diff --git a/Cargo.toml b/Cargo.toml
index 9e3894af..6456c21a 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -18,7 +18,7 @@ members = [
 ]
 
 [workspace.package]
-version = "0.4.0"
+version = "0.4.1"
 edition = "2021"
 license = "Apache-2.0 OR MIT"
 repository = "https://github.com/RightNow-AI/openfang"
diff --git a/crates/openfang-api/src/routes.rs b/crates/openfang-api/src/routes.rs
index 68e72867..a132f6d1 100644
--- a/crates/openfang-api/src/routes.rs
+++ b/crates/openfang-api/src/routes.rs
@@ -7007,8 +7007,8 @@ pub async fn set_provider_key(
                 "\n[default_model]\nprovider = \"{}\"\nmodel = \"{}\"\napi_key_env = \"{}\"\n",
                 name, model_id, env_var
             );
+            backup_config(&config_path);
             if let Ok(existing) = std::fs::read_to_string(&config_path) {
-                // Remove existing [default_model] section if present, then append
                 let cleaned = remove_toml_section(&existing, "default_model");
                 let _ = std::fs::write(&config_path, format!("{}\n{}", cleaned.trim(), update_toml));
             } else {
@@ -10839,6 +10839,12 @@ pub async fn auth_check(
 }
 
 /// Remove a `[section]` and its contents from a TOML string.
+#[allow(dead_code)]
+fn backup_config(config_path: &std::path::Path) {
+    let backup = config_path.with_extension("toml.bak");
+    let _ = std::fs::copy(config_path, backup);
+}
+
 fn remove_toml_section(content: &str, section: &str) -> String {
     let header = format!("[{}]", section);
     let mut result = String::new();
diff --git a/crates/openfang-api/static/js/pages/agents.js b/crates/openfang-api/static/js/pages/agents.js
index 21f506f1..5a3c7df4 100644
--- a/crates/openfang-api/static/js/pages/agents.js
+++ b/crates/openfang-api/static/js/pages/agents.js
@@ -393,7 +393,7 @@ function agentsPage() {
     },
 
     // ── Multi-step wizard navigation ──
-    openSpawnWizard() {
+    async openSpawnWizard() {
       this.showSpawnModal = true;
       this.spawnStep = 1;
       this.spawnMode = 'wizard';
@@ -401,8 +401,18 @@ function agentsPage() {
       this.selectedPreset = '';
       this.soulContent = '';
       this.spawnForm.name = '';
+      this.spawnForm.provider = 'groq';
+      this.spawnForm.model = 'llama-3.3-70b-versatile';
       this.spawnForm.systemPrompt = 'You are a helpful assistant.';
       this.spawnForm.profile = 'full';
+      try {
+        var res = await fetch('/api/status');
+        if (res.ok) {
+          var status = await res.json();
+          if (status.default_provider) this.spawnForm.provider = status.default_provider;
+          if (status.default_model) this.spawnForm.model = status.default_model;
+        }
+      } catch(e) { /* keep hardcoded defaults */ }
     },
 
     nextStep() {
diff --git a/crates/openfang-channels/src/bridge.rs b/crates/openfang-channels/src/bridge.rs
index 60b29562..390b367f 100644
--- a/crates/openfang-channels/src/bridge.rs
+++ b/crates/openfang-channels/src/bridge.rs
@@ -780,7 +780,7 @@ async fn dispatch_message(
                 send_lifecycle_reaction(adapter, &message.sender, msg_id, AgentPhase::Error).await;
             }
             warn!("Agent error for {agent_id}: {e}");
-            let err_msg = format!("Agent error: {e}");
+            let err_msg = sanitize_agent_error(&e.to_string());
             send_response(
                 adapter,
                 &message.sender,
@@ -803,6 +803,76 @@ async fn dispatch_message(
     }
 }
 
+fn sanitize_agent_error(raw: &str) -> String {
+    let lower = raw.to_lowercase();
+
+    if lower.contains("rate limit")
+        || lower.contains("rate_limit")
+        || lower.contains("429")
+        || lower.contains("too many requests")
+        || lower.contains("resource_exhausted")
+    {
+        return "Rate limit reached, please try again later.".to_string();
+    }
+
+    if lower.contains("authentication")
+        || lower.contains("unauthorized")
+        || lower.contains("invalid api key")
+        || lower.contains("invalid x-goog-api-key")
+        || lower.contains("incorrect api key")
+        || lower.contains("permission denied")
+        || lower.contains("billing")
+        || lower.contains("quota exceeded")
+    {
+        return "Service temporarily unavailable.".to_string();
+    }
+
+    if lower.contains("context length")
+        || lower.contains("token limit")
+        || lower.contains("too many tokens")
+        || lower.contains("maximum context")
+        || lower.contains("max_tokens")
+        || lower.contains("context window")
+    {
+        return "Message too long, try a shorter request.".to_string();
+    }
+
+    if lower.contains("overloaded")
+        || lower.contains("503")
+        || lower.contains("502")
+        || lower.contains("server error")
+        || lower.contains("internal error")
+    {
+        return "The AI service is temporarily overloaded, please try again shortly.".to_string();
+    }
+
+    if lower.contains("timeout") || lower.contains("timed out") || lower.contains("deadline") {
+        return "Request timed out, please try again.".to_string();
+    }
+
+    if lower.contains("model not found") || lower.contains("model_not_found") {
+        return "The requested model is currently unavailable.".to_string();
+    }
+
+    let cleaned = raw
+        .strip_prefix("LLM driver error: ")
+        .or_else(|| raw.strip_prefix("Agent error: "))
+        .unwrap_or(raw);
+
+    if let Some(first_sentence_end) = cleaned.find(". ") {
+        let first = &cleaned[..=first_sentence_end];
+        if first.len() < cleaned.len() / 2 {
+            return format!("Agent error: {first}");
+        }
+    }
+
+    if cleaned.contains('{') || cleaned.len() > 200 {
+        return "Something went wrong processing your request. Please try again.".to_string();
+    }
+
+    format!("Agent error: {cleaned}")
+}
+
 /// Detect image format from the first few magic bytes.
 ///
 /// Returns `Some("image/...")` for JPEG, PNG, GIF, and WebP.
@@ -1012,7 +1082,7 @@ async fn dispatch_with_blocks(
                 send_lifecycle_reaction(adapter, &message.sender, msg_id, AgentPhase::Error).await;
             }
             warn!("Agent error for {agent_id}: {e}");
-            let err_msg = format!("Agent error: {e}");
+            let err_msg = sanitize_agent_error(&e.to_string());
             send_response(
                 adapter,
                 &message.sender,
diff --git a/crates/openfang-cli/src/main.rs b/crates/openfang-cli/src/main.rs
index 6ab71db4..7bfa88ac 100644
--- a/crates/openfang-cli/src/main.rs
+++ b/crates/openfang-cli/src/main.rs
@@ -794,11 +794,36 @@ enum SystemCommands {
     },
 }
 
+fn config_log_level() -> String {
+    let config_path = if let Ok(home) = std::env::var("OPENFANG_HOME") {
+        std::path::PathBuf::from(home).join("config.toml")
+    } else {
+        dirs::home_dir()
+            .unwrap_or_else(std::env::temp_dir)
+            .join(".openfang")
+            .join("config.toml")
+    };
+    if let Ok(content) = std::fs::read_to_string(config_path) {
+        for line in content.lines() {
+            let trimmed = line.trim();
+            if trimmed.starts_with("log_level") {
+                if let Some(val) = trimmed.split('=').nth(1) {
+                    let level = val.trim().trim_matches('"').trim_matches('\'');
+                    if !level.is_empty() {
+                        return level.to_string();
+                    }
+                }
+            }
+        }
+    }
+    "info".to_string()
+}
+
 fn init_tracing_stderr() {
     tracing_subscriber::fmt()
         .with_env_filter(
             tracing_subscriber::EnvFilter::try_from_default_env()
-                .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new("info")),
+                .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new(config_log_level())),
         )
         .init();
 }
@@ -824,7 +849,7 @@ fn init_tracing_file() {
             tracing_subscriber::fmt()
                 .with_env_filter(
                     tracing_subscriber::EnvFilter::try_from_default_env()
-                        .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new("info")),
+                        .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new(config_log_level())),
                 )
                 .with_writer(std::sync::Mutex::new(file))
                 .with_ansi(false)
@@ -4691,6 +4716,8 @@ fn cmd_config_set(key: &str, value: &str) {
         std::process::exit(1);
     });
 
