verbose live UI + tool-level SSE events + Groq default + regression tests

Makefile

@@ -1,4 +1,4 @@
-.PHONY: kind tilt-up tilt-down seed evals \
+.PHONY: kind tilt-up tilt-down seed seed-fetch seed-push evals test \
         compose-up compose-down compose-clean compose-seed compose-evals \
         docs-graphs
 
@@ -17,11 +17,29 @@ tilt-up:
 tilt-down:
 	cd ctrl && tilt down $(KCTX)
 
-seed:
+seed-fetch:
+	bash ctrl/seed/download.sh
+
+# Copy the downloaded SQLite into the api pod. The file is excluded from the
+# docker build context (.dockerignore) to keep the image small, so we push
+# it explicitly the first time.
+seed-push:
+	@POD=$$(kubectl $(KCTX) $(KNS) get pod -l app=api -o jsonpath='{.items[0].metadata.name}'); \
+	echo "pushing financial.sqlite → $$POD"; \
+	kubectl $(KCTX) $(KNS) exec $$POD -- mkdir -p /app/seed/data; \
+	kubectl $(KCTX) $(KNS) cp ctrl/seed/data/financial.sqlite $$POD:/app/seed/data/financial.sqlite
+
+seed: seed-fetch seed-push
 	kubectl $(KCTX) $(KNS) exec deploy/api -- uv run python -m seed.load_bird
 
 evals:
-	kubectl $(KCTX) $(KNS) exec deploy/api -- uv run python -m evals.run_evals
+	kubectl $(KCTX) $(KNS) exec deploy/api -- uv run python -m api.evals.run_evals
+
+# Unit tests run on the host venv (no postgres/llm calls needed).
+# Ensures dev extras are installed first; uv is idempotent on no-ops.
+test:
+	uv sync --extra dev
+	uv run pytest -q
 
 # ── docker-compose (alternative path) ──────────────────────
 
@@ -34,11 +52,11 @@ compose-down:
 compose-clean:
 	$(COMPOSE) down -v
 
-compose-seed:
+compose-seed: seed-fetch
 	$(COMPOSE) exec api uv run python -m seed.load_bird
 
 compose-evals:
-	$(COMPOSE) exec api uv run python -m evals.run_evals
+	$(COMPOSE) exec api uv run python -m api.evals.run_evals
 
 # ── docs ───────────────────────────────────────────────────