Skip to content

Personal AI, On Personal Devices

Python

feat: auto-pull all Qwen3.5 models that fit on user's hardware

Jon Saad-Falcondfb66625a79c
diff --git a/desktop/src-tauri/Cargo.lock b/desktop/src-tauri/Cargo.lock
index ba5f0965..55112ab4 100644
--- a/desktop/src-tauri/Cargo.lock
+++ b/desktop/src-tauri/Cargo.lock
@@ -2573,7 +2573,7 @@ dependencies = [
[[package]]
name = "openjarvis-desktop"
-version = "1.0.0"
+version = "0.1.0"
dependencies = [
"reqwest 0.12.28",
"serde",
diff --git a/desktop/src-tauri/src/lib.rs b/desktop/src-tauri/src/lib.rs
index d1f26918..4ea87c43 100644
--- a/desktop/src-tauri/src/lib.rs
+++ b/desktop/src-tauri/src/lib.rs
@@ -8,7 +8,75 @@ use tokio::sync::Mutex;
const OLLAMA_PORT: u16 = 11434;
const JARVIS_PORT: u16 = 8222;
-const DEFAULT_MODEL: &str = "qwen3:0.6b";
+
+/// Qwen3.5 model variants, ordered smallest to largest.
+/// Each entry is (ollama_tag, approximate_download_size_gb, min_ram_gb).
+/// min_ram_gb includes ~2GB headroom above model size for OS/runtime.
+const QWEN35_MODELS: &[(&str, f64, f64)] = &[
+ ("qwen3.5:0.8b", 1.0, 4.0),
+ ("qwen3.5:2b", 2.7, 6.0),
+ ("qwen3.5:4b", 3.4, 8.0),
+ ("qwen3.5:9b", 6.6, 12.0),
+ ("qwen3.5:27b", 17.0, 24.0),
+ ("qwen3.5:35b", 24.0, 32.0),
+ // 122b needs ~81GB, only for 96GB+ machines
+ ("qwen3.5:122b", 81.0, 96.0),
+];
+
+/// Fallback model if we cannot determine RAM or no Qwen3.5 fits.
+const FALLBACK_MODEL: &str = "qwen3:0.6b";
+
+/// Get total system RAM in GB.
+fn total_ram_gb() -> f64 {
+ #[cfg(target_os = "macos")]
+ {
+ use std::process::Command;
+ if let Ok(output) = Command::new("sysctl").args(["-n", "hw.memsize"]).output() {
+ if let Ok(s) = String::from_utf8(output.stdout) {
+ if let Ok(bytes) = s.trim().parse::<u64>() {
+ return bytes as f64 / (1024.0 * 1024.0 * 1024.0);
+ }
+ }
+ }
+ }
+ #[cfg(target_os = "linux")]
+ {
+ if let Ok(contents) = std::fs::read_to_string("/proc/meminfo") {
+ for line in contents.lines() {
+ if line.starts_with("MemTotal:") {
+ if let Some(kb_str) = line.split_whitespace().nth(1) {
+ if let Ok(kb) = kb_str.parse::<u64>() {
+ return kb as f64 / (1024.0 * 1024.0);
+ }
+ }
+ }
+ }
+ }
+ }
+ // Fallback: assume 8GB (conservative)
+ 8.0
+}
+
+/// Return the list of Qwen3.5 models that fit on this machine, smallest first.
+fn models_that_fit() -> Vec<&'static str> {
+ let ram = total_ram_gb();
+ QWEN35_MODELS
+ .iter()
+ .filter(|(_, _, min_ram)| ram >= *min_ram)
+ .map(|(tag, _, _)| *tag)
+ .collect()
+}
+
+/// Return the largest Qwen3.5 model that fits, or the fallback.
+fn best_default_model() -> &'static str {
+ let ram = total_ram_gb();
+ QWEN35_MODELS
+ .iter()
+ .rev()
+ .find(|(_, _, min_ram)| ram >= *min_ram)
+ .map(|(tag, _, _)| *tag)
+ .unwrap_or(&FALLBACK_MODEL)
+}
/// Resolve full path to a binary by checking common locations.
/// macOS .app bundles don't inherit the shell PATH, so we probe manually.
@@ -212,7 +280,13 @@ async fn ollama_has_model(model: &str) -> bool {
return models.iter().any(|m| {
m.get("name")
.and_then(|n| n.as_str())
- .map(|n| n.starts_with(model.split(':').next().unwrap_or(model)))
+ .map(|n| {
+ // Exact match on full tag, or match without
+ // Ollama's ":latest" suffix
+ n == model
+ || n.strip_suffix(":latest") == Some(model)