+    let _ = std::fs::copy(&config_path, config_path.with_extension("toml.bak"));
+
     std::fs::write(&config_path, &serialized).unwrap_or_else(|e| {
         ui::error(&format!("Failed to write config: {e}"));
         std::process::exit(1);
@@ -4757,6 +4784,8 @@ fn cmd_config_unset(key: &str) {
         std::process::exit(1);
     });
 
+    let _ = std::fs::copy(&config_path, config_path.with_extension("toml.bak"));
+
     std::fs::write(&config_path, &serialized).unwrap_or_else(|e| {
         ui::error(&format!("Failed to write config: {e}"));
         std::process::exit(1);
diff --git a/crates/openfang-hands/src/bundled.rs b/crates/openfang-hands/src/bundled.rs
index b54b9feb..bce5af2d 100644
--- a/crates/openfang-hands/src/bundled.rs
+++ b/crates/openfang-hands/src/bundled.rs
@@ -1,6 +1,6 @@
 //! Compile-time embedded Hand definitions.
 
-use crate::{HandDefinition, HandError};
+use crate::{parse_hand_toml, HandDefinition, HandError};
 
 /// Returns all bundled hand definitions as (id, HAND.toml content, SKILL.md content).
 pub fn bundled_hands() -> Vec<(&'static str, &'static str, &'static str)> {
@@ -55,7 +55,7 @@ pub fn parse_bundled(
     skill_content: &str,
 ) -> Result<HandDefinition, HandError> {
     let mut def: HandDefinition =
-        toml::from_str(toml_content).map_err(|e| HandError::TomlParse(e.to_string()))?;
+        parse_hand_toml(toml_content).map_err(|e| HandError::TomlParse(e.to_string()))?;
     if !skill_content.is_empty() {
         def.skill_content = Some(skill_content.to_string());
     }
diff --git a/crates/openfang-hands/src/lib.rs b/crates/openfang-hands/src/lib.rs
index f2d3e339..a96dfd44 100644
--- a/crates/openfang-hands/src/lib.rs
+++ b/crates/openfang-hands/src/lib.rs
@@ -306,6 +306,20 @@ fn default_temperature() -> f32 {
     0.7
 }
 
