Most self-hosting guides amount to gluing together half a dozen projects, matching ports, and wrestling drivers. Dream Server takes the opposite approach: package a full local-first AI stack into an opinionated, single-command installer so you can run inference, chat, voice, agents, RAG, and image generation on your own hardware without stitching components manually.
What Sets It Apart
- Full-stack, not just inference: includes a web chat UI, llama-server inference gateway, embeddings and Qdrant for RAG, ComfyUI for image generation, Whisper/Kokoro for STT/TTS, agent frameworks (Hermes/OpenClaw), n8n workflows, and observability tooling — so you get an end-to-end stack that interoperates out of the box. This means less integration work and fewer runtime surprises when combining features like agents + RAG.
- One-command bootstrap with hardware auto-detection: the installer detects GPU/CPU capabilities, selects an appropriate model tier, and starts a tiny 1.5B bootstrap model for immediate use while larger GGUF models download in the background. Practically, you can be chatting in minutes even if the final model takes hours to fetch.
- Local-first with optional cloud/hybrid fallback: designed to run entirely on-device (no subscriptions or external telemetry), but offers cloud or hybrid modes that route to provider APIs if you choose. That gives a clear path from fully sovereign setups to pragmatic hybrid deployments.
- Extension-first architecture and safety tooling: services are modular extensions (manifest+compose fragments) and the project includes privacy-proxy tooling, token usage monitoring, and an Agent Policy Engine for auditing autonomous tool calls — useful for teams that need governance around local agents.
Who it's for — tradeoffs
Great fit if you want a single, pre-wired self-hosted AI stack for experimentation, privacy, or edge deployment and you have at least moderate hardware (GPU recommended). It's also suited for home labs and research teams who value local inference, modular extensions, and an integrated dashboard. Look elsewhere if you need a minimal inference-only binary, strict enterprise support SLAs, or want a fully managed cloud solution without any host maintenance. Self-hosting still requires disk space, model downloads, occasional updates, and basic ops knowledge (Docker/ports/drive space); very small or ephemeral devices may rely on cloud mode instead.
Where it fits
Dream Server targets users who want the full workflow (chat, RAG, agents, TTS/STT, image gen) without manually wiring multiple projects. Compared to single-purpose tools (llama-server, LocalAI, Open WebUI), Dream Server trades minimal footprint for an integrated, opinionated experience that simplifies multi-service interactions.
