Research your existing archive directly
When AI agents need to find information in local data, traditional RAG systems are commonly used. These systems typically process the collection into a vector database, which can involve considerable preparation. FinditAgent takes a different approach. The AI is given access to a very powerful search system: Findit6. This locally running program can read practically any file format and simply tell the AI which documents contain the requested terms. The AI can construct suitable search queries from the user’s question and have Findit run them. No vectorization, not even an index – yet searches are very fast.
- No indexing required
- No vectorization or embeddings needed
- No vector database to set up
- No preprocessing or reorganization of your archive
- New and modified files are immediately searchable
Findit6 supports Office documents, PDFs, emails, ZIP and TAR archives, source code and documents whose text has been recognized by OCR, among many other formats. This lets the AI agent work with file collections and folder structures built up over many years.
The Archivist Principle
FinditAgent acts as an intelligent archivist between your files and the AI.
🔒 AI cannot access your files
The language model has no direct access to your files or file system. It cannot open, modify or delete original files itself.
📄 Only the text excerpts needed
The agent asks Findit6 to perform searches, receives results lists and, when necessary, requests specific document contents or excerpts for analysis. The AI is given only extracted copies of text in UTF-8 format .
Your original files remain under your control. If you use an external AI provider, the text copies selected for the research are sent to that service. With sufficiently powerful hardware and a local model via Ollama, all AI processing can also stay on your computer.
Which AI models can be used?
FinditAgent has built-in interfaces for Claude, GPT, Gemini, Grok and Perplexity. As is usual with agentic systems, FinditAgent uses the providers’ APIs rather than their ordinary chat interfaces. These are easy to configure with the respective provider; usage is billed based on data traffic and token consumption. Because FinditAgent searches documents locally and only sends relevant search results to the AI, token use remains moderate. Depending on the model and complexity, a research session can cost between around 10 cents and a few euros in extreme cases.
Local AI with OLLAMA is another option. For maximum privacy, every bit of your data can stay on your own computer. Local models of about 14 billion parameters (14B) and above can work with Findit6. Because they generally have considerably smaller context windows than major cloud models, FinditAgent uses a special Guided Mode: the AI is led through many individual research steps. User questions generally need to be a little more precise. Results from smaller local LLMs can nevertheless be comparable to those of smaller Claude, GPT and similar models. Note: 14B models need at least 12 GB of graphics memory, preferably 16 GB.
What can FinditAgent do?
Unlike traditional AI document assistants, FinditAgent needs neither a prepared document collection nor a vector database. The agent uses direct search results from your existing archive to:
- Find individual facts – the proverbial needle in a haystack.
- Compare multiple documents and weigh statements against each other.
- Detect relationships and patterns across different sources.
- Bring information together and produce a detailed research report supported by sources.
If desired, the AI can also produce comprehensive reports summarizing all documents in your archive relevant to the question. Depending on the capabilities of the model selected, these reports can be remarkably precise, concrete and detailed.
Traceable and verifiable research
FinditAgent is designed so that you do not simply have to trust an AI answer: you can check the evidence yourself.
🔎 Direct source references
The key statements in the final report are linked to the original files found. You can open the documents with a single click to verify them.
📝 Research log
Search queries, result lists, text excerpts read and the agent’s individual steps are logged. You can review and verify the research process later.
This does not automatically make AI results error-free. But documented sources and a traceable research process make potential mistakes much easier to identify.
Quick start on Windows
1. Download and install
Findit Suite includes Findit6 and FinditAgent, the local reranker and the necessary Python runtime. No separate setup of these components is required.
2. Connect an AI model
You need either an API key from a supported cloud provider (such as OpenAI, Anthropic or Gemini) or a working Ollamainstallation with a suitable model for local use.
Detailed instructions on setup, Agent Mode and Guided Mode are available in the documentation and first steps.
FinditAgent in action
A brief look at a real agent research session: