v2.5.1 v2.5.2
Rouf0x/splatfpvMay 26, 2026 → May 26, 20260 days between releases
What's New
v2.5.2 fixes iOS detection and dashboard DNS, but is deprecated for a missing image tag and security advisories
Breaking Changes
- The release pins a hermes-agent image tag that no longer exists, so fresh installs and upgrades fail.
- The release is affected by security advisories GHSA-vqpg-pvjj-4cmq and GHSA-v3hj-52g2-2hmc.
- LLM profile selection now prefers the minitap default over OpenAI config, which can change behavior for existing OpenAI-configured deployments.
New Features
- The minitap default LLM profile is now preferred over OpenAI configuration.
Improvements
- Discrete NVIDIA GPUs with less than 4GB VRAM now default to the CPU/Tier 0 fallback instead of entering a crash-looping CUDA install.
Bug Fixes
- iOS device detection now checks for Darwin instead of Linux.
- Dashboard nginx re-resolves the dashboard-api service through Docker DNS at request time, preventing stale container IPs after lifecycle recreation.
Upgrade Notes
Do not install or upgrade to v2.5.2 as published: its pinned hermes-agent image tag is missing and it carries known security advisories. Install from the README instead. If testing the release, verify LLM profile precedence because minitap defaults now override OpenAI config.
The Two Releases
Dream Server 2.5.1 - Fresh Install & Bootstrap Recovery
May 26, 2026· 5 months ago
Dream Server 2.5.1 is the fresh-install and bootstrap-recovery patch release validated after the 2.5.0 audit hardening cycle. It improves bootstrap full-model download resume/retry behavior, extends Hermes timeouts for slow local backends, includes Dream Talk owner-portal work, hardens lifecycle and reinstall paths, and expands operator documentation.
Full release notesDream Server 2.5.2 - Lifecycle DNS & Tiny GPU Fallback
May 26, 2026· 5 months ago
Dream Server 2.5.2 fixes lifecycle DNS behavior by having dashboard nginx re-resolve dashboard-api through Docker DNS at request time, preventing stale container IPs after lifecycle recreation. It also routes discrete NVIDIA GPUs with less than 4GB VRAM to the CPU/Tier 0 fallback by default to avoid crash-looping CUDA llama-server installs. The release is now deprecated due to a missing hermes-agent image tag and security advisories.
Full release notes