{
  "fixes": {
    "slug": "open-webui",
    "name": "Open WebUI",
    "listing": "https://www.anchorterminal.com/tools/open-webui",
    "markdown": "# Fix list: Open WebUI\n\nFrom Anchor Terminal's listing at https://www.anchorterminal.com/tools/open-webui, the October 2026 research run, assessed 3 October 2026. Grade D, 52 out of 100.\n\nThis is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.\n\nFor a coding agent working on Open WebUI: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.\n\n## 1. Payments \u0026 pricing, 20 out of 100, up to 10 more on the total\n\nWhy it scored 20: Read with the self-hosted rule, scoring the paid option, as the Marmot dossier does with its hosted plan. The paid option is the enterprise licence (white-labelling, SLA-backed support, Terminals), sold to registered organisations through sales. No payment protocol (0). No published price (0). Self-hosting is free with no account at Open WebUI Inc. and no card, for any number of users who keep the branding (20). The licence goes through sales, and on a self-hosted instance an administrator has to turn API keys on before an agent can get one (0).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):\n\nThe published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).\n\n- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.\n- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for \"contact sales\" or prices behind a login.\n- 20, a free tier or trial that doesn't need a card.\n- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).\n\nPayment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.\n\nOpen-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.\n\n## 2. Agent ergonomics, 54 out of 100, up to 7.5 more on the total\n\nWhy it scored 54: Graded on the HTTP API. Chat completions take OpenAI's sizing controls such as `max_tokens` and can stream, but `/api/models` returns every model with its full metadata and nothing selects fields (14 of 25). File, knowledge, chat and model lists page with `page`, file search takes `skip` and `limit` (1 to 1,000), and files, chats and knowledge can be searched (13 of 20). Errors are FastAPI `detail` strings from a shared message table with HTTP codes, and upstream provider errors pass through. Readable, and mostly undocumented (12 of 20). Reads are GETs, and we found no idempotency keys or retry guidance for writes (5 of 20). A completion needs two fields and OpenAI's SDKs work against the chat routes, but Open WebUI publishes no SDK of its own (10 of 15).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):\n\n- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).\n- 20, pagination, filtering and output-size controls.\n- 20, actionable, documented error responses, codes and messages an agent can recover from.\n- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.\n- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.\n\nModels are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.\n\n## 3. Schema \u0026 documentation, 60 out of 100, up to 6.5 more on the total\n\nWhy it scored 60: Graded on the HTTP API an agent calls, the OpenAI-compatible /api/chat/completions and /api/models plus the /api/v1 file, knowledge and retrieval routes. Open WebUI is an MCP client and runs no MCP server. FastAPI's OpenAPI document and Swagger UI exist only with `ENV=dev`, which the docs explain, while the chat and model routes follow OpenAI's published request shapes (12 of 25). llms.txt with an index of every page, and a Markdown copy of each page (10). The API reference at docs.openwebui.com/reference/api-endpoints explains seven route groups (models, chat completions, the Anthropic Messages route with token counting, the Ollama proxy, file upload with processing status, knowledge and web retrieval) and what each is for, out of hundreds of routes (10 of 20). `/api/chat/completions` takes a free-form `form_data: dict`, while most /api/v1 routes use Pydantic forms (6 of 15). curl and Python examples for the documented routes, and the reference names 400, 403 and 502 for a few cases (9 of 15). /api/v1 prefixes on most routes and a dated CHANGELOG.md entry for every release (13 of 15).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):\n\nAPIs and MCP servers.\n\n- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).\n- 10, llms.txt or Markdown docs served for agents.\n- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.\n- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.\n- 0 to 15, examples and documented error responses.\n- 15, versioning and a public changelog.\n\nModels are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.