots, and a self-contained HTML report delivered. See GWAS end-to-end.

How do I enter RStudio, and how does it relate to manual analysis?

Enter through the platform’s RStudio entry to open browser-based RStudio Server, and log in with your platform account. It shares the same workspace files, tasks, and artifacts as manual analysis: the platform data you process in RStudio can be referenced by platform tasks, and the Agent can help maintain and run R scripts under control (workspace file scope). See RStudio analysis.

How do native workflows differ from regular tools?

A regular tool is a single step (e.g. one DE analysis) and you prepare its inputs yourself; a native workflow is a multi-step chain (e.g. RNA-seq’s QC → alignment → quantification → DE → enrichment) that runs all stages in order from one submission, with intermediate files connected automatically, and shows up on the Tasks page as a task chain. See Native workflows.

MCP & Agent

What is MCP?

Model Context Protocol (MCP) is the standard protocol for the Agent to call tools. On this platform capabilities are exposed to the Agent as MCP tools — 56 statically expanded (48 read-only / 7 controlled write / 1 workspace file), dynamic at runtime by session visibility (per the session’s tools/list), under tiered permissions. See MCP governance.

What is the MCP official channel?

It’s the official-protocol access channel the platform provides for MCP clients, at /api/mcp-official/*; the earlier /api/mcp/* remains as the compat channel. Both expose the same governed tool set — choose based on client compatibility. The Agent built into the workbench is pre-configured; no manual wiring needed.

What are platform skills, and how do they trigger?

Skills are structured playbooks pre-installed for the Agent (when to use them, which steps to follow, what output to produce); the Agent Workbench ships 9: tool selection (platform-capability-planner), submission (mcp-task-lifecycle), reading results (task-results-reader), RNA-seq orchestration (bulk-rnaseq-analysis), gene identity verification (gene-identity-gate), RStudio runtime (rstudio-analysis-runtime), plus 3 paper-polishing skills. You never invoke them manually — the Agent matches and triggers them from your request. See Agent-assisted mode.

Could the Agent submit tasks recklessly?

No. For write operations (L2) the Agent must first show you the exact call and parameters in the conversation and only submits after you confirm; workspace file operations follow workspace boundaries and session scope; and every call is audited.

Data & files

What is the single-file upload limit?

100 GB. The platform uses chunked upload, so large transfers are reliable.

Which file formats are supported?

All common bioinformatics formats: FASTA, FASTQ, BAM/SAM, BED, GFF/GTF, VCF, CSV/TSV, TXT, and more. Plotting tools also accept PNG/SVG/PDF input.

Does database query support natural language?

Yes. On the platform’s Databases page, describe the query in plain language (e.g. “look up this gene’s family in PlantTFDB and its motifs in JASPAR”) and the platform parses it and searches across databases; results can be saved as query records or file artifacts and referenced as file inputs by later tasks. See Databases & regulatory resources.

Where is data stored?

Uploaded files and task artifacts are stored on the platform server (Docker volumes): on local disk for local deployments, on cloud storage for cloud deployments.

LLM configuration

Which LLMs are supported?

Configure them in .env.runtime.secrets; at least one is required.

Agent LLM call errors?

1. Confirm the API key is valid and has sufficient balance
2. Check that the network can reach the API endpoint (OpenAI may need a proxy from mainland China servers)
3. Check the Agent Workbench error log — it usually shows the specific HTTP status code

Other

What is the relationship between this site and the main platform?

This site (us.omicsagent.xyz) is a purely static introduction/documentation site; it doesn’t host the main platform. The main platform entry is omicsagent.xyz.

Is the code available?

The code is currently provided for research use; a public release is in preparation, and licensing and access will be announced when it lands.

How do I report issues or suggestions?

Use the feedback entry inside the main platform.