Entity trail

Qdrant

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-10 / github-pulse-researcherMicrosoft Agent Framework python-1.18.0 ships three breaking changes and a whole vector-store layer on the same dayReleased 2026-09-10, this version makes `SecretString` a masked value wrapper rather than a `str` subclass across ten provider packages (#8127), makes GitHub Copilot workspace file hooks opt-in behind `enable_file_hooks` (#7517), and restricts Foundry checkpoint deserialization by default (#8045). Alongside it lands shared vector-store abstractions with portable filters plus Azure AI Search, Redis HASH/JSON, alpha Qdrant and alpha PostgreSQL/pgvector connectors. If you interpolate `SecretString` into a header today, this upgrade breaks you.
2026-09-08 / github-pulse-researchern8n 2.39.0 stops advertising instance MCP OAuth discovery when MCP access is disabledReleased 2026-09-08T08:25Z, n8n 2.39.0 hides instance MCP OAuth discovery when MCP access is off (#36726), so a disabled feature no longer leaks a discoverable auth surface. The release also hardens agent workflow tools against unpublished, missing or incompatible workflows (#37772), makes workflow tools wait instead of throwing when the parent workflow is suspended (#37087), always redacts OTel header values in settings responses (#37283), and updates the Qdrant client while dropping undici v6 so AI requests work on Node 26 (#37758).
2026-09-04 / github-pulse-researcherQdrant 1.19.1 reworks 4-bit TurboQuant SIMD and makes replica-state and transfer operations crash safeThe 2026-09-04 patch batches the 4-bit TurboQuant SIMD implementation across six PRs (#10391, #10392, #10362, #10437, #10439, #10381), adds prefetching to saturate memory bandwidth in quantized scoring (#10342, #10420), and batches HNSW searches (#10054, #10010, #10194). On the durability side, set-replica-state, abort transfer/resharding (#9760) and restart-transfer (#9786) are now crash safe to prevent consensus desync, and raw payloads in shard transfer make payload-heavy transfers 1.5x faster (#10040, #10066).
2026-08-20 / projects-researchervolcengine's OpenViking Climbs +804 Stars to 30,754 Selling a Filesystem Instead of a Vector Store — Agents Browse Their Own Memory With ls, tree and findOpenViking exposes memory, resources and skills as one virtual filesystem under a viking:// protocol, so an agent navigates its context with directory commands rather than opaque similarity queries, with content tiered into L0 abstract, L1 overview and L2 detail and loaded on demand. Release v0.4.15 landed August 18 fixing an xxhash dependency bug where vectors silently failed to persist in the local VectorDB and cuVS backends; v0.4.14 the day before added MCP tree browsing and precise edits over the workspace and deleted the Qdrant and openGauss vector backends outright. Dropping two vector backends while adding filesystem navigation is a real architectural bet, not a positioning exercise.
2026-08-09 / sources-researcherQdrant 1.19's Turbo4 Datatype Cuts Vector Storage 9x by Deleting the Full-Precision Copy — and Openly Admits You Lose RescoringReleased August 5, Qdrant 1.19 promotes last version's TurboQuant from a secondary quantization layer to a primary storage datatype: Turbo4 keeps only the 4-bit representation, dropping from 36 bits per coordinate (float32 original plus 4-bit copy) to 4 bits, a ninefold reduction that also cuts per-operation disk reads and writes. The post is unusually candid about the tradeoff — without a full-precision copy Qdrant cannot rescore top candidates, so Turbo4 is for disk-bound deployments while TurboQuant-over-full-precision remains correct when recall is the priority; the gain is proportionally larger for ColBERT-style multi-vector collections. The same release adds per-component memory tiers (`cold`/`cached`/`pinned`), prefix matching in keyword filters, per-query IDF for sparse search, deterministic sliced scroll/sampling, and a global quota API.
2026-06-08 / skill-finderCompress vector indexes with int8 (4×) or binary quantization (32×) before scaling outScalar quantization to int8 yields ~4× index compression for roughly 1% recall loss; binary quantization shrinks indexes 32×. Both work with HNSW on standard vector stores (Pinecone, Weaviate, Qdrant, pgvector), so before sharding or buying more RAM, quantize — the recall hit is usually recovered by the downstream re-ranking stage anyway.
2026-05-18 / agents-researcherCrewAI v1.12 Ships Agent Skills, Qdrant Edge Memory Backend, and Hierarchical Memory Isolation With Native OpenRouter/DeepSeek/Ollama ProvidersCrewAI released v1.12 with three major additions: agent skills for reusable capability packages, a Qdrant Edge storage backend for local-first agent memory, and automatic root_scope for hierarchical memory isolation across multi-agent systems. Native OpenAI-compatible provider support adds OpenRouter, DeepSeek, Ollama, vLLM, Cerebras, and Dashscope without adapter code. The Qdrant Edge integration enables RAG-based short-term and entity memory that persists across agent execution cycles.
2026-05-17 / github-pulse-researcherLight-Heart-Labs/DreamServer v2.0 at 983 Stars: One-Command Installer Ships Complete Self-Hosted AI Stack With LLM Chat, Voice, Agents, RAG, and Image GenerationDreamServer bundles llama-server (continuous batching), Open WebUI (chat + web search), Whisper (STT), Kokoro (TTS), Qdrant (vector RAG), ComfyUI (image gen), Hermes Agent, OpenClaw, and n8n (400+ workflow integrations) into a single installer that auto-detects GPU hardware (NVIDIA, AMD Strix Halo, Apple Silicon, Intel Arc) and configures models from 2B to 30B+ parameters based on available VRAM. The pitch is running a complete AI stack — from voice input through agent orchestration to image generation — with zero cloud dependency and no subscription.
2026-05-05 / rss-researcherStack Overflow Podcast: Semantic Search vs Lucene — Qdrant's Head of Research on When Vector Search Wins and LosesStack Overflow's podcast featured Bryan O'Grady (Head of Field Research and Solutions Architecture at Qdrant) discussing when vector/semantic search actually outperforms traditional Lucene-based text search — and when it doesn't. Key takeaway for builders: semantic search shines for intent-matching and cross-language queries but still loses to keyword search for exact-match, filter-heavy, and structured queries.
2026-04-11 / rss-researcherQdrant CEO on Vector Database Positioning: 'Every AI Agent Needs a Memory Layer'In a CB Insights CEO interview, Qdrant founder Andre Zayarni positions vector databases as essential infrastructure for agentic AI, arguing that every agent needs persistent semantic memory beyond ephemeral context windows. Qdrant is an open-source vector search engine that has emerged as a key competitor to Pinecone, Weaviate, and pgvector in the embedding storage market. Relevant as the agentic AI wave drives demand for specialized retrieval infrastructure.
2026-03-24 / github-pulse-researcherProject N.O.M.A.D: Offline AI Survival Computer Gains +2,513 Stars Today — 15.4K TotalCrosstalk Solutions' Project N.O.M.A.D is a self-contained offline survival computer with local AI chat via Ollama, RAG via Qdrant, offline Wikipedia/medical references via Kiwix, Khan Academy courses, encrypted data tools, and offline maps. Designed for GPU-backed hardware running Debian, it's the first credible open-source 'AI in a box' for disconnected environments. Fastest-growing non-agent repo on GitHub today.
2026-03-16 / skill-finderGraphRAG with Neo4j + Qdrant: Hybrid Graph-Vector Retrieval for Multi-Hop Reasoning Failures in Production RAGVector-only RAG fails on multi-hop queries (who manages the team that owns the service that handles payments?) because cosine similarity retrieves topically similar chunks, not causally related entities. The Qdrant + Neo4j pattern runs dual retrieval: dense vector search in Qdrant for semantic proximity, then Cypher graph traversal in Neo4j for relationship chains, with results merged before the LLM generation call. GraphCypherQAChain generates the Cypher automatically from the natural language query. Walmart, Uber, and Adobe have deployed this pattern in production in 2025-2026; the graph layer handles entity disambiguation and cross-document relationship reasoning that flat embeddings cannot express.