\n\n## 4. Security \u0026 auth, 63 out of 100, up to 6.5 more on the total\n\nWhy it scored 63: Graded on the HTTP API with the tool lines. API keys stay off until an administrator sets `ENABLE_API_KEYS`, and a group permission decides who may create one. Each user has one `sk-` key, regenerable, stored as plain text with a last-used time and no expiry set by the API, sent as a Bearer token or an `x-api-key` header. An administrator can hold every key to an endpoint allowlist, but keys have no scopes of their own and carry their user's rights, and session JWTs last four weeks by default. No secret in a query string (22 of 30). Roles, groups, per-feature permissions, per-model and per-knowledge access grants and the allowlist can keep a key to chat and models, and sign-up closes once the first account becomes admin. An admin's key reaches everything, API deletes ask for no confirmation, and the per-call tool approval added in 0.11.1 works in saved chats in the interface, off by default, not on the API. More than half of the past year's advisories were access-control or authorisation flaws in this permission model, all fixed (10 of 20). Retrieved context is wrapped in tags and a debug log flags context that contains them, and we found no injection guidance (5 of 15). An audit log at metadata, request or request-and-response level, off by default (`AUDIT_LOG_LEVEL=NONE`), events for key creation and deletion, and opt-in OpenTelemetry (12 of 15). SECURITY.md, a security.txt valid to 30 June 2027, reports only through GitHub advisories, and each accepted report published as an advisory after the fix. No bounty and no SOC 2, and the policy refuses reports for flaws already fixed in the open, which then get no advisory (14 of 20).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-security):\n\n- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.\n- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.\n- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.\n- 0 to 15, audit logs or per-call visibility for the operator.\n- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.\n\nModels are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.\n\n## 5. Reliability, 68 out of 100, up to 6.4 more on the total\n\nWhy it scored 68: Read with the local-software lines, as the Goose and Aider calibration dossiers do. `open-webui` on PyPI for Python 3.11 and 3.12, Docker images (`:main`, `:cuda`, `:ollama` and a slim build of about 175 MB), Compose, Kustomize, Helm and desktop apps (20). The frontend workflow lints, builds and runs unit tests on pushes to main and dev, while the backend workflow on push only checks Ruff formatting and the backend suite in open-webui/tests runs on release pull requests. Codespell and both lint workflows are switched off. The runs we saw passed (18 of 25). 234 open issues against issue numbers past 30,000, recent bug reports labelled `bug` and `confirmed issue` within a day, and 0.11.3 made a failed upgrade stop at the migration error after reports of half-upgraded instances (#29280) (20 of 25). CHANGELOG.md follows Keep a Changelog and claims semver, with dated entries and migration warnings, but schema migrations ship in patch releases (0.10.2, 0.11.1) and no 2026 entry carries a breaking-change label (10 of 15). 0.11.4, pre-1.0, and classed 4 - Beta on PyPI (0).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):\n\nHosted APIs, MCP servers, models and platforms.\n\n- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).\n- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.\n- 15, rate limits documented with numbers.\n- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.\n- 10, an SLA published for any paid tier.\n- 10, the surface agents use is generally available, not beta or preview.\n\nLocal packages, SDKs, frameworks and stdio MCP servers.\n\n- 20, installs from an official package with supported runtimes stated.\n- 25, a public CI and test suite, passing on the default branch.\n- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).\n- 15, semver discipline and breaking changes called out in a changelog.\n- 15, version 1.0 or later, or declared stable.\n\nProtocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.\n\n## 6. Transparency \u0026 trust, 73 out of 100, up to 2.4 more on the total\n\nMade of editorial 66, provenance 79.