Dedicated GPU servers built for speed and reliability.
Miristream is the dedicated hardware that runs every Miricam platform — dual 3U rackmount nodes with NVIDIA RTX 5080 GPUs, on-site AI inference, sub-second WebRTC streaming, and active/standby failover. Included with Miricam. Also available for the small number of bespoke projects we build with clients. Not offered as general hosting.
GPU at every layer that needs it. No round-trips to anyone else's data centre.
The platform feels fast because the hardware is local, the GPU is dedicated, and the AI is on-site. There is no shared cloud queue, no API rate limit, and no transatlantic round-trip in front of every request.
Hardware video encode (NVENC)
Transcoding runs on the RTX 5080's dedicated encode blocks, not the CPU. The hardware can sustain multiple simultaneous 4K encodes in the time CPU-only servers would need for one. The CPU stays free for platform logic, billing, and session work.
WebRTC, terminated server-side
Miristream terminates WebRTC connections on the server rather than relying on browser-to-browser peer connections. Streams stay stable as rooms fill up, and routing variability from direct peer paths is removed entirely.
On-site LLM inference
Models stay loaded on the GPUs, so requests do not pay a cold-start cost. Responses to AI features — ID verification, profile writing, AutoSEO, data queries — typically complete in 3–8 seconds rather than the 20–60 seconds that CPU-only or cloud-routed inference can take.
Local network access to your data
The AI reads directly from your platform's database over the local network. No exports, no ETL, no data shipped to an analytics warehouse. Live data, queried in place, returned in seconds.
Two physical nodes, monitored continuously, with no third-party dependencies in the critical path.
Reliability comes from straightforward decisions: dedicated hardware we own, a warm standby always ready, and AI that keeps running regardless of what is happening at someone else's data centre.
Active/standby dual-node
Two physical servers run in tandem. If the active node has a hardware issue, traffic shifts to the warm standby — no cold-start delay, no complex multi-region routing. Straightforward redundancy that actually works.
No external API dependency
AI features run on hardware we own. When OpenAI, Anthropic, or another cloud AI provider has an outage — and they all do, regularly — your platform keeps working. The AI does not stop because someone else's service stopped.
Daily backups, offsite copy
Your database is backed up every 24 hours with a copy stored offsite. If something goes wrong, recovery is to the previous day's state — not a long, fragile reconstruction.
Continuous uptime monitoring
Every layer of the stack — platform, streaming, GPU/AI — is monitored continuously. If something stops responding, we find out immediately and start fixing it. You do not need to open a ticket to tell us first.
Isolated AI layer
The AI workloads are isolated from the core platform. If the AI server has a problem, member access, live video, and billing keep running. AI features are enhancements, not a dependency that takes the whole site down.
We control model versions
Cloud AI providers retire and update models on their own schedule. On our hardware, model versions change when we decide they should — so the AI features on your platform behave the same way month to month.
Three workloads, kept separate.
Running a web application, streaming live video, and doing AI inference at the same time on a single shared server is a recipe for everything performing poorly. We keep these workloads separated so none of them slow the others down.
Web, Database & Admin
- Web application and PHP processing
- MySQL — members, sessions, billing, analytics
- Operator admin panel and dashboards
- API endpoints and cron automation
- SSL certificates and DNS management
- Daily database backups with offsite copy
Miristream Video Pipeline
- WebRTC signal handling and stream distribution
- RTMP & RTSP ingest — OBS, IP cameras, encoders
- HLS and DASH output for broad device support
- GPU transcoding — H.264, H.265, AV1
- Adaptive bitrate (5 quality rungs from one stream)
- Server-side recording and VOD delivery
- Token-based stream authentication
On-Site AI Inference
- Dual NVIDIA RTX 5080 GPUs
- LLMs running entirely on-site
- No cloud API calls — your data stays local
- Concurrent inference across platform features
- Typical completions in 3–8 seconds
- GPU capacity expandable as you grow
The AI features the on-site GPUs power.
Beyond video encoding, the same GPUs run inference for the platform's AI features. None of these depend on a third-party AI service, and none of them add a usage charge to your bill.
Photo ID Age Verification
When performers or members submit a government-issued ID, the document is analysed on your own servers by the on-site LLM. It never gets transmitted to an external AI service and never touches a third-party database. For an adult site operating under age verification regulations, that is a meaningful difference.
Performer Profile Writing
Performers fill in a short form, and the AI turns it into a proper profile — readable, engaging, and consistent with the tone of your platform. Blank or thin profiles hurt conversion; this keeps every performer's page working from day one. Operators review and publish with one click.
