Published by Gigabrain Labs · Updated
Keep your project knowledge ready for the next task
Gigabrain turns your files and data into knowledge your AI agent can retrieve and use across sessions. It combines ingestion, semantic and lexical search, a source-linked knowledge graph and verification tools. Your agent can trace this context back to the original material. You control the files and connections.
How a piece of knowledge becomes useful again
- Choose one project and the sources it may use.
- Record facts with their origin, scope and date.
- Connect relevant entities and relationships.
- Retrieve evidence for the next task and inspect the supporting source.
- Record corrections and rerun the question rather than trusting an old answer.
Check answers against their sources
Your agent can still make mistakes, especially when sources are missing, conflict or have become outdated. Check its answer against the evidence. Check the tested agents and versions to see how your agent calls Gigabrain’s tools. The demonstrations show the steps and results.
What stays local, and what can use the network
| Operation | Where it runs / network use | What to verify |
|---|---|---|
| Source files and notes | Files in your chosen local project or vault. | Remote sync, backup and shared folders have their own destinations. |
| Agent reasoning | Your chosen agent; a hosted agent sends selected context to its provider. | A local vault does not make a hosted agent offline. |
| Retrieval | Search and citation checks run locally after model setup. | Hosted agents can still send the context you share to their provider. |
| Graph authoring | Optional cloud graph authoring can send selected source excerpts to your chosen provider. | Check which provider receives the material and what you are approving before enabling it. |
| Scheduled operations | An enabled schedule can execute approved operations without another prompt. | Check what runs, where data goes, how often it runs and how to stop it. |
| Updates and remote Git | Downloading updates or sending changes to remote Git storage needs network access. | Choose your remote and inspect changes before applying them. |
Your notes and the retrieval index have different jobs
| Layer | Files | Why it exists |
|---|---|---|
| Knowledge | Sources/, MOCs/, Process/ and curated atom notes | Human-readable knowledge, relationships and decisions; versioned in Git. |
| Retrieval | Corpus/ and _retrieval/ | Generated chunks and indexes; disposable and excluded from the graph and Git. |
A retrieved quotation is checked against its indexed source span. That checks the quotation’s provenance; it does not establish that the source itself is correct. Retain original documents in raw/ and use the validation tools when checking source changes.
Find the meaning, match the detail
Semantic retrieval uses a local embedding model to find passages with related meaning, even when your question uses different words. Lexical search matches words, names and identifiers. Gigabrain combines their rankings, then can follow graph connections to find more source passages.
Ingestion preserves source locations as it prepares material for indexing. Retrieved passages carry quotations and citation details you can verify. You can rebuild the generated indexes without discarding the knowledge you have curated.
Bring the files you already use
Prepare Word documents, Excel workbooks and email exports as readable source copies. Your originals stay unchanged. PDF and Apple Pages imports use installed local readers. Google Docs need a connection or export. A cloud shortcut does not contain the document. Read the connection guide in your download.
Search recordings as well as documents
Transcribe audio and video soundtracks locally after installing the listed tools and speech model. Transcripts keep timestamps for checking the recording. This reads speech, not the images in a video. Review names, numbers and unclear speech before relying on them.
Keep useful memory across sessions
Your agent can suggest facts, preferences, decisions and task steps to save. You approve the exact changes before they become stored memory. Source facts stay separate from your own statements and the conclusions your agent draws. You can correct, retire or restore records. Conversation storage stays off unless you choose it.
Connect through one local interface
The optional MCP server lets an agent call Gigabrain tools through a shared protocol. It supports search, source checks, graph inspection and approved memory changes. Read access is the default. Check the tested integration details before setting it up.
Use the right sources for the task
Your agent can enrich its answers with relevant public research, even when your vault already has useful evidence. It uses deeper research when a task needs several sources or careful checks. It must keep private details private and label outside findings clearly. Public guidance does not prove a fact about your own case. If the evidence still falls short, it should explain what remains unknown. Web access depends on your agent and permissions.