\n\nWhy it scored 73: The Open WebUI License is BSD-3-Clause with a clause that forbids changing the branding in deployments over 50 users in a rolling 30 days without written permission or an enterprise licence. It isn't OSI-approved and contributors sign a CLA, but the source is public, the terms are short, commercial use is allowed, and older code stays under MIT or BSD-3-Clause per LICENSE_HISTORY (22 of 30). The FAQ says Open WebUI doesn't send data to external services and makes no external calls by default, while the server checks api.github.com for releases by default and downloads its embedding model at start unless `OFFLINE_MODE` is on. The privacy policy of 31 December 2025 covers only openwebui.com and gives no retention periods, and the chat data privacy page sets out who can read chats without retention defaults (16 of 30). Dated changelog entries flag migrations, renamed settings keep deprecated aliases and Pipelines is marked legacy, with no deprecation policy (10 of 20). We found no product telemetry to the vendor. The Docker image sets `DO_NOT_TRACK`, `SCARF_NO_ANALYTICS` and `ANONYMIZED_TELEMETRY=false`, Chroma's telemetry is off in code, and `ENABLE_VERSION_UPDATE_CHECK=false` or `OFFLINE_MODE` stops the release check (18 of 20).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):\n\n- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.\n- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).\n- 0 to 20, a deprecation policy or notices with dates.\n- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).\n\nThe other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.\n\nProvenance checks not met in full (half of this category, computed from checked facts):\n\n- Domain age: openwebui.com, registered 2024-02-17 (2 years) (7 of 15)\n- Status page: not found (0 of 10)\n\n## 7. Maintenance \u0026 community, 91 out of 100, up to 0.8 more on the total\n\nWhy it scored 91: 0.11.4 on 21 September 2026 (30). Five releases in the 90 days to 3 October, 0.11.0 (27 July) to 0.11.4 (20). Recent bug reports are labelled and confirmed within a day and pull requests merge daily, but we couldn't see reply times (20 of 25). Read as current official builds, as the calibration dossiers read this line for local software. PyPI, GHCR and desktop builds ship from each release, and there's no SDK of its own (13 of 15). Dependabot monthly for uv, pip, npm and Actions, a uv.lock, and CI passing, with codespell and both lint workflows switched off (8 of 10).\n\nThe checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):\n\n- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.\n- 20, at least three releases or dated changelog entries in the last 90 days.\n- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.\n- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).\n- 10, package health, current dependencies and CI.\n\nModels are read for deprecation notice periods and model churn rather than release counts.\n\n## Deductions\n\nEach comes off the total. A fixed and documented problem counts for less at the next check.\n\n- 2026-05-05. GHSA-2r4p-jpmg-48f4 (CVE-2026-44551, Critical, 9.1). LDAP sign-in accepted an empty password where the directory allows unauthenticated binds, giving full access to the victim's account. It affects 0.8.12 and earlier and was fixed in 0.9.0 (21 April 2026) before publication, -2. https://github.com/open-webui/open-webui/security/advisories/GHSA-2r4p-jpmg-48f4\n- 2026-07-02. GHSA-74h3-cxq7-vc5q (CVE-2026-59216, 7.7). A signed-in low-privilege user could run code and tools in another user's session through an unchecked Socket.IO session_id, which against an administrator meant code execution as the server process (root in default containers). Fixed in 0.10.0 on 29 June 2026 and published three days later, -2. https://github.com/open-webui/open-webui/security/advisories/GHSA-74h3-cxq7-vc5q\n- 2026-08-02 and 2026-09-04. Two account takeovers through OAuth, both High. GHSA-rq84-p6rr-vf89 accepted tokens issued to any client in the OAuth token exchange (fixed in 0.11.0), and GHSA-wpmr-8h3q-fwj7 (8.1) matched OAuth and OIDC subjects by substring on SQLite, so a crafted subject could sign in as an existing account, administrators included (fixed in 0.11.1 on 25 August). Both fixed before publication, -2. https://github.com/open-webui/open-webui/security/advisories/GHSA-rq84-p6rr-vf89; https://github.com/open-webui/open-webui/security/advisories/GHSA-wpmr-8h3q-fwj7\n- 2025-10-03 to 2026-10-03. The rest of the year's record. GitHub reviewed 143 of the repository's advisories in the 12 months to 3 October 2026, and 129 cover flaws fixed in releases from 0.6.35 (6 November 2025) on, 58 High and 1 Critical, more than half of them access-control or authorisation flaws by their titles and CWE tags. July to September alone brought 52 (18 High, 29 Moderate, 5 Low) in batches published on 2 July, 2 August and 4 September. Each was fixed in a release before publication, so the 125 beyond the four above count together, -2. https://github.com/open-webui/open-webui/security/advisories; https://github.com/advisories?query=open-webui+type%3Areviewed+ecosystem%3Apip\n\n## What we couldn't check\n\nWhat we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.