Privacy, compliance and security for your vector stores.
Know what your AI retrieves. VecShield inspects the content, permissions and retrieval behavior behind RAG applications—helping your team find sensitive data, review compliance evidence and investigate security risks.
An open-source Security & Compliance Console, with authorised red teaming for vector stores and RAG retrieval. Evidence for your team, not another opaque score.
Security & Compliance Console
Review sensitive-data detections, tenant-isolation issues and vector-store vulnerabilities in one place. Each finding shows the evidence and coverage behind it, so your team can decide what needs attention and track the next action.
RAG gives a language model context from your documents. Those documents are split into chunks, indexed in a vector store and retrieved when someone asks a question. Sensitive information, poisoned content and missing access controls can travel along that same path. VecShield examines the store and retrieval layer where those risks begin.
Sources
Documents, tickets, wikis
Ingestion
Chunk · embed · write
Vector store
Chunks, metadata, embeddings
Retrieval
Similarity search, filters
LLM
Answer generation
$ git clone https://github.com/Vecshield/vecshield $ cd vecshield/demo $ docker compose up --build
The demo includes a synthetic estate and the full console at 127.0.0.1:8088. A generated admin password is printed on first boot.
VecShield runs alongside the vector infrastructure you already use. Scans and findings stay inside your environment.
Stored prompt-injection indicators, suspicious chunks, provenance gaps and anomalous embeddings.
Personal, health, payment and credential data stored in chunks and associated metadata.
Checks whether one tenant or namespace can retrieve data belonging to another.
Network exposure, TLS, encryption settings, known vulnerable dependencies and configuration issues.
API keys, excessive permissions, stale credentials and access-control mismatches.
Bulk reads, enumeration attempts and unusual collection-wide retrieval patterns.
Content inspection
Inspect chunks for personal data, health information, payment details and secrets. Review redaction previews and check whether splitting content across chunks hides a sensitive value—before deciding how that content should be handled.
Finding sensitive data is the starting point. VecShield maps observations to relevant frameworks, including GDPR, HIPAA and PCI-DSS, to support evidence review—not to declare compliance automatically.
Explore compliance evidence →Research, experiments and practical notes on securing retrieval-augmented systems.
View all research →Retrieval Security10 min read
RAG security shouldn't start at the LLM. Why identity checks, tenant and classification filters, provenance and retention belong in the retrieval layer, before chunks ever reach the model.
Read →
Retrieval Security4 min read
Document permissions often disappear when content enters a vector store. Why RAG access controls, tenant isolation and least privilege need evidence, not assumptions.
Read →
RAG Poisoning6 min read
Indirect prompt injection and RAG poisoning turn the knowledge base itself into an attack surface. Why provenance and retrieval-layer visibility are security controls.
Read →
Authorised red teaming tests whether poisoned knowledge can influence an answer, whether one tenant can retrieve another's data, and whether queries can extract a collection. These security checks complement sensitive-data discovery and compliance evidence review.
Attacker-controlled content
RAG ingestion
Poisoned knowledge base
Malicious / incorrect retrieval
LLM
Tenant A query
One tenant's request
Vector store
Shared collection
Tenant B chunk
Cross-tenant data exposure
Repeated semantic queries
Progressive recovery
Content collected across queries
Stored knowledge
Collection contents extracted
VecShield focuses on the retrieval and vector-store layer. It is not a generic LLM jailbreak detector.
Read the risk model →
Chroma
pgvector
Pinecone
Qdrant
Weaviate
Milvus
Redis
OpenSearch
MongoDB
OpenAI vector store
Azure AI Search
Amazon Bedrock Knowledge Bases
Vertex AIConnectors read from your vector stores. Optional monitoring records retrieval events. Detectors turn both into findings for data exposure, RAG integrity and access posture.
VecShield runs in your environment and does not require sending vector-store contents to a VecShield-hosted service. Scans run against the store on their own schedule, not between your application and its queries.
Read the architecture →Vector stores
Connectors
Read-only inspection
Retrieval events
Monitoring
Optional event capture
Detectors
Data exposure · RAG integrity · access posture
Findings & evidence
Coverage, masked previews and audit records
A result states what was scanned, what was sampled, what was skipped, what evidence supports a finding, what evidence argues against it, and what the detector cannot conclude. When a check samples a collection instead of reading all of it, the result is marked partial. There is no overall score.
Detection is not a legal determination. VecShield produces evidence; humans make the final decision.