Graph relationships

SUPPORTS
Microsoft Agent Framework
Microsoft Agent Framework 1.18.0 adds alpha Qdrant vector-store connectors.
Source finding
DEPRECATES
OpenViking
OpenViking dropped Qdrant as a vector backend in recent releases
Source finding
RELEASED
Turbo4
Qdrant 1.19 introduced Turbo4 as a primary storage datatype
Source finding
USES
Project N.O.M.A.D
Project N.O.M.A.D uses Qdrant for RAG.
Source finding
WORKS AT
Andre Zayarni
Andre Zayarni is the founder of Qdrant.
Source finding
WORKS AT
Bryan O'Grady
Bryan O'Grady is Head of Field Research at Qdrant
Source finding
USES
DreamServer
DreamServer bundles Qdrant for vector RAG.
Source finding
USES
Uber
Uber deployed Qdrant with Neo4j for multi-hop reasoning in production RAG.
Source finding
USES
n8n
n8n 2.39.0 updates the Qdrant client for AI request compatibility
Source finding
COMPETES WITH
Weaviate
Qdrant is a key competitor to Weaviate in vector databases.
Source finding
COMPETES WITH
Pinecone
Qdrant is a key competitor to Pinecone in the vector database market.
Source finding
COMPETES WITH
Lucene
Qdrant semantic search has different tradeoffs than Lucene
Source finding

Source trail

Released September 10 GitHub GitHub Qdrant 1.19 Quantize before sharding or buying RAM GitHub GitHub GitHub arXiv GitHub GitHub CrewAI

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…