\n\n- Our advisory counts come from the GitHub Advisory Database records that link to each repository advisory, because the repository's pages paginate out of order. Those records carry GitHub's review dates, so we took publication dates (2 July, 2 August, 4 September) from the repository advisories we opened\n- Whether the regression suite passed on the 0.11.4 release pull request. The runs we saw were contributor pull requests\n- Unchecked: the terms of service at openwebui.com/terms\n- Unchecked: release cadence of the desktop app, which lives in a separate repository (open-webui/desktop)\n- Whether the enterprise licence's SLA has published terms. The enterprise page mentions SLA-backed support without numbers\n\n## Weaknesses\n\n- API keys are off by default, and each user gets one key with no scopes or expiry\n- The API reference covers seven route groups, and the OpenAPI file and Swagger UI need `ENV=dev`\n- 52 advisories published from July to September 2026, 18 of them High, including cross-user code execution (CVE-2026-59216) and two OAuth account takeovers, all fixed\n- Pre-1.0 (0.11), with database migrations in patch releases and no rolling updates during them\n- The branding clause makes the licence non-OSI, and the enterprise licence has no published price\n\n## What costs an agent a turn today\n\nThe notes we give agents before they call it. Each one is a workaround an agent shouldn't need.\n\n- Ask the administrator to set `ENABLE_API_KEYS=true` and let your group create keys. `sk-` keys are refused until then\n- Send OpenAI's request shape to `/api/chat/completions` with a Bearer key, or use `x-api-key` behind a proxy that takes `Authorization` for itself\n- Call `/api/models` first and use an `id` from it. Model IDs depend on the instance's connections\n- Poll `GET /api/v1/files/{id}/process/status` until it reads `completed` before adding a file to a knowledge base\n- Expect a 403 on routes outside `API_KEYS_ALLOWED_ENDPOINTS` when the administrator has set an allowlist\n\n## What the review panel asked for\n\n- breaking-change labels\n- a deprecation policy\n- scoped keys with expiry\n- tool approval on the API\n\n## When it's done\n\nSend what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `\"kind\": \"dispute\"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.\n",
    "grade": "D",
    "score": 52,
    "assessed": "2026-10-03",
    "run": "October 2026 research run",
    "categories": [
      {
        "key": "payments",
        "name": "Payments \u0026 pricing",
        "score": 20,
        "maxGain": 10,
        "reason": "Read with the self-hosted rule, scoring the paid option, as the Marmot dossier does with its hosted plan. The paid option is the enterprise licence (white-labelling, SLA-backed support, Terminals), sold to registered organisations through sales. No payment protocol (0). No published price (0). Self-hosting is free with no account at Open WebUI Inc. and no card, for any number of users who keep the branding (20). The licence goes through sales, and on a self-hosted instance an administrator has to turn API keys on before an agent can get one (0).",
        "checklist": [
          "The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).",
          "- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.\n- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for \"contact sales\" or prices behind a login.\n- 20, a free tier or trial that doesn't need a card.\n- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).",
          "Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.",
          "Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-payments"
      },
      {
        "key": "ergonomics",
        "name": "Agent ergonomics",
        "score": 54,
        "maxGain": 7.5,
        "reason": "Graded on the HTTP API. Chat completions take OpenAI's sizing controls such as `max_tokens` and can stream, but `/api/models` returns every model with its full metadata and nothing selects fields (14 of 25). File, knowledge, chat and model lists page with `page`, file search takes `skip` and `limit` (1 to 1,000), and files, chats and knowledge can be searched (13 of 20). Errors are FastAPI `detail` strings from a shared message table with HTTP codes, and upstream provider errors pass through. Readable, and mostly undocumented (12 of 20). Reads are GETs, and we found no idempotency keys or retry guidance for writes (5 of 20). A completion needs two fields and OpenAI's SDKs work against the chat routes, but Open WebUI publishes no SDK of its own (10 of 15).",
        "checklist": [
