FAQ
Deployment
Q: What's the minimum hardware required?
A 4C8G CPU server is enough to run the full stack to start (Milvus has a high memory footprint, so 8G+ RAM is recommended). MinerU GPU parsing additionally needs a GPU with 8G+ VRAM.
Q: Are offline / intranet environments supported?
Yes. Every service can be self-hosted, images can be imported offline, models can be replaced with open-source models deployed on your intranet, and nothing depends on external network access.
Q: How do I upgrade?
Pull the new version's images, run database migrations, then restart the services (docker compose exec server ./ollmo migrate, or make migrate locally). For production, back up MySQL and MinIO data before upgrading.
Q: Are HTTPS and custom domains supported?
Yes. Put an Nginx or Caddy reverse proxy in front of Web / API and configure the certificate, then update CORS_ORIGINS in .env.
Models & retrieval
Q: Do I have to use SiliconFlow?
No. Any OpenAI-compatible endpoint can be connected, and self-hosted inference services like Ollama and vLLM are also supported. The install wizard simply provides a ready-to-use default path.
Q: What if retrieval quality is poor?
- Check whether a Rerank model is configured (it has the largest impact on accuracy);
- On the "Retrieval testing" page, compare the balance of keyword and semantic weights;
- Adjust the chunk size (use parent-child chunking for complex documents);
- Add curated Q&A for high-frequency questions — exact matches return the standard answer.
Q: What file formats are supported?
All common formats: PDF, DOCX, Markdown, TXT, and more. MinerU parsing handles complex layouts (tables, formulas, scanned documents) even better.
Q: Will the answers hallucinate?
ollmo's answers are based on retrieved chunks and carry citations. When retrieval confidence is insufficient, the standard Agent template routes to the "direct reply" branch and honestly reports that no content was found, rather than fabricating.
Data & security
Q: Where is my document data stored?
Entirely on your servers: MySQL (metadata), Milvus (vectors), MinIO (raw files and parsing artifacts) — fully under your control.
Q: Is data isolated between tenants?
Yes. Every data table carries tenant_id, and MySQL, Milvus, and MinIO are logically isolated end to end, so teams cannot see each other's data.
Q: Is there auditing and usage management?
Yes. The admin console provides audit logs, usage and cost statistics, and team quota management (Free / Pro / Enterprise).
Licensing & support
Q: Can I use it commercially?
Yes. Internal enterprise use, commercial application development, and on-premises delivery are free; operating a multi-tenant SaaS similar to ollmo requires a commercial license. See License.
Q: Where do I report issues?
GitHub Issues. For business and licensing inquiries: contact@ollmo.com.