self-host without docker on Macos

[tne-ops]

I know you don't support non-docker instances, but on MacOS, docker has a bit of a penalty (particularly when running large models where the fix allocation to a linux VM is a bit painful).

But I worked out how to get it to install on MacOS, happy to provide a README.md (or do you not want it because it gets people in trouble?). I did a branch from your main that I'll maintain for Makefile based off of master for now, but can share if interested

.PHONY: install run clean help kill stop

# Configuration
PORT ?= 6333

# Default target
## help: Show this help message.
help:
	@sed -n 's/^##//p' $(MAKEFILE_LIST) | column -t -s ':' |  sed -e 's/^/ /'

## install: Install dependencies and build Qdrant locally
install: build
	brew install rustup
	rustup component add rustfmt
	brew install protobuf
	cargo build --release --bin qdrant
	./tools/sync-web-ui.sh

## kill/stop: Stop running Qdrant server
kill stop:
	pkill qdrant || echo "No Qdrant server was running"

## run: Run Qdrant locally (PORT=6333 by default)
run: kill
	QDRANT__HTTP_PORT=$(PORT) ./target/release/qdrant
	@echo "Web UI available at: http://localhost:$(PORT)/dashboard"

## clean: Clean build artifacts
clean:
	cargo clean

[tne-ops]

Oh this is so helpful. Thank you so much. We are doing the same thing, we want a very small fast lightweight pipelines that can come up in ms. Qdrant is so awesome. Please Linke to the Makefil mods and show. It's all part of open source, we all give a little so we can all get better together. Yes love to see the mods and how this works and happy to contribute ours as well. Edge deployments are so key and these chunk drift and so forth is so useful. Please let us know the links, we are happy to contribute out knowledge and draft a PR if appropriate for the rest of the Qdrant community :-) Sent via Superhuman ( ***@***.*** )

…

On Tue, Jul 29, 2025 at 1:53 PM, PSBigBig < ***@***.*** > wrote: Hey! Great to see others navigating non-Docker MacOS setups — I’ve also run into similar friction when trying to integrate Qdrant into lightweight RAG pipelines that need native performance. Actually, we’ve recently mapped out a bunch of production-grade LLM failure modes (like #1 Hallucination & Chunk Drift ( https://github.com/onestardao/WFGY/blob/main/ProblemMap/README.md ) ) that often show up during edge deployments — and found that tweaking embedding behavior and memory handoff matters more than people expect. Happy to link you to the exact Makefile mods or show how we manage non-containerized setups in our Reasoning Engine if useful. Let me know — and kudos for posting this, this will help way more people than it looks! — Reply to this email directly, view it on GitHub ( #6943 (comment) ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/ATK6MOPSATLC2UDRP4Q7ZAD3K4D7FAVCNFSM6AAAAACCOK2IRKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTGOJRG44TCOI ). You are receiving this because you authored the thread. Message ID: <qdrant/qdrant/repo-discussions/6943/comments/13917919 @ github. com> -- You received this message because you are subscribed to the Google Groups "Architects and Developers" group. To unsubscribe from this group and stop receiving emails from it, send an email to dev+unsubscribe@ tne. ai ( ***@***.*** ). To view this discussion visit https:/ / groups. google. com/ a/ tne. ai/ d/ msgid/ dev/ qdrant/ qdrant/ repo-discussions/ 6943/ comments/ 13917919%40github. com ( https://groups.google.com/a/tne.ai/d/msgid/dev/qdrant/qdrant/repo-discussions/6943/comments/13917919%40github.com?utm_medium=email&utm_source=footer ).

[onestardao]

Haha I was this close to believing I was truly off-topic
(But then again… maybe sharing non-Docker RAG setup insights was too spicy for GitHub to handle.)

Anyway—super appreciate your thoughtful reply.
Totally hear you on needing ultra-lightweight, fast RAG pipelines — especially with Qdrant, where chunk behavior and handoff timing matter a lot during edge deployments.

As you already found the link I shared above (I won’t repost it, or I’ll get kicked out for Off-Topic v2 😅), that map includes not just chunk-related pitfalls like No.1 Hallucination & Chunk Drift, but also some Makefile tricks we used internally to patch performance + embedding tweaks without containers.

Happy to swap notes or walk you through the setup if helpful! Let’s build fast and sane LLM stacks together.