+ || model.strip_suffix(":latest") == Some(n)
+ })
.unwrap_or(false)
});
}
@@ -285,19 +359,26 @@ async fn boot_backend(backend: SharedBackend, status: SharedStatus) {
s.detail = "Inference engine ready.".into();
}
- // Phase 2: Ensure a default model exists
+ // Phase 2: Ensure models are available
+ let fitting = models_that_fit();
+ let default_model = best_default_model();
+
+ // Pull the smallest model first so the app becomes usable quickly.
+ // Then the server starts while remaining models download in the background.
+ let first_model = fitting.first().copied().unwrap_or(FALLBACK_MODEL);
+
{
let mut s = status.lock().await;
s.phase = "model".into();
- s.detail = format!("Checking for model {}...", DEFAULT_MODEL);
+ s.detail = format!("Checking for model {}...", first_model);
}
- if !ollama_has_model(DEFAULT_MODEL).await {
+ if !ollama_has_model(first_model).await {
{
let mut s = status.lock().await;
- s.detail = format!("Downloading {}... (this may take a minute)", DEFAULT_MODEL);
+ s.detail = format!("Downloading {}... (this may take a minute)", first_model);
}
- if let Err(e) = pull_model(DEFAULT_MODEL).await {
+ if let Err(e) = pull_model(first_model).await {
let mut s = status.lock().await;
s.error = Some(format!("Failed to download model: {}", e));
return;
@@ -310,6 +391,13 @@ async fn boot_backend(backend: SharedBackend, status: SharedStatus) {
s.detail = "Model ready.".into();
}
+ // Collect remaining models to pull in the background after the server starts.
+ let remaining_models: Vec<&str> = fitting
+ .iter()
+ .filter(|m| **m != first_model)
+ .copied()
+ .collect();
+
// Phase 3: Start jarvis serve
{
let mut s = status.lock().await;
@@ -332,7 +420,12 @@ async fn boot_backend(backend: SharedBackend, status: SharedStatus) {
}
let mut cmd = tokio::process::Command::new(&uv_bin);
- cmd.args(["run", "jarvis", "serve", "--port", &JARVIS_PORT.to_string(), "--agent", "simple"])
+ cmd.args([
+ "run", "jarvis", "serve",
+ "--port", &JARVIS_PORT.to_string(),
+ "--model", default_model,
+ "--agent", "simple",
+ ])
.stdout(std::process::Stdio::null())
.stderr(std::process::Stdio::null())
.current_dir(project_root.as_ref().unwrap());
@@ -368,6 +461,19 @@ async fn boot_backend(backend: SharedBackend, status: SharedStatus) {
s.phase = "ready".into();
s.detail = "All systems ready.".into();
}
+
+ // Phase 4: Pull remaining Qwen3.5 models in the background.
+ // The server is already running with the first model, so the user
+ // can start chatting immediately while larger models download.
+ if !remaining_models.is_empty() {
+ tokio::spawn(async move {
+ for model in remaining_models {
+ if !ollama_has_model(model).await {
+ let _ = pull_model(model).await;
+ }
+ }
+ });
+ }
}
// ---------------------------------------------------------------------------
diff --git a/frontend/package-lock.json b/frontend/package-lock.json
index fb450ed7..697dcc97 100644
--- a/frontend/package-lock.json
+++ b/frontend/package-lock.json
@@ -1,12 +1,12 @@
{
"name": "openjarvis-chat",
- "version": "1.0.0",
+ "version": "0.1.0",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "openjarvis-chat",
- "version": "1.0.0",
+ "version": "0.1.0",
"dependencies": {
"@tailwindcss/vite": "^4.2.1",
"@tauri-apps/api": "^2",