SAG Wiki · Product demo01
Fast local search
Add a folder to find content, preview source text and highlight matches. Quick search needs no API key.
03 / SAG WIKILOCAL KNOWLEDGE BASE
FOR PEOPLE & AGENTS
SAG Wiki
Your local files. Ready to search.
Add a folder to search documents, notes and code. Ask across files and check every source. Built for people and agents.
14.0MB
macOS installer · v0.1.7
0
Model calls for local indexing
SAG2
Retrieval across source documents
Installer size excludes optional document converters and recognition models downloaded on demand.
01 / CAPABILITIES
Search, ask and check in one place
Keep your files organized as they are. The knowledge base indexes them for reading, cited answers and agent access.
02
Cited answers across files
Answer with multiple documents and use @ to select files. Open citations to see the supporting passages.
03
Keep up with file changes
Watch local files and update the index incrementally. Files stay in place, even when a folder is removed from the library.
04
Read and edit
Edit Markdown and preview HTML in the app, moving from search to reading and editing.
05
Explore related documents
Find documents with similar content and inspect their shared evidence. Relationship calculations run locally.
06
Give agents the same knowledge
Use a standalone CLI and Skill. The desktop and CLI share settings and the library; the CLI also works on its own.
Direct indexing: Markdown, text, RTF, supported code and configuration filesPDF / Office: enable the optional document converter
02 / EVERYDAY WORKFLOW
From a folder to an answer you can verify
Select a step to inspect the actual interface. Follow the links between files, search results and answer citations.
03 / POWERED BY SAG2
Lighter preparation. More complete retrieval.
Build a keyword index directly from the source, without LLMs or embeddings. Jev judges useful evidence at query time, and terms from that evidence drive further searches.
01Keyword search
02Jev evidence judgment
03Expand the search
04Return source passages
Published results for the underlying SAG2 methodRecall@10
MuSiQue
86.62%
SAG2
86.62
BM25
54.41
2WikiMultiHopQA
97.43%
SAG2
97.43
BM25
71.50
HotpotQA
99.00%
SAG2
99.00
BM25
86.90
Recall@10 measures evidence found in the first ten results. SAG2 uses 1,000 questions per benchmark; MuSiQue was used for method development. These are method results; validate performance on your own files.
Source: SAG2 paper, Table 104 / FOR YOUR AGENT
Let your agent search the project
Send the setup instruction to your agent to configure the Skill, index the current project and retrieve relevant passages. The CLI works independently and shares settings and the library with the desktop app.
Claude CodeCodexOpenCode
View the official SkillSEND TO YOUR AGENT
Install https://github.com/Zleap-AI/SAG-Wiki/blob/main/cli/skills/sag/SKILL.md and follow its setup flow for my current project.
The agent follows the setup instructions to configure the CLI and index your project.Install the CLI manually
Requires Node.js 18+. Configure your Jev API key before the first semantic search.
npm install -g @zleap-ai/sag
sag add
sag check
sag "Where is permission checked?" . -m 505 / LOCAL FILES, YOUR SERVICES
Local files. Services you configure.
Original files and indexes remain local. Semantic retrieval and Q&A send the required question, passages or context to your configured services. Provider charges depend on the services you choose.
Read the setup guide- Indexing and quick search
- Local, with no API key required
- SAG2 semantic retrieval
- Configure a Jev service and API key
- Cited Q&A
- Configure an answer model, plus Jev when using SAG2
06 / GET STARTED
Download. Add your first folder.
SAG Wiki is free to use and open source under MIT.
The current macOS build is not Apple-notarized and may show a security prompt on first launch. Installers exclude optional conversion components.