ODS, formerly DreamServer, v2.6.0 is an installation and control layer for a local AI server. It selects an inference path for the detected hardware and combines Open WebUI, LiteLLM, agents, workflows, RAG, voice, image, and observability services.
Source: github.com/Osmantic/ODS/tree/v2.6.0 View on GitHub →The installer selects an inference backend, GGUF model, and context profile from the machine platform, GPU, and memory tier. It then connects inference, Open WebUI, LiteLLM, n8n, agents, RAG, speech, image, and observability components behind one control path.
Source: github.com/Osmantic/ODS/tree/v2.6.0The installer selects an inference backend, GGUF file, and context configuration from the platform, GPU, and memory tier.
Source: github.com/Osmantic/ODS/tree/v2.6.0It connects inference, Open WebUI, LiteLLM, n8n, agents, RAG, voice, image, and observability components.
Source: github.com/Osmantic/ODS/tree/v2.6.0Local mode is the default. Users can explicitly switch to cloud or hybrid mode and configure an external provider.
Source: github.com/Osmantic/ODS/tree/v2.6.0The platform installer creates the.env file and credentials, while control scripts manage containers and native inference services. Open WebUI reaches local or cloud models through LiteLLM and the selected route. Each extension service retains its own network and credential boundary.
Source: github.com/Osmantic/ODS/tree/v2.6.0DockerWindows WSL2, Apple Silicon, or a su…Disk space for models and imagesODS suits users who want to install and maintain inference, a WebUI, workflows, and agents together on one workstation. A narrower tool is a better fit when the only need is document chat, container orchestration already exists, or the user does not want the upgrade surface of a multi-service stack.
Source: github.com/Osmantic/ODS；https://github.com/Mintplex-Labs/anything-llmv2.6.0 was published on 2026-07-28. Its changes include remote providers, model switching, rollback for GPU reassignment, rootless fixes, native Windows and macOS paths, and DNS rebinding and SSRF hardening.
Source: github.com/Osmantic/ODS/releases/tag/v2.6.0status: docs-only | checked_at: 2026-09-23 | environment: Docs only. Performed a static review of the v2.6.0 release, README, Installer Trust notes, and platform and port documentation. | limitations: Installation, image pulls, GPU inference startup, and data-volume behavior were not tested. | observed_output: Official documentation defines the status command, WebUI port 3000, platform inference ports, and green validation across six hosts. It also states that the long-running six-round model matrix is exempt.
Source: github.com/Osmantic/ODS/releases/tag/v2.6.0AnythingLLM is an end-user application centered on documents, workspaces, chat, and agents. ODS is a broader local AI server stack and control layer that also orchestrates inference, workflows, speech, images, and operations components.
Source: github.com/Mintplex-Labs/anything-llm；https://github.com/Osmantic/ODS…The project is now published as ODS, so older DreamServer instructions may point to outdated paths. For a long-lived machine, pin v2.6.0 and review the installer. Verify the first chat, restart recovery, and actual routing before enabling cloud services or third-party integrations.
Source: github.com/Osmantic/ODS/tree/v2.6.0Yes. The old repository now redirects to Osmantic/ODS, and the current project name, documentation, and commands use ODS. New installations should pin a stable ODS tag.
Docker must be installed and running. Windows requires the WSL2 backend in Docker Desktop; macOS supports Apple Silicon; Linux GPU and distribution support must be checked against the official matrix.
Local mode is the default, and cloud access is not required. Requests leave the machine through LiteLLM only after cloud or hybrid mode and an external provider are configured.
Run the platform-specific status command, open localhost:3000 for a short chat, and check the inference port.