Telegram & scheduled jobs
Human requests and automated collection enter through narrowly scoped interfaces.
Private compute.
Public proof.
A self-hosted AI platform that runs local inference, Telegram summaries, autonomous news analysis, and a dedicated coding model without exposing the compute plane to the public internet.
Loading the latest scheduled status snapshot…
How the platform moves work
Interfaces and agents remain separate from the shared inference layer, keeping each component replaceable.
Purpose-built agents
Application code handles workflow, storage, validation, and deterministic decisions.
Local model servers
Shared llama.cpp inference stays private and uses dedicated production and coding models.
Sanitized outputs
Only selected static artifacts leave the platform for public viewing.
Platform signals
These are delayed, sanitized states from the scheduled publication cycle—not a live control interface.
Resource snapshot
Utilization is deliberately rounded. Detailed host telemetry remains private.
CPU utilization
—
Memory utilization
—
AI storage utilization
—
Current uptime
—
Engineering priorities
No public model endpoint, no inbound home firewall rule, and no administrative controls in this showcase.
Existing scheduled work produces this page. There is no additional monitoring stack, database, or persistent collector.
Applications are separated from inference, scheduled work is locked against overlap, and public output remains available if the home platform is offline.
The dashboard is public. The system is not.
This page receives a small static snapshot during the existing publication cycle. A visitor never connects to the home server, model runtime, message store, or administration layer.
✓Inference and application data remain on private infrastructure.
✓No messages, identifiers, logs, addresses, or service commands are published.
✓Status values are rounded and delayed by design.