+#[derive(Deserialize)]
+struct HandTomlWrapper {
+    hand: HandDefinition,
+}
+
+/// Parse HAND.toml content, supporting both flat format and `[hand]` table format.
+pub fn parse_hand_toml(content: &str) -> Result<HandDefinition, toml::de::Error> {
+    if let Ok(def) = toml::from_str::<HandDefinition>(content) {
+        return Ok(def);
+    }
+    let wrapper: HandTomlWrapper = toml::from_str(content)?;
+    Ok(wrapper.hand)
+}
+
 /// Complete Hand definition — parsed from HAND.toml.
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct HandDefinition {
@@ -798,4 +812,50 @@ metrics = []
         assert!(install.macos.is_none());
         assert!(install.windows.is_none());
     }
+
+    #[test]
+    fn parse_hand_toml_flat_format() {
+        let toml_str = r#"
+id = "test"
+name = "Test Hand"
+description = "A test hand"
+category = "content"
+tools = ["shell_exec"]
+
+[agent]
+name = "test-hand"
+description = "Test agent"
+system_prompt = "You are a test agent."
+
+[dashboard]
+metrics = []
+"#;
+        let def = parse_hand_toml(toml_str).unwrap();
+        assert_eq!(def.id, "test");
+        assert_eq!(def.name, "Test Hand");
+    }
+
+    #[test]
+    fn parse_hand_toml_wrapped_format() {
+        let toml_str = r#"
+[hand]
+id = "test"
+name = "Test Hand"
+description = "A test hand"
+category = "content"
+tools = ["shell_exec"]
+
+[hand.agent]
+name = "test-hand"
+description = "Test agent"
+system_prompt = "You are a test agent."
+
+[hand.dashboard]
+metrics = []
+"#;
+        let def = parse_hand_toml(toml_str).unwrap();
+        assert_eq!(def.id, "test");
+        assert_eq!(def.name, "Test Hand");
+        assert_eq!(def.agent.name, "test-hand");
+    }
 }
diff --git a/crates/openfang-kernel/src/kernel.rs b/crates/openfang-kernel/src/kernel.rs
index 99f5de91..4ecf3587 100644
--- a/crates/openfang-kernel/src/kernel.rs
+++ b/crates/openfang-kernel/src/kernel.rs
@@ -1602,7 +1602,7 @@ impl OpenFangKernel {
                 label: None,
             });
 
-        // Check if auto-compaction is needed: message-count OR token-count trigger
+        // Check if auto-compaction is needed: message-count OR token-count OR quota-headroom trigger
         let needs_compact = {
             use openfang_runtime::compactor::{
                 estimate_token_count, needs_compaction as check_compact,
@@ -1624,7 +1624,23 @@ impl OpenFangKernel {
                     "Token-based compaction triggered (messages below threshold but tokens above)"
                 );
             }
-            by_messages || by_tokens
+            let by_quota = if let Some(headroom) = self.scheduler.token_headroom(agent_id) {
+                let threshold = (headroom as f64 * 0.8) as u64;
+                if estimated as u64 > threshold && session.messages.len() > 4 {
+                    info!(
+                        agent_id = %agent_id,
+                        estimated_tokens = estimated,
+                        quota_headroom = headroom,
+                        "Quota-headroom compaction triggered (session would consume >80% of remaining quota)"
+                    );
+                    true
+                } else {
+                    false
+                }
+            } else {
+                false
+            };
+            by_messages || by_tokens || by_quota
         };
 
         let tools = self.available_tools(agent_id);
@@ -2094,6 +2110,42 @@ impl OpenFangKernel {
                 label: None,
             });
 
+        // Pre-emptive compaction: compact before LLM call if session is large or quota headroom is low
+        {
+            use openfang_runtime::compactor::{
+                estimate_token_count, needs_compaction as check_compact,
+                needs_compaction_by_tokens, CompactionConfig,
+            };
+            let config = CompactionConfig::default();
+            let by_messages = check_compact(&session, &config);
+            let estimated = estimate_token_count(
+                &session.messages,
+                Some(&entry.manifest.model.system_prompt),
+                None,
+            );
+            let by_tokens = needs_compaction_by_tokens(estimated, &config);
+            let by_quota = if let Some(headroom) = self.scheduler.token_headroom(agent_id) {
+                let threshold = (headroom as f64 * 0.8) as u64;
+                estimated as u64 > threshold && session.messages.len() > 4
+            } else {
+                false
+            };
+            if by_messages || by_tokens || by_quota {
+                info!(agent_id = %agent_id, messages = session.messages.len(), estimated_tokens = estimated, "Pre-emptive compaction before LLM call");
+                match self.compact_agent_session(agent_id).await {
+                    Ok(msg) => {
+                        info!(agent_id = %agent_id, "{msg}");
+                        if let Ok(Some(reloaded)) = self.memory.get_session(session.id) {
+                            session = reloaded;
+                        }
+                    }
+                    Err(e) => {
+                        warn!(agent_id = %agent_id, "Pre-emptive compaction failed: {e}");
+                    }
+                }
+            }
+        }
+
         let messages_before = session.messages.len();
 
