A personal project for asking questions of your own files, with the passage behind each claim.
Source-available · Self-hosted · Personal project
Your Private AI
Knowledge Companion
Ask about your documents. See the source of each claim. The knowledge store stays on your machine.
Source-available under FSL-1.1-ALv2. Each release becomes Apache-2.0 two years after publication.
A short tour
Claim verification, the knowledge map, the wiki, and local sources.
Verified answers, Constellation explore & drill-down, wiki, TrustScore, and key features — captured from the live product.
Claim verification & refutation, map explore, local sources — short cut for social and quick scans.
Self-hosted. Knowledge stores stay local. Source-available under FSL-1.1-ALv2. Get Core free
How it works
Import files, ask a question, read the citation.
1
Import your files
Drop a folder or upload files. Cerid parses PDFs, docs, code, emails, and archives.
2
Ask a question
A question searches the knowledge you imported.
3
See which documents answered
Each claim points at the passage it came from.
4
The record accumulates
Facts, decisions, and preferences stay in the store and can be forgotten.
What the repository does
The public tree. Tap a card for a longer note.
Know where answers come from
Every answer includes inline citations so you can see which document, and which passage, backs each claim.
Streaming claim verification with verified, partial, and refuted outcomes. Click a footnote for the source and the reasoning.
Explore your knowledge as a map
Constellation shows entities and clusters you can zoom and drill into.
A force-layout map of the corpus: hubs, domains, and neighbors, with a wiki page behind a node.
A wiki that accumulates
Readable entity pages grow from your sources. Browse by domain, open an article, return later.
Wiki entities with a summary, co-mentions, references, and a verification status. Pages refresh as sources update.
Disclosed TrustScore
A single score with the methodology in the open: the components, and what a drop means.
A straight mean of measured components, including retrieval, memory recall, verification coverage, and preservation health. No learned weights.
Finds the right part of your documents
A question searches the relevant sections across your files, including when the answer spans more than one document.
Hybrid keyword and vector search, with a cross-encoder rerank. Retrieval depth follows the question.
Works with the model you choose
Use a hosted model, or run a local one. The corpus stays on the machine either way.
The pipeline can run against a local model. A hosted model is optional, and only the chosen context is sent.
Your knowledge stores stay on your device
Self-hosted by design. Telemetry is off unless you turn it on.
Docker stack with Chroma, Neo4j, and Redis on your machine. Optional encryption at rest. Source-available under FSL-1.1-ALv2.
Privacy by design
The knowledge base stays on your machine. Telemetry is off unless you turn it on.
- Knowledge stores stay on your machine
- Credentials can be encrypted at rest
- A local model can answer with nothing sent out
- Source-available under FSL-1.1-ALv2
- A hosted model is optional
Stored locally
Documents, embeddings, and metadata stay on your machine.
Encrypted at rest
API keys and credentials can be protected with Fernet encryption.
Source-available
FSL-1.1-ALv2. Each release converts to Apache-2.0 after two years.
The repository
Clone it and run the setup script. There is no account and no subscription.
Terminal
$ git clone https://github.com/Cerid-AI/cerid-ai.git
$ cd cerid-ai && ./scripts/start-cerid.sh
$ cd cerid-ai && ./scripts/start-cerid.sh