          "- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).\n- 20, pagination, filtering and output-size controls.\n- 20, actionable, documented error responses, codes and messages an agent can recover from.\n- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.\n- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.",
          "Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-ergonomics"
      },
      {
        "key": "schema",
        "name": "Schema \u0026 documentation",
        "score": 60,
        "maxGain": 6.5,
        "reason": "Graded on the HTTP API an agent calls, the OpenAI-compatible /api/chat/completions and /api/models plus the /api/v1 file, knowledge and retrieval routes. Open WebUI is an MCP client and runs no MCP server. FastAPI's OpenAPI document and Swagger UI exist only with `ENV=dev`, which the docs explain, while the chat and model routes follow OpenAI's published request shapes (12 of 25). llms.txt with an index of every page, and a Markdown copy of each page (10). The API reference at docs.openwebui.com/reference/api-endpoints explains seven route groups (models, chat completions, the Anthropic Messages route with token counting, the Ollama proxy, file upload with processing status, knowledge and web retrieval) and what each is for, out of hundreds of routes (10 of 20). `/api/chat/completions` takes a free-form `form_data: dict`, while most /api/v1 routes use Pydantic forms (6 of 15). curl and Python examples for the documented routes, and the reference names 400, 403 and 502 for a few cases (9 of 15). /api/v1 prefixes on most routes and a dated CHANGELOG.md entry for every release (13 of 15).",
        "checklist": [
          "APIs and MCP servers.",
          "- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).\n- 10, llms.txt or Markdown docs served for agents.\n- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.\n- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.\n- 0 to 15, examples and documented error responses.\n- 15, versioning and a public changelog.",
          "Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-schema"
      },
      {
        "key": "security",
        "name": "Security \u0026 auth",
        "score": 63,
        "maxGain": 6.5,
        "reason": "Graded on the HTTP API with the tool lines. API keys stay off until an administrator sets `ENABLE_API_KEYS`, and a group permission decides who may create one. Each user has one `sk-` key, regenerable, stored as plain text with a last-used time and no expiry set by the API, sent as a Bearer token or an `x-api-key` header. An administrator can hold every key to an endpoint allowlist, but keys have no scopes of their own and carry their user's rights, and session JWTs last four weeks by default. No secret in a query string (22 of 30). Roles, groups, per-feature permissions, per-model and per-knowledge access grants and the allowlist can keep a key to chat and models, and sign-up closes once the first account becomes admin. An admin's key reaches everything, API deletes ask for no confirmation, and the per-call tool approval added in 0.11.1 works in saved chats in the interface, off by default, not on the API. More than half of the past year's advisories were access-control or authorisation flaws in this permission model, all fixed (10 of 20). Retrieved context is wrapped in tags and a debug log flags context that contains them, and we found no injection guidance (5 of 15). An audit log at metadata, request or request-and-response level, off by default (`AUDIT_LOG_LEVEL=NONE`), events for key creation and deletion, and opt-in OpenTelemetry (12 of 15). SECURITY.md, a security.txt valid to 30 June 2027, reports only through GitHub advisories, and each accepted report published as an advisory after the fix. No bounty and no SOC 2, and the policy refuses reports for flaws already fixed in the open, which then get no advisory (14 of 20).",
        "checklist": [
          "- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.\n- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.\n- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.\n- 0 to 15, audit logs or per-call visibility for the operator.\n- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.",
          "Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-security"
      },
      {
        "key": "reliability",
        "name": "Reliability",
        "score": 68,
        "maxGain": 6.4,
