Clojurians Log v2

oh shit, i never put it together that you created yaml, no wonder you're such a wiz at yaml

> HEY EVERYONE!!! I CREATED YAML!!! HOLY SHIT!!! Not really my style. Maybe it should be 🤷

I got some recruiter spam a while back that said I was in the top 1% of Clojure code authors in the world.

If it's true, it's gotta be because I write so much bad Clojure. 🙂

hah you have it in your profile so it's on me for not noticing before

You mean "bad ass" Clojure, I'm sure.

aww

We were a bit red-faced in #clj-yaml... we were getting into a bit of a YAML-diss session when @ingy dropped in... you were very cool about it, Ingy! ❤️

I recall that. I never even bothered to read your disses. I've seen my share.

btw I don't think there were any Clojure YAMLs on https://yaml.org/libraries/ Now we have 3 🙂

YAMLScript ships its Clojure and Java assets to Clojars. But YAMLStar ships its Java to Maven Central... That was a bit of a pain, but all sorted/automated now. I recall asking about how to do this once in #clojure and iirc we ended up realizing that (likely) of all the Clojurists here, only @alexmiller does that. Now there are 2! 😄

Oh. YAMLStar works on Windows (YAMLScript doesn't yet)

And also on Fortran 😄

Also check out this new YAML parser comparison site: https://play.yaml.com/

Hi, I'm using eca-emacs and want to stay with it, I need only memory function (chroma) so that I don't have to repeat myself again and again, is there a way to setup just the memory part?

Hello, Mateusz!

Yes, you could. In the docker-compose.yaml there are multiple dependencies. One of them is chroma. You could just up that one or milvus. Start an emacs session (could be deamon mode) -- you won't be using the multi-agent setup. E.g., you don't need to touch emacs if you want. Then, start hive-mcp using start-mcp.sh the start-bb-mcp.sh would need you to clone somewhere the bb-cmp (which would enable multiplexing the JVM cost down on new claude sessions). But if you are strictly using an agent, you could use start-mcp.sh as entrypoint. It's the setup where you spawn the heavy-JVM etc. I would advise cloning the bb-mcp anyway and then using:

# Required: Paths to repositories (adjust to your clone locations)
export HIVE_MCP_DIR="$HOME/projects/hive-mcp"
export BB_MCP_DIR="$HOME/projects/bb-mcp"

# Register hive-mcp as "hive" MCP server
claude mcp add hive --scope user -- \
  $HIVE_MCP_DIR/start-bb-mcp.sh

;;; hive-mcp minimal configuration

;; 1. Add hive-mcp to load-path (REQUIRED)
;; The order matters: elisp/ BEFORE elisp/addons/
(add-to-list 'load-path "/path/to/hive-mcp/elisp")
(add-to-list 'load-path "/path/to/hive-mcp/elisp/addons")

;; 2. Install required packages (if not present)
(unless (package-installed-p 'websocket)
  (package-refresh-contents)
  (package-install 'websocket))

;; 3. Load core module (REQUIRED)
(require 'hive-mcp)

;; 4. Enable the global minor mode (REQUIRED)
(hive-mcp-mode 1)

;; 5. Start Emacs server for emacsclient (REQUIRED)
;; Idempotent: safe to call multiple times
(unless (server-running-p)
  (server-start))

docker compose up chroma -d

You could also ask her, of couse, lookup memory for. When I explicitly want her to explore something, but use memory first, currently I do:

<so and so> (lookup memories)

Oh, I forgot. But, you either/or need to run ollama nomic-embed-text:latest model, or setup the OPENROUTER_API_KEY and use it.

When using OpenRouter, the default embedding model is qwen/qwen3-embedding-8b (4096 dimensions, 33k context, free tier).

  Supported OpenRouter Embedding Models
  ┌────────────────────────────────┬────────────┬────────────────────┐
  │             Model              │ Dimensions │       Notes        │
  ├────────────────────────────────┼────────────┼────────────────────┤
  │ qwen/qwen3-embedding-8b        │ 4096       │ Default, free tier │
  ├────────────────────────────────┼────────────┼────────────────────┤
  │ openai/text-embedding-3-small  │ 1536       │ Paid               │
  ├────────────────────────────────┼────────────┼────────────────────┤
  │ openai/text-embedding-3-large  │ 3072       │ Paid               │
  ├────────────────────────────────┼────────────┼────────────────────┤
  │ cohere/embed-english-v3.0      │ 1024       │ Paid               │
  ├────────────────────────────────┼────────────┼────────────────────┤
  │ cohere/embed-multilingual-v3.0 │ 1024       │ Paid               │
  └────────────────────────────────┴────────────┴────────────────────┘
  Configuration

  From src/hive_mcp/embeddings/openrouter.clj:110-129:

  ;; Default (uses OPENROUTER_API_KEY env var)
  (openrouter/->provider)

  ;; Explicit model
  (openrouter/->provider {:model "openai/text-embedding-3-small"})

  The API key is read from OPENROUTER_API_KEY environment variable by default.

I use a split solution, for memories I use nomic-embed-text:latest (4K context) and for presets (bigger sized text, in general) I use the qwen/qwen3-embedding-8b (supports up to 32K context) It's not free, btw. It's just cheap to use the embedding qwen model - the information above is inaccurate, regarding this.

I created #hive-mcp because I will be updating it frequently there. May even hook a way to auto-release notes on what was worked on through versions eventually. Anyone that have questions or need help setting it up. We can schedule a call and record it to share with others interested. I'm up for it. 😁

I am happy to discuss this in #datahike, including how to expose it through its versioning bindings.