Open WebUI
π Quick overview
Open WebUI is a self-hosted web platform for exposing local models via Ollama, vLLM, or any OpenAI-compatible API. The project presents itself as extensible, feature-rich, and capable of running entirely offline12.
π‘ Why this project interests us
Open WebUI occupies the βinternal ChatGPT portalβ place in an on-premise stack. It does not replace the inference engine: it orchestrates model access, interface, users, files, and plugins.
In this vault, it is the reference solution for the simple multi-user scenario: a local backend, a web interface, permissions, easy adoption.
β Strengths
- Self-hosted: Docker, Kubernetes, Python, images with Ollama or CUDA as needed1.
- Provider-agnostic: Ollama, OpenAI-compatible APIs, vLLM, and other backends2.
- Mature user experience: history, files, RAG, plugins, multiple models.
- Credible local deployment: can be linked to Ollama via
OLLAMA_BASE_URL4. - Controllable observability: OpenTelemetry available for your own traces/logs in production4.
- Recent team features: versions 0.10 (June 2026) and 0.11 (July 2026) add shared folders with per-group permissions, external knowledge bases, automatic context compaction, sub-agents, and LDAP group synchronization: features that bring Open WebUI closer to an enterprise portal, and just as many surfaces to administer5.
β οΈ Limitations and risks
- Not an inference engine: Ollama/vLLM must be sized separately.
- Administration surface: accounts, plugins, CORS, secrets, and network exposure must be hardened.
- Cloud provider possible: if connected to OpenAI/Anthropic, data follows the chosen provider.
- Telemetry/analytics to verify: the official Docker image sets
SCARF_NO_ANALYTICS=true,DO_NOT_TRACK=true, andANONYMIZED_TELEMETRY=falseby default; for a pip install or a rebuilt image, set them explicitly and enableOFFLINE_MODE=truein a strict context46.
π Sovereignty and privacy
- Data: stored in the self-hosted instance.
- Model: local if
OLLAMA_BASE_URL/ local vLLM; cloud if external provider configured. - Memory: history and RAG in the instance.
- Telemetry: disable recommended via environment variables4.
- 100% offline mode: yes if images, models, and dependencies are preloaded;
OFFLINE_MODE=trueandHF_HUB_OFFLINE=1cut downloads and network checks4. - Verdict: β οΈ configurable β excellent on-prem if hardened, but multi-provider by nature.
See the full grid: Sovereignty & Privacy.
π Possible integration in this vault
Open WebUI is a good companion to the blueprints:
- Scenario A: local personal UI in front of Ollama.
- Scenario B: SME portal for a few users.
- Scenario D: access front on vLLM/TensorRT-LLM behind proxy.
π Project maturity
Very widely used and actively maintained project (about 154,000 GitHub stars, v0.13.4 as of 2026-11-17), with a large GitHub community and plugin ecosystem. Product maturity is good, but the attack surface follows: 88 security advisories published between June and September 2026, including about thirty of high severity (account takeover via OAuth, SSRF toward internal services, cross-user tool execution) and three exploitable without an account. Fixes ship only in current releases: an SME must follow the release train without ever staying below 0.13.1 (2026-11-08). Like 0.12.4 and 0.12.5, it ships security fixes whose advisories the vendor will only publish once CVEs are assigned. The vendor recommends going straight to 0.13.4 (2026-11-17), after a database backup: 0.13.1 changes the schema, which requires updating all instances at the same time. 0.13.4 also removes Direct Connections, that is, the models and tool servers that each user added in their settings. They now go through the Tether desktop app, free for personal non-commercial use but subject to an Enterprise license in a business setting7. It must not pin a version and must apply the official hardening guide8.
π See also
- Ollama Β· vLLM β inference backends
- π§© RAG & Agents Β· βοΈ Inference engines
- Scenario B Β· Scenario D
- AnythingLLM Β· Khoj β RAG/UI alternatives
- π Inference security Β· π§ͺ Evaluate a model
π Sources
Footnotes
-
Open WebUI GitHub β self-hosted offline platform, Ollama support, and Docker/Kubernetes images. https://github.com/open-webui/open-webui β© β©2
-
Open WebUI Docs β home, providers, and features. https://docs.openwebui.com/ β© β©2
-
Open WebUI β License (BSD-3 up to v0.6.5; branding clause since v0.6.6 of 2025-04-19, β50 users or fewer over 30 daysβ exemption, not OSI-approved), accessed 2026-10-10. https://docs.openwebui.com/license/ β©
-
Open WebUI β Environment Variable Configuration:
OLLAMA_BASE_URL,OFFLINE_MODE,HF_HUB_OFFLINE,ENABLE_OTEL*,OTEL_EXPORTER_OTLP_ENDPOINT, read on 2026-10-09. https://docs.openwebui.com/reference/env-configuration/ β© β©2 β©3 β©4 β©5 -
Open WebUI β Releases v0.10.0 (2026-06-29: shared folders, external knowledge bases, context compaction) and v0.11.0 (2026-07-27: sub-agents, LDAP group synchronization). https://github.com/open-webui/open-webui/releases/tag/v0.10.0 Β· https://github.com/open-webui/open-webui/releases/tag/v0.11.0 β©
-
Open WebUI β Dockerfile (
SCARF_NO_ANALYTICS=true,DO_NOT_TRACK=true,ANONYMIZED_TELEMETRY=falseby default), read on 2026-10-09. https://github.com/open-webui/open-webui/blob/main/Dockerfile β© -
Open WebUI, Open WebUI v0.13.4: AI that does the work, not just answers (βgo straight to v0.13.4β; v0.13.1 includes security fixes whose advisories will be published once CVEs are assigned; schema change, no gradual update; Direct Connections replaced in v0.13.4 by Tether, free for personal non-commercial use, Enterprise license required in an organization), 2026-11-19 Β· Open WebUI v0.12.4 and v0.12.5 (security fixes in both versions), 2026-11-04 Β· Releases (v0.13.1 of 2026-11-08, v0.13.4 of 2026-11-17). https://openwebui.com/blog/v0-13-4-ai-that-does-the-work-not-just-answers Β· https://openwebui.com/blog/v0-12-4-and-v0-12-5-accessibility-and-shared-notes Β· https://github.com/open-webui/open-webui/releases β©
-
Open WebUI β Security advisories (88 advisories published between 2026-06-11 and 2026-09-28: 30 high, 53 medium, 5 low; GitHub API count of 2026-10-10) and Hardening guide. https://github.com/open-webui/open-webui/security/advisories Β· https://docs.openwebui.com/getting-started/advanced-topics/hardening β©