        "reason": "Read with the local-software lines, as the Goose and Aider calibration dossiers do. `open-webui` on PyPI for Python 3.11 and 3.12, Docker images (`:main`, `:cuda`, `:ollama` and a slim build of about 175 MB), Compose, Kustomize, Helm and desktop apps (20). The frontend workflow lints, builds and runs unit tests on pushes to main and dev, while the backend workflow on push only checks Ruff formatting and the backend suite in open-webui/tests runs on release pull requests. Codespell and both lint workflows are switched off. The runs we saw passed (18 of 25). 234 open issues against issue numbers past 30,000, recent bug reports labelled `bug` and `confirmed issue` within a day, and 0.11.3 made a failed upgrade stop at the migration error after reports of half-upgraded instances (#29280) (20 of 25). CHANGELOG.md follows Keep a Changelog and claims semver, with dated entries and migration warnings, but schema migrations ship in patch releases (0.10.2, 0.11.1) and no 2026 entry carries a breaking-change label (10 of 15). 0.11.4, pre-1.0, and classed 4 - Beta on PyPI (0).",
        "checklist": [
          "Hosted APIs, MCP servers, models and platforms.",
          "- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).\n- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.\n- 15, rate limits documented with numbers.\n- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.\n- 10, an SLA published for any paid tier.\n- 10, the surface agents use is generally available, not beta or preview.",
          "Local packages, SDKs, frameworks and stdio MCP servers.",
          "- 20, installs from an official package with supported runtimes stated.\n- 25, a public CI and test suite, passing on the default branch.\n- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).\n- 15, semver discipline and breaking changes called out in a changelog.\n- 15, version 1.0 or later, or declared stable.",
          "Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-reliability"
      },
      {
        "key": "transparency",
        "name": "Transparency \u0026 trust",
        "score": 73,
        "maxGain": 2.4,
        "reason": "The Open WebUI License is BSD-3-Clause with a clause that forbids changing the branding in deployments over 50 users in a rolling 30 days without written permission or an enterprise licence. It isn't OSI-approved and contributors sign a CLA, but the source is public, the terms are short, commercial use is allowed, and older code stays under MIT or BSD-3-Clause per LICENSE_HISTORY (22 of 30). The FAQ says Open WebUI doesn't send data to external services and makes no external calls by default, while the server checks api.github.com for releases by default and downloads its embedding model at start unless `OFFLINE_MODE` is on. The privacy policy of 31 December 2025 covers only openwebui.com and gives no retention periods, and the chat data privacy page sets out who can read chats without retention defaults (16 of 30). Dated changelog entries flag migrations, renamed settings keep deprecated aliases and Pipelines is marked legacy, with no deprecation policy (10 of 20). We found no product telemetry to the vendor. The Docker image sets `DO_NOT_TRACK`, `SCARF_NO_ANALYTICS` and `ANONYMIZED_TELEMETRY=false`, Chroma's telemetry is off in code, and `ENABLE_VERSION_UPDATE_CHECK=false` or `OFFLINE_MODE` stops the release check (18 of 20).",
        "blend": "editorial 66, provenance 79",
        "checklist": [
          "- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.\n- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).\n- 0 to 20, a deprecation policy or notices with dates.\n- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).",
          "The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-transparency"
      },
      {
        "key": "maintenance",
        "name": "Maintenance \u0026 community",
        "score": 91,
        "maxGain": 0.8,
        "reason": "0.11.4 on 21 September 2026 (30). Five releases in the 90 days to 3 October, 0.11.0 (27 July) to 0.11.4 (20). Recent bug reports are labelled and confirmed within a day and pull requests merge daily, but we couldn't see reply times (20 of 25). Read as current official builds, as the calibration dossiers read this line for local software. PyPI, GHCR and desktop builds ship from each release, and there's no SDK of its own (13 of 15). Dependabot monthly for uv, pip, npm and Actions, a uv.lock, and CI passing, with codespell and both lint workflows switched off (8 of 10).",
        "checklist": [
          "- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.\n- 20, at least three releases or dated changelog entries in the last 90 days.\n- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.\n- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).\n- 10, package health, current dependencies and CI.",
          "Models are read for deprecation notice periods and model churn rather than release counts."
