feat: switch default Ollama model to granite4:350m

granite4:350m is ~676 MB on disk vs ~1.5 GB for granite3.3:2b, making
the bundled provider far quicker to bootstrap on a fresh machine while
still supporting tool calling.

- ollama_bootstrap.sh: LANGFLOW_OLLAMA_MODEL default
- docker_example/ollama-granite: ollama-init pull command, README
- ollama_constants.py: catalog entry replaced
- Makefile: ollama_up help text

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: Empreiteiro

Makefile

@@ -233,7 +233,7 @@ lint: install_backend ## run linters
 
 
 
-ollama_up: ## start a local Ollama with granite3.3:2b for the bundled model provider
+ollama_up: ## start a local Ollama with granite4:350m for the bundled model provider
 	@bash scripts/setup/ollama_bootstrap.sh
 
 ollama_down: ## stop and remove the bundled Ollama container (keeps the model volume)

docker_example/ollama-granite/README.md

@@ -1,6 +1,6 @@
-# Langflow + Ollama (granite3.3:2b) preset
+# Langflow + Ollama (granite4:350m) preset
 
-Turnkey stack: Langflow, Postgres, and an Ollama server with `granite3.3:2b` pre-pulled. The Langflow service sees `OLLAMA_BASE_URL=http://ollama:11434`, so the Ollama provider is auto-enabled in **Model Providers** on first login.
+Turnkey stack: Langflow, Postgres, and an Ollama server with `granite4:350m` pre-pulled. The Langflow service sees `OLLAMA_BASE_URL=http://ollama:11434`, so the Ollama provider is auto-enabled in **Model Providers** on first login.
 
 ## Run
 
@@ -14,14 +14,14 @@ Or, with Podman:
 podman compose up        # podman-compose / podman compose plugin
 ```
 
-Open http://localhost:7860. The first start downloads the model (~1.5 GB), so allow a few minutes.
+Open http://localhost:7860. The first start downloads the model (a few hundred MB for `granite4:350m`).
 
 ## Services
 
 - **langflow** — `localhost:7860`, with `OLLAMA_BASE_URL` already set
 - **postgres** — `localhost:5432` (`langflow/langflow`)
 - **ollama** — `localhost:11434`
-- **ollama-init** — one-shot, pulls `granite3.3:2b` then exits
+- **ollama-init** — one-shot, pulls `granite4:350m` then exits
 
 ## GPU
 

docker_example/ollama-granite/docker-compose.yml

@@ -58,7 +58,7 @@ services:
     environment:
       - OLLAMA_HOST=http://ollama:11434
     command: >
-      "ollama pull granite3.3:2b && echo 'granite3.3:2b ready'"
+      "ollama pull granite4:350m && echo 'granite4:350m ready'"
     restart: "no"
 
 volumes:

scripts/setup/ollama_bootstrap.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# Bootstrap a local Ollama server with the granite3.3:2b model so that
+# Bootstrap a local Ollama server with the granite4:350m model so that
 # `make run_cli` exposes a pre-configured Ollama provider in Langflow.
 #
 # Resolution order:
@@ -16,7 +16,7 @@
 # Environment variables:
 #   OLLAMA_PORT                 default 11434
 #   OLLAMA_HOST                 default 127.0.0.1
-#   LANGFLOW_OLLAMA_MODEL       default granite3.3:2b
+#   LANGFLOW_OLLAMA_MODEL       default granite4:350m
 #   LANGFLOW_OLLAMA_ENGINE      override container engine: docker | podman
 #                               (auto-detected when unset)
 #   LANGFLOW_OLLAMA_NO_INSTALL  set to 1 to disable brew autoinstall
@@ -28,7 +28,7 @@ set -euo pipefail
 
 OLLAMA_PORT="${OLLAMA_PORT:-11434}"
 OLLAMA_HOST="${OLLAMA_HOST:-127.0.0.1}"
-OLLAMA_MODEL="${LANGFLOW_OLLAMA_MODEL:-granite3.3:2b}"
+OLLAMA_MODEL="${LANGFLOW_OLLAMA_MODEL:-granite4:350m}"
 CONTAINER_NAME="langflow-ollama"
 VOLUME_NAME="langflow-ollama-data"
 HEALTH_TIMEOUT_S=60

src/lfx/src/lfx/base/models/ollama_constants.py

@@ -137,7 +137,7 @@
     ),
     create_model_metadata(
         provider="Ollama",
-        name="granite3.3:2b",
+        name="granite4:350m",
         icon="Ollama",
         tool_calling=True,
     ),