AutoSEO
AutoSEO writes optimised meta titles and descriptions across performer profiles, category pages, and site content. It pulls Google Search Console data to focus on keywords that are already getting impressions — the ones closest to actually ranking. It runs on a schedule, not as a one-time audit.
Chat with Your Data
Administrators can ask questions about their platform in plain English and get real answers from the live data. Which performers are bringing in the most revenue, where members drop off in the funnel, what chat patterns correlate with paid bookings. The AI reads directly from the database — no exports, no analyst, no dashboard required.
Dual 3U servers, NVIDIA RTX 5080 GPUs, 10 GbE.
Two physical machines, both GPU-equipped, both managed by us. One handles active ingest and distribution; the second provides failover and runs AI inference workloads so transcoding throughput stays unaffected. Both carry NVIDIA RTX 5080 GPUs with dedicated NVENC and NVDEC blocks.
For Miricam operators this is transparent — the platform handles it. But if you want to know what is actually running the video and AI side of your business, this is it.
| Component | Specification |
|---|---|
| Form factor | Dual 3U rackmount servers |
| GPU | NVIDIA GeForce RTX 5080 (per node) |
| GPU memory | 16 GB GDDR7 (per node) |
| Video encode | Hardware NVENC — H.264, H.265, AV1 |
| Video decode | Hardware NVDEC — H.264, H.265, AV1, VP9 |
| Network | 10 GbE uplink |
| Redundancy | Active/standby node pair |
| Management | Fully managed by 2MUCH.NET |
This is not a general hosting service.
We do not sell shared hosting, VPS rentals, cPanel accounts, or cloud resale. Miristream exists to run Miricam platforms with the speed, security, and reliability we promise — and that is the standard we want to keep. Spreading the same hardware across general hosting customers would compromise both, so we do not.
Available for bespoke projects. If you have a specialised application — moderation tooling, age verification, video, a custom AI workload, something we would build with you from scratch — and you want it running on hardware where the AI is on-site and the data is not being shipped to third parties, get in touch. We take on a small number of these projects per year. We will tell you honestly whether yours is a fit.
Questions operators tend to ask.
Can I rent Miristream as general hosting?
No. We do not offer shared hosting, VPS rentals, or cloud resale. Miristream runs Miricam platforms and a small number of bespoke projects we build with clients. If you need general hosting, we are not the right fit and will say so up front.
Do I pay extra for Miristream?
No. It is part of every Miricam plan — no separate streaming server charge, no per-stream fee, no cloud egress billing. The hardware is provisioned and managed by 2MUCH.NET.
What is the latency for viewers?
WebRTC streams typically deliver 200–500 ms glass-to-glass under normal conditions — sub-second in practice. HLS and DASH outputs are also available for broad device compatibility but add 6–15 seconds by design. Interactive sessions use WebRTC.
Why on-site AI instead of cloud APIs?
Speed, cost, reliability, and data security. On-site GPU inference completes in 3–8 seconds with no API queue and no per-token billing. Sensitive material — IDs, chat content, billing data — stays on your infrastructure rather than going to a third-party AI provider. Your AI features keep working when external APIs go down.
Are the AI models always loaded?
Yes. Models are kept loaded on the GPU hardware, so there is no cold-start delay when a feature is triggered. Responses begin generating within milliseconds rather than waiting for a model to spin up from scratch.
Does it support OBS Studio?
Yes. RTMP ingest is the standard protocol OBS Studio, Streamlabs, and most hardware encoders use. Performers on Enterprise plans can broadcast professional setups directly to the Miristream endpoint. Standard performers on all plans use browser-based WebRTC.
Can sessions be recorded server-side?
Yes, on Enterprise plans. Recordings happen at the server as MP4 files independent of the performer's device. Useful for replay subscriptions, compliance archiving, and operator review.
What happens if the AI server has a problem?
The AI layer is monitored alongside everything else and is isolated from the core platform. If it has a problem, member access, live video, and billing keep running. AI features are enhancements, not a dependency.
Is the infrastructure shared between Miricam operators?
No. Each Miricam platform runs on its own dedicated infrastructure. Stream endpoints, recordings, AI workloads, and database access are completely isolated per platform.
What does it cost to expand GPU capacity?
GPU upgrades are priced at cost and added to your monthly subscription. We quote before making any changes and let you know in advance when you are getting close to the limits of your current setup, rather than waiting for things to slow down.
Real GPU hardware. Real on-site AI. One monthly price.
Dedicated servers, on-site AI inference, Miristream video, bandwidth, SSL, daily backups, uptime monitoring, and support from the team that built it — all included with your Miricam subscription. Or talk to us about a bespoke project.