Using a development version of chachaml from another Clojure repo

You don't need to publish chachaml to a Maven repository to try it from another project. There are three reliable options depending on the build tool and how tightly you want the two repos coupled.

What's available right now

master covers M1 through M6 — the complete v0.1 MLOps loop:

• Run lifecycle: with-run, start-run!, end-run!
• Logging: log-params, log-param, log-metrics, log-metric
• Artifacts: log-artifact, log-file, load-artifact, list-artifacts
• Tracking macro: chachaml.tracking/deftracked — defn-shaped, wraps the body in with-run automatically
• Model registry (chachaml.registry): register-model, models, model, model-versions, promote!, load-model
• REPL helpers (chachaml.repl): runs-table, inspect, compare-runs, print-comparison
• Querying: runs, run, last-run
• Store binding: use-store!, with-store
• Default SQLite store at ./chachaml.db plus artifact directory at ./chachaml-artifacts/ (both auto-created)

Option 1 — :local/root in deps.edn (recommended)

In the consuming project's deps.edn, add chachaml as a local-root dependency:

{:deps
 {chachaml/chachaml {:local/root "/Users/jiriknesl/Documents/src/chachaml"}}}

Then from a REPL:

(require '[chachaml.core :as ml]
         '[chachaml.registry :as reg])

(ml/with-run {:experiment "demo"}
  (ml/log-params {:lr 0.01})
  (ml/log-metric :acc 0.91)
  (ml/log-artifact "model" {:weights [1.0 2.0] :bias 0.3})
  (reg/register-model "demo-classifier"
                      {:artifact "model" :stage :staging}))

(ml/last-run)
;; => {:id "…", :experiment "demo", :status :completed, …}

(:params (ml/run (:id (ml/last-run))))
;; => {:lr 0.01}

(reg/load-model "demo-classifier" {:stage :staging})
;; => {:weights [1.0 2.0] :bias 0.3}

(reg/promote! "demo-classifier" 1 :production)
(reg/load-model "demo-classifier")  ;; => latest production

REPL helpers

(require '[chachaml.repl :as repl])

(repl/runs-table)
;; ID          EXPERIMENT              NAME              STATUS      STARTED               DURATION
;; abc12345    linear-regression       synthetic         completed   2026-04-16 21:30:15   12.4s

(repl/inspect (:id (ml/last-run)))     ; full run summary
(repl/inspect "demo-classifier")        ; model summary by name

(repl/compare-runs (mapv :id (ml/runs)))
;; => {:runs [...] :params {...} :metrics {...}}
;;   each section split into :same / :differ / :partial

(repl/print-comparison (repl/compare-runs ["id1" "id2"]))

deftracked

(require '[chachaml.tracking :refer [deftracked]])

(deftracked train [config]
  (ml/log-params config)
  (let [model (do-training config)]
    (ml/log-metric :acc (eval-model model))
    (ml/log-artifact "model" model)
    model))

(train {:lr 0.01 :epochs 50})  ; auto-creates a run, completes it

Edits to chachaml's source are picked up the next time the consuming JVM is started (or after (require '[chachaml.core :as ml] :reload) in the REPL). No publish step.

Option 2 — lein install to your local Maven repo

If the consuming project uses Leiningen and you want to depend on a versioned snapshot:

cd /Users/jiriknesl/Documents/src/chachaml
lein install

This installs chachaml-0.1.0-SNAPSHOT.jar into ~/.m2/repository/. Then in the consumer's project.clj:

:dependencies [[chachaml "0.1.0-SNAPSHOT"]]

Re-run lein install after each chachaml change.

Option 3 — Leiningen checkouts (live source)

Leiningen supports a checkouts/ directory that symlinks into another project's source for live editing without lein install:

cd /path/to/consumer
mkdir -p checkouts
ln -s /Users/jiriknesl/Documents/src/chachaml checkouts/chachaml

You still need a placeholder dependency in project.clj:

:dependencies [[chachaml "0.1.0-SNAPSHOT"]]

Restart the JVM after symlinking. Subsequent edits to chachaml are visible immediately to the REPL.

A 30-second smoke test

Anywhere you have Clojure CLI installed:

mkdir -p /tmp/chachaml-smoke && cd /tmp/chachaml-smoke

cat > deps.edn <<EOF
{:deps {chachaml/chachaml {:local/root "/Users/jiriknesl/Documents/src/chachaml"}}}
EOF

clojure -M -e '
(require (quote [chachaml.core :as ml]))
(ml/with-run {:experiment "smoke"}
  (ml/log-params {:lr 0.1})
  (ml/log-metric :acc 0.9))
(prn (ml/last-run))
'

Expected: a printed run map with :status :completed, plus a fresh /tmp/chachaml-smoke/chachaml.db file.

To inspect the database directly:

sqlite3 /tmp/chachaml-smoke/chachaml.db \
  "SELECT id, experiment, status, name FROM runs;"

Choosing the store location

By default chachaml writes to ./chachaml.db in the consumer's working directory. To override:

(ml/use-store! {:path "/path/to/runs.db"})

For tests, use an in-memory store that disappears at JVM exit:

(ml/use-store! {:in-memory? true})

Or scope it locally:

(require '[chachaml.store.sqlite :as sqlite])

(ml/with-store (sqlite/open {:in-memory? true})
  (ml/with-run {} (ml/log-metric :acc 1.0))
  (ml/last-run))

Caveats while chachaml is pre-1.0

• Public API may change between milestones; check CHANGELOG.md before upgrading.
• The default store path is project-local. Add chachaml.db* and chachaml-artifacts/ to your consumer's .gitignore (the entries are already in chachaml's own .gitignore for reference).
• :in-memory? mode holds a single open Connection and is not thread-safe; it's intended for tests.