        ],
        "checklistUrl": "https://www.anchorterminal.com/benchmark/#checklist-maintenance"
      }
    ],
    "provenance": [
      {
        "label": "Domain age",
        "value": "openwebui.com, registered 2024-02-17 (2 years)",
        "points": 7,
        "max": 15
      },
      {
        "label": "Status page",
        "value": "not found",
        "points": 0,
        "max": 10
      }
    ],
    "deductions": [
      "2026-05-05. GHSA-2r4p-jpmg-48f4 (CVE-2026-44551, Critical, 9.1). LDAP sign-in accepted an empty password where the directory allows unauthenticated binds, giving full access to the victim's account. It affects 0.8.12 and earlier and was fixed in 0.9.0 (21 April 2026) before publication, -2. https://github.com/open-webui/open-webui/security/advisories/GHSA-2r4p-jpmg-48f4",
      "2026-07-02. GHSA-74h3-cxq7-vc5q (CVE-2026-59216, 7.7). A signed-in low-privilege user could run code and tools in another user's session through an unchecked Socket.IO session_id, which against an administrator meant code execution as the server process (root in default containers). Fixed in 0.10.0 on 29 June 2026 and published three days later, -2. https://github.com/open-webui/open-webui/security/advisories/GHSA-74h3-cxq7-vc5q",
      "2026-08-02 and 2026-09-04. Two account takeovers through OAuth, both High. GHSA-rq84-p6rr-vf89 accepted tokens issued to any client in the OAuth token exchange (fixed in 0.11.0), and GHSA-wpmr-8h3q-fwj7 (8.1) matched OAuth and OIDC subjects by substring on SQLite, so a crafted subject could sign in as an existing account, administrators included (fixed in 0.11.1 on 25 August). Both fixed before publication, -2. https://github.com/open-webui/open-webui/security/advisories/GHSA-rq84-p6rr-vf89; https://github.com/open-webui/open-webui/security/advisories/GHSA-wpmr-8h3q-fwj7",
      "2025-10-03 to 2026-10-03. The rest of the year's record. GitHub reviewed 143 of the repository's advisories in the 12 months to 3 October 2026, and 129 cover flaws fixed in releases from 0.6.35 (6 November 2025) on, 58 High and 1 Critical, more than half of them access-control or authorisation flaws by their titles and CWE tags. July to September alone brought 52 (18 High, 29 Moderate, 5 Low) in batches published on 2 July, 2 August and 4 September. Each was fixed in a release before publication, so the 125 beyond the four above count together, -2. https://github.com/open-webui/open-webui/security/advisories; https://github.com/advisories?query=open-webui+type%3Areviewed+ecosystem%3Apip"
    ],
    "unchecked": [
      "Our advisory counts come from the GitHub Advisory Database records that link to each repository advisory, because the repository's pages paginate out of order. Those records carry GitHub's review dates, so we took publication dates (2 July, 2 August, 4 September) from the repository advisories we opened",
      "Whether the regression suite passed on the 0.11.4 release pull request. The runs we saw were contributor pull requests",
      "Unchecked: the terms of service at openwebui.com/terms",
      "Unchecked: release cadence of the desktop app, which lives in a separate repository (open-webui/desktop)",
      "Whether the enterprise licence's SLA has published terms. The enterprise page mentions SLA-backed support without numbers"
    ],
    "weaknesses": [
      "API keys are off by default, and each user gets one key with no scopes or expiry",
      "The API reference covers seven route groups, and the OpenAPI file and Swagger UI need `ENV=dev`",
      "52 advisories published from July to September 2026, 18 of them High, including cross-user code execution (CVE-2026-59216) and two OAuth account takeovers, all fixed",
      "Pre-1.0 (0.11), with database migrations in patch releases and no rolling updates during them",
      "The branding clause makes the licence non-OSI, and the enterprise licence has no published price"
    ],
    "agentNotes": [
      "Ask the administrator to set `ENABLE_API_KEYS=true` and let your group create keys. `sk-` keys are refused until then",
      "Send OpenAI's request shape to `/api/chat/completions` with a Bearer key, or use `x-api-key` behind a proxy that takes `Authorization` for itself",
      "Call `/api/models` first and use an `id` from it. Model IDs depend on the instance's connections",
      "Poll `GET /api/v1/files/{id}/process/status` until it reads `completed` before adding a file to a knowledge base",
      "Expect a 403 on routes outside `API_KEYS_ALLOWED_ENDPOINTS` when the administrator has set an allowlist"
    ],
    "requests": [
      {
        "text": "breaking-change labels",
        "reviews": 1
      },
      {
        "text": "a deprecation policy",
        "reviews": 1
      },
      {
        "text": "scoped keys with expiry",
        "reviews": 1
      },
      {
        "text": "tool approval on the API",
        "reviews": 1
      }
    ],
    "recheck": "https://www.anchorterminal.com/builders/#disputes"
  },
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-10-04",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.3",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run"
  }
}