Tip: select any text on this page to bring up the Ask tooltip and question the Synedrix OS assistant about it.
What Qdrant is
Qdrant is a vector database. Its documentation frames the purpose as vector search that goes beyond keyword matching to find data based on semantic meaning, turning unstructured data into dense vector embeddings mapped into a high-dimensional space where similar items cluster.
The data model is small: a collection holds points, and a point is an id, one or more vectors, and a payload of arbitrary JSON metadata. Four distance metrics are supported, with cosine normalisation applied automatically at upload.
Payload filtering is the load-bearing feature. Conditions combine under must, should, and must_not operators that nest recursively, and filters are applied inside the HNSW traversal rather than before or after it.
What it is normally used for
How Qdrant is used generally, before Synedrix OS enters the picture.
- Semantic search: retrieval by meaning rather than keyword overlap
- Retrieval-augmented generation: supplying a model with grounded context at query time
- Hybrid search: combining semantic and lexical matching through sparse and dense vectors
- Recommendation and relevance feedback over the same vector space
- Multi-modal retrieval: separate image and text embeddings attached to one point
What it owns, and what it does not
Every platform in the stack has a non-overlapping responsibility. The boundary is as much a part of the design as the capability.
Owns
- Semantic retrieval collections
- Similarity search
- Metadata-aware filtered vector queries
- Pattern recall
Does not own
- Canonical relational truth: entities, state, approvals
- → Supabase PG17 (Layer 7)
- Explicit relationships: lineage and influence networks
- → Neo4j (Layer 7)
- Temporal truth: time-series metrics and telemetry
- → QuestDB (Layer 7)
- Ephemeral hot state and deduplication
- → Redis (Layer 7)
- LLM traces, prompt versions, cost attribution
- → Langfuse (Layer 10)
- Producing the embeddings themselves
- → Ollama / vLLM in Modes 2 and 3
Why Qdrant
Filters apply while searching, not after
Semantic search needs to respect boundaries like tenant or scope, not just similarity. Qdrant applies those filters during the search itself, rather than fetching too many results and filtering them afterwards, which would be slower and less accurate.
Fully self-hostable, with no licence to unlock
Running Synedrix OS self-hosted or air-gapped means every data store has to run locally with nothing gated behind a paid licence. Qdrant is Apache 2.0 licensed and runs as a single binary or container, meeting that requirement without exception.
No external dependency by default
A self-hosted Qdrant instance only ever talks to the systems calling it; nothing else. That's exactly what an air-gapped deployment requires.
Memory use can be tuned, not just accepted
On hardware with fixed memory, Qdrant's documented compression settings can be the difference between a collection fitting in memory or not, turning a hardware limit into something that can be tuned.
Qdrant questions, answered
Short, citable answers drawn from this page.
What role does Qdrant play in Synedrix OS?
Qdrant (Vector memory) is mapped to Layer 7: Memory & Data in Synedrix OS. Answers what is similar. Filtering happens inside the search, not around it.
What does Qdrant own in Synedrix OS?
Qdrant owns: Semantic retrieval collections; Similarity search; Metadata-aware filtered vector queries; Pattern recall.
What does Qdrant not own in Synedrix OS?
Qdrant does not own: Canonical relational truth: entities, state, approvals (owned by Supabase PG17 (Layer 7)); Explicit relationships: lineage and influence networks (owned by Neo4j (Layer 7)); Temporal truth: time-series metrics and telemetry (owned by QuestDB (Layer 7)); Ephemeral hot state and deduplication (owned by Redis (Layer 7)); LLM traces, prompt versions, cost attribution (owned by Langfuse (Layer 10)); Producing the embeddings themselves (owned by Ollama / vLLM in Modes 2 and 3).
The full reference page, with the operating model, deployment detail, and links to every canonical specification this platform touches, lives in the documentation repository, which is private while the platform is being built.