         let tools = self.available_tools(agent_id);
diff --git a/crates/openfang-kernel/src/scheduler.rs b/crates/openfang-kernel/src/scheduler.rs
index b0fb5245..c75fb869 100644
--- a/crates/openfang-kernel/src/scheduler.rs
+++ b/crates/openfang-kernel/src/scheduler.rs
@@ -130,6 +130,19 @@ impl AgentScheduler {
             .get(&agent_id)
             .map(|t| (t.total_tokens, t.tool_calls))
     }
+
+    /// Returns remaining token headroom before quota is hit.
+    /// Returns `None` if no token quota is configured (unlimited).
+    pub fn token_headroom(&self, agent_id: AgentId) -> Option<u64> {
+        let quota = self.quotas.get(&agent_id)?;
+        if quota.max_llm_tokens_per_hour == 0 {
+            return None;
+        }
+        let mut tracker = self.usage.get_mut(&agent_id)?;
+        tracker.reset_if_expired();
+        let used = tracker.total_tokens;
+        Some(quota.max_llm_tokens_per_hour.saturating_sub(used))
+    }
 }
 
 impl Default for AgentScheduler {
diff --git a/crates/openfang-runtime/src/prompt_builder.rs b/crates/openfang-runtime/src/prompt_builder.rs
index e0a8bd2a..4e984490 100644
--- a/crates/openfang-runtime/src/prompt_builder.rs
+++ b/crates/openfang-runtime/src/prompt_builder.rs
@@ -275,12 +275,18 @@ pub fn build_canonical_context_message(ctx: &PromptContext) -> Option<String> {
 ///
 /// Also used by `agent_loop.rs` to append recalled memories after DB lookup.
 pub fn build_memory_section(memories: &[(String, String)]) -> String {
-    let mut out = String::from(
-        "## Memory\n\
-         - When the user asks about something from a previous conversation, use memory_recall first.\n\
-         - Store important preferences, decisions, and context with memory_store for future use.",
-    );
-    if !memories.is_empty() {
+    let mut out = String::from("## Memory\n");
+    if memories.is_empty() {
+        out.push_str(
+            "- When the user asks about something from a previous conversation, use memory_recall first.\n\
+             - Store important preferences, decisions, and context with memory_store for future use.",
+        );
+    } else {
+        out.push_str(
+            "- Use the recalled memories below to inform your responses.\n\
+             - Only call memory_recall if you need information not already shown here.\n\
+             - Store important preferences, decisions, and context with memory_store for future use.",
+        );
         out.push_str("\n\nRecalled memories:\n");
         for (key, content) in memories.iter().take(5) {
             let capped = cap_str(content, 500);
@@ -728,7 +734,7 @@ mod tests {
     fn test_memory_section_empty() {
         let section = build_memory_section(&[]);
         assert!(section.contains("## Memory"));
-        assert!(section.contains("memory_recall"));
+        assert!(section.contains("use memory_recall first"));
         assert!(!section.contains("Recalled memories"));
     }
 
@@ -742,6 +748,8 @@ mod tests {
         assert!(section.contains("Recalled memories"));
         assert!(section.contains("[pref] User likes dark mode"));
         assert!(section.contains("[ctx] Working on Rust project"));
+        assert!(section.contains("Use the recalled memories below"));
+        assert!(!section.contains("use memory_recall first"));
     }
 
     #[test]
diff --git a/crates/openfang-types/src/error.rs b/crates/openfang-types/src/error.rs
index beadcf0c..4f6be014 100644
--- a/crates/openfang-types/src/error.rs
+++ b/crates/openfang-types/src/error.rs
@@ -72,7 +72,7 @@ pub enum OpenFangError {
     Serialization(String),
 
     /// The agent loop exceeded the maximum iteration count.
-    #[error("Max iterations exceeded: {0}")]
+    #[error("Max iterations exceeded ({0}). Configure a higher limit in agent.toml under [autonomous] max_iterations")]
     MaxIterationsExceeded(u32),
 
     /// The kernel is shutting down.