[KongchangAI]
ProductSurfSense Agent

SurfSense

An open-source AI agent product positioned as a NotebookLM with an agent. It can retrieve documents and generate and edit editable files such as Word, PDF, PPT, and Excel documents, and it supports MCP plugins and multiple model backends.

Timeline (last 90 days)

Oct 31

SurfSense is an open-source Agent project under the Apache-2.0 license that supports Windows, macOS and Linux.

Unverified50%
Oct 31

SurfSense can generate Word, PDF, PPT, Excel and web page files, and can edit files the user already has.

Unverified50%
Oct 31

SurfSense's model can be a local model, any OpenAI-compatible API, or directly reuse the user's ChatGPT/Claude subscription plan.

Unverified50%
Oct 31

When SurfSense edits user files, all changes are delivered as a new copy and the original file is never written to; Word edits are returned in track-changes mode plus comments, Excel updates formulas, and PDFs can have forms filled in.

Unverified50%
Oct 31

SurfSense supports MCP plugins and can add external MCP servers via URL, such as Notion and DeepWiki.

Unverified50%
Oct 31

SurfSense uses opencode as the execution engine for all model-driven Agents; if the user uses a Claude plan, execution is taken over by Claude Code, and both paths share the same set of tools and rules.

Unverified50%
Oct 31

For security reasons SurfSense has disabled shell, code execution, sub-Agents and web access; Agents can only use a fixed tool set: search, document tools and PDF tools.

Unverified50%
Oct 31

SurfSense adopts a "document as script" design: the Agent writes each document as a script, SurfSense runs it in a separate process, keeps every version, and sends page previews back to the Agent to check and correct the output.

Unverified50%
Oct 31

In SurfSense each conversation has its own folder storing the text of the selected source materials, which the Agent can directly read and grep.

Unverified50%
Oct 31

SurfSense's Agent mode requires the model to support tool calling and a context of at least 32K; smaller models can still be used for ordinary Q&A.

Unverified50%

2 more timeline events

All Facts (12)

Unverified

SurfSense is an open-source Agent project under the Apache-2.0 license that supports Windows, macOS and Linux.

50%
Unverified

SurfSense can generate Word, PDF, PPT, Excel and web page files, and can edit files the user already has.

50%
Unverified

SurfSense's model can be a local model, any OpenAI-compatible API, or directly reuse the user's ChatGPT/Claude subscription plan.

50%
Unverified

When SurfSense edits user files, all changes are delivered as a new copy and the original file is never written to; Word edits are returned in track-changes mode plus comments, Excel updates formulas, and PDFs can have forms filled in.

50%
Unverified

SurfSense supports MCP plugins and can add external MCP servers via URL, such as Notion and DeepWiki.

50%
Unverified

SurfSense uses opencode as the execution engine for all model-driven Agents; if the user uses a Claude plan, execution is taken over by Claude Code, and both paths share the same set of tools and rules.

50%
Unverified

For security reasons SurfSense has disabled shell, code execution, sub-Agents and web access; Agents can only use a fixed tool set: search, document tools and PDF tools.

50%
Unverified

SurfSense adopts a "document as script" design: the Agent writes each document as a script, SurfSense runs it in a separate process, keeps every version, and sends page previews back to the Agent to check and correct the output.

50%
Unverified

In SurfSense each conversation has its own folder storing the text of the selected source materials, which the Agent can directly read and grep.

50%
Unverified

SurfSense's Agent mode requires the model to support tool calling and a context of at least 32K; smaller models can still be used for ordinary Q&A.

50%
Unverified

SurfSense can be downloaded at surfsense.com/downloads, and the source code is in the MODSetter/SurfSense repository on GitHub.

50%
Unverified

In a Reddit post, the SurfSense maintainer raised two open questions: 1) whether documents as version-managed scripts combined with preview self-checking is the right pattern, or whether the model should edit files directly; 2) how to decide which tools should require approval.

50%

Source Articles