display.dev for vLLM
vLLM serves the model on your GPUs. display.dev is where its output lands – at a permanent URL, behind your company login when you want it. vLLM gives you high-throughput, OpenAI-compatible inference on your own infrastructure. What it doesn't give you is a hosted place to put the HTML and Markdown your jobs produce, where the right people can open it without an account. display.dev is that place. One command, permanent URL, gated by your company's Google or Microsoft login.
vLLM serves inference. It doesn't host output.
You stood up vLLM for a reason: PagedAttention throughput, an OpenAI-compatible endpoint, models running on GPUs you control instead of a vendor API. Llama, Qwen, DeepSeek, gpt-oss – whatever you serve, it stays on your infrastructure.
That stack ends at the completion. When a batch job or a serving endpoint produces a report, a dashboard, or a research brief, vLLM's job is done and the output is a file on a worker. There's no hosted way to hand that file to a colleague who has no account anywhere – because there's no vendor in the loop.
display.dev is the publish layer that vLLM doesn't include, and isn't trying to.
From a vLLM job to a company URL
Publish output from a batch job:
python run_batch.py --endpoint http://vllm:8000/v1 --out ./report.html
dsp publish ./report.html --name "batch-report-$(date +%Y-%m-%d)"Publish as an agent tool call (MCP):
{
"name": "publish",
"arguments": { "content": "<html>...", "name": "vllm-output", "visibility": "company" }
}Publish a whole output directory:
dsp publish ./output/ # serves index.html at the root, resolves relative assetsThe URL is permanent and authenticated. Paste it anywhere. The viewer signs in once with their company email and sees the full rendered result.
Viewers sign in with their work email. Nothing else.
Self-hosting on vLLM means there's no vendor with a "share with my org" feature – which is the point of self-hosting. It also means the colleague who needs to read the output has no account to authenticate against.
display.dev is vendor-independent. Viewers sign in with the Google or Microsoft email they already use for work. No display.dev account, no model account, nothing to install. They click the link and read the page.
The best model changes. The home doesn't.
You'll swap the model behind your vLLM endpoint as the open-weight field moves. display.dev sits above the serving layer – change the served model, the engine, even the whole stack, and everything you published keeps its URL, versions, and comment threads.
Pricing
Try it. No credit card.
- Included: 10 gated artifacts
- Included: Unlimited public artifacts*
- Included: 50MB storage
- Not included: Version history
- Not included: Google + Microsoft SSO
- Not included: Private artifacts
- Not included: Audit logs
display.dev branded
Individuals. Unlimited gated sharing.
- Included: Unlimited gated artifacts
- Included: Unlimited public artifacts
- Included: 1GB storage
- Included: 10 versions
- Not included: Google + Microsoft SSO
- Not included: Private artifacts
- Not included: Audit logs
For teams that need company auth.
- Included: Unlimited gated artifacts
- Included: Unlimited public artifacts
- Included: 25GB storage
- Included: 50 versions
- Included: Google + Microsoft SSO
- Included: Private artifacts
- Included: 90 days audit logs
For compliance-driven organizations.
From €499/mo
- Included: Unlimited gated artifacts
- Included: Unlimited public artifacts
- Included: Custom storage
- Included: Unlimited versions
- Included: Google + Microsoft SSO
- Included: Private artifacts
- Included: 365 days audit logs
Features
Viewers click a link and sign in with their Google or Microsoft account, or a one-time password. No app to install. No account to create. No IT ticket.
Inline comments on every artifact. Your agent reads them via MCP, fetches the current version, republishes with short_id and base_version, and resolves the thread. The artifact stays a living document, not a one-shot screenshot.
dsp publish ./file.html from your terminal, or one MCP tool call. No git repo, no deploy pipeline, no project to configure.
Every artifact gets a URL that keeps working. Share in Slack, link in Notion, paste in email. It still works six months later.
No per-seat pricing at any tier. Share with your PM, exec, legal team, or designer for the same flat price.
Whatever vLLM is serving – plus Claude Code, Cursor, Codex, or your own scripts.
Common questions.
Does this expose my vLLM endpoint?
No. display.dev hosts only the file you explicitly publish. Your vLLM server, GPUs, and weights stay entirely on your own infrastructure. You're publishing the output, not the endpoint.
Do viewers need access to my infrastructure or an AI account?
No. Viewers sign in with their company Google or Microsoft email. No infrastructure access, no model account, no display.dev account on their side.
Can a batch job publish automatically?
Yes. Add dsp publish at the end of the job, or call display.dev's MCP/REST API from the job so it publishes its own output and returns the URL.
Can I publish a multi-file output directory?
Yes. dsp publish ./output/ serves index.html at the root URL and resolves all relative asset paths.
What happens when I change the served model?
Nothing moves. display.dev is independent of what produced the output, so every URL, version, and comment thread stays put.
Publish your first artifact in 15 seconds.
Free tier. No credit card. One-time password auth on free, Google + Microsoft SSO on Pro (€49/month flat).