AI Potluck
Product / UX / UI & API

SurfSense

SurfSense

AI research assistant that connects a model to a personal or team knowledge base plus external sources and answers over them with hybrid retrieval, reachable through one REST API or an MCP server. Operators bring their own model, embedding and reranker choices, including local ones, and run it on their own infrastructure. It is positioned against NotebookLM, Perplexity and Glean.

The connector count and per-feature depth come from the maintainer's README and were not independently checked. Licensing has since split across the tree, and the openness axis was re-derived against the split. Verified 2026-08-13 via the MODSetter/SurfSense repository.

Openness

2 high confidence
2.0
license
Apache-2.0(OSI, everything outside the proprietary directory)+BSL-1.1(Business Source License 1.1 - surfsense_backend/app/proprietary)
source
public(github.com/MODSetter/SurfSense)
self-host
primary(Docker/Compose)
core-gated
ungated

The license is a two-part compound rather than the plain Apache-2.0 it is often taken for. The root LICENSE opens "Portions of this software are licensed as follows" and puts everything under surfsense_backend/app/proprietary/ under the Business Source License 1.1 defined in that directory's own LICENSE, with everything outside it under Apache-2.0. That directory is live rather than a stub - it holds the platform and web-crawler code that is the live-data connector surface the product now leads on. GitHub reads the split file as unclassified. The most restrictive part of a compound license sets the tier, and BSL-1.1 restricts competing use, so public source under it lands at source-available. You can read the whole tree, and the license's terms do not let you run the proprietary half in production. Nothing is withheld from the published source - the restricted directory ships inside it - and the project still says self-hosting stays free and open source, with self-hosted installs running with billing off.

  • https://github.com/MODSetter/SurfSense/blob/main/LICENSE recorded 2026-06-17

    Apache License 2.0

  • https://raw.githubusercontent.com/MODSetter/SurfSense/main/LICENSE recorded 2026-08-14

    Root LICENSE, read rather than assumed. 'Portions of this software are licensed as follows - All content that resides under the "surfsense_backend/app/proprietary/" directory of this repository is licensed under the Business Source License 1.1 as defined in "surfsense_backend/app/proprietary/LICENSE". ... Content outside of the above-mentioned directory is available under the Apache License, Version 2.0', followed by the full Apache text. Two governing licenses, one repository.

  • https://api.github.com/repos/MODSetter/SurfSense/contents/surfsense_backend/app/proprietary recorded 2026-08-14

    The BSL-covered directory exists and ships code - LICENSE, __init__.py, platforms/ and web_crawler/. It is the connector surface rather than a placeholder.

  • https://github.com/MODSetter/SurfSense recorded 2026-08-14

    Repository page, 15,906 stars. GitHub records the license as spdxId NOASSERTION / 'Other', which is GitHub declining to classify the split LICENSE file. The page still reads 'self-hosting stays free and open source' and 'Self-hosted installs run with billing off', so nothing is withheld from the published tree for a paid tier.

Adoption

3 medium confidence
3.0

MODSetter/SurfSense shows 15,902 GitHub stars, and no download or deployment telemetry is available for a self-hosted web app, so stars are the only signal - which caps the level at 3 and falls in the >10K stars band. Star growth is rapid, but real-world deployment is unproven.

Capability

4 low confidence
4.0

An unusually broad connector, model and collaboration surface on paper, but per-feature depth and quality are unverified - almost all of it is README-sourced - and the project is far less proven than Open WebUI-class tools, hence 4 rather than the top and low confidence. The project has repositioned toward being an open web research platform for AI agents, with live data connectors to Reddit, YouTube, Instagram, TikTok, Amazon, Walmart, Google Maps, Google Search and Indeed reachable over one REST API or an MCP server, plus scheduled and event-triggered agents. The maintainers state that the knowledge base, cited chat, reports, podcasts, presentations, automations and collaborative chats all keep working and that self-hosting stays free, so nothing the score rests on has been withdrawn.

Verified 2026-08-13