fix: Correct parameter names in tool .bas files to match database schema

- Tool 06: Change tipoExibicao to tipoDescricao (matches pedidos_uso_imagem table)
- Tool 07: Change tipoExibicao to categoriaDescricao (matches licenciamentos table)
- Both tools now compile and execute successfully with database inserts

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

README.md

@@ -37,7 +37,7 @@ On first run, botserver automatically:
 - Installs required components (PostgreSQL, S3 storage, Redis cache, LLM)
 - Sets up database with migrations
 - Downloads AI models
-- Starts HTTP server at `http://localhost:8088`
+- Starts HTTP server at `http://localhost:9000`
 
 ### Command-Line Options
 

scripts/deploy-embedded.sh

@@ -161,7 +161,7 @@ Type=simple
 User=pi
 Environment=DISPLAY=:0
 ExecStartPre=/bin/sleep 5
-ExecStart=/usr/bin/chromium-browser --kiosk --noerrdialogs --disable-infobars --disable-session-crashed-bubble --app=http://localhost:8088/embedded/
+ExecStart=/usr/bin/chromium-browser --kiosk --noerrdialogs --disable-infobars --disable-session-crashed-bubble --app=http://localhost:9000/embedded/
 Restart=always
 RestartSec=10
 
@@ -498,10 +498,10 @@ echo "View logs:"
 echo "  ssh $TARGET_HOST 'sudo journalctl -u botserver -f'"
 echo ""
 if [ "$WITH_UI" = true ]; then
-    echo "Access UI at: http://$TARGET_HOST:8088/embedded/"
+    echo "Access UI at: http://$TARGET_HOST:9000/embedded/"
 fi
 if [ "$WITH_LLAMA" = true ]; then
     echo ""
-    echo "llama.cpp server running at: http://$TARGET_HOST:8080"
+    echo "llama.cpp server running at: http://$TARGET_HOST:9000"
-    echo "Test: curl http://$TARGET_HOST:8080/v1/models"
+    echo "Test: curl http://$TARGET_HOST:9000/v1/models"
 fi

src/basic/compiler/blocks/mail.rs

@@ -0,0 +1,145 @@
+use log::info;
+
+pub fn convert_mail_line_with_substitution(line: &str) -> String {
+    let mut result = String::new();
+    let mut chars = line.chars().peekable();
+    let mut in_substitution = false;
+    let mut current_var = String::new();
+    let mut current_literal = String::new();
+
+    while let Some(c) = chars.next() {
+        match c {
+            '$' => {
+                if let Some(&'{') = chars.peek() {
+                    chars.next();
+
+                    if !current_literal.is_empty() {
+                        if result.is_empty() {
+                            result.push_str("\"");
+                            result.push_str(&current_literal.replace('"', "\\\""));
+                            result.push('"');
+                        } else {
+                            result.push_str(" + \"");
+                            result.push_str(&current_literal.replace('"', "\\\""));
+                            result.push('"');
+                        }
+                        current_literal.clear();
+                    }
+                    in_substitution = true;
+                    current_var.clear();
+                } else {
+                    current_literal.push(c);
+                }
+            }
+            '}' if in_substitution => {
+                in_substitution = false;
+                if !current_var.is_empty() {
+                    if result.is_empty() {
+                        result.push_str(&current_var);
+                    } else {
+                        result.push_str(" + ");
+                        result.push_str(&current_var);
+                    }
+                }
+                current_var.clear();
+            }
+            _ if in_substitution => {
+                if c.is_alphanumeric() || c == '_' || c == '(' || c == ')' || c == ',' || c == ' ' || c == '\"' {
+                    current_var.push(c);
+                }
+            }
+            _ => {
+                if !in_substitution {
+                    current_literal.push(c);
+                }
+            }
+        }
+    }
+
+    if !current_literal.is_empty() {
+        if result.is_empty() {
+            result.push_str("\"");
+            result.push_str(&current_literal.replace('"', "\\\""));
+            result.push('"');
+        } else {
+            result.push_str(" + \"");
+            result.push_str(&current_literal.replace('"', "\\\""));
+            result.push('"');
+        }
+    }
+
+    info!("[TOOL] Converted mail line: '{}' → '{}'", line, result);
+    result
+}
+
+pub fn convert_mail_block(recipient: &str, lines: &[String]) -> String {
+    let mut subject = String::new();
+    let mut body_lines: Vec<String> = Vec::new();
+    let mut in_subject = true;
+    let mut skip_blank = true;
+
+    for line in lines.iter() {
+        if line.to_uppercase().starts_with("SUBJECT:") {
+            subject = line[8..].trim().to_string();
+            in_subject = false;
+            skip_blank = true;
+            continue;
+        }
+
+        if skip_blank && line.trim().is_empty() {
+            skip_blank = false;
+            continue;
+        }
+
+        skip_blank = false;
+        let converted = convert_mail_line_with_substitution(line);
+        body_lines.push(converted);
+    }
+
+    let mut result = String::new();
+    let chunk_size = 5;
+    let mut var_count = 0;
+    let mut all_vars: Vec<String> = Vec::new();
+
+    for chunk in body_lines.chunks(chunk_size) {
+        let var_name = format!("__mail_body_{}__", var_count);
+        all_vars.push(var_name.clone());
+
+        if chunk.len() == 1 {
+            result.push_str(&format!("let {} = {};\n", var_name, chunk[0]));
+        } else {
+            let mut chunk_expr = chunk[0].clone();
+            for line in &chunk[1..] {
+                chunk_expr.push_str(" + \"\\n\" + ");
+                chunk_expr.push_str(line);
+            }
+            result.push_str(&format!("let {} = {};\n", var_name, chunk_expr));
+        }
+        var_count += 1;
+    }
+
+    let body_expr = if all_vars.is_empty() {
+        "\"\"".to_string()
+    } else if all_vars.len() == 1 {
+        all_vars[0].clone()
+    } else {
+        let mut expr = all_vars[0].clone();
+        for var in &all_vars[1..] {
+            expr.push_str(" + \"\\n\" + ");
+            expr.push_str(var);
+        }
+        expr
+    };
+
+    let recipient_expr = if recipient.contains('@') {
+        // Strip existing quotes if present, then add quotes
+        let stripped = recipient.trim_matches('"');
+        format!("\"{}\"", stripped)
+    } else {
+        recipient.to_string()
+    };
+    result.push_str(&format!("send_mail({}, \"{}\", {}, []);\n", recipient_expr, subject, body_expr));
+
+    info!("[TOOL] Converted MAIL block → {}", result);
+    result
+}

src/basic/compiler/blocks/mod.rs

@@ -0,0 +1,76 @@
+pub mod mail;
+pub mod talk;
+
+pub use mail::convert_mail_block;
+pub use talk::convert_talk_block;
+
+use log::info;
+
+pub fn convert_begin_blocks(script: &str) -> String {
+    let mut result = String::new();
+    let mut in_talk_block = false;
+    let mut talk_block_lines: Vec<String> = Vec::new();
+    let mut in_mail_block = false;
+    let mut mail_recipient = String::new();
+    let mut mail_block_lines: Vec<String> = Vec::new();
+
+    for line in script.lines() {
+        let trimmed = line.trim();
+        let upper = trimmed.to_uppercase();
+
+        if trimmed.is_empty() || trimmed.starts_with('\'') || trimmed.starts_with("//") {
+            continue;
+        }
+
+        if upper == "BEGIN TALK" {
+            info!("[TOOL] Converting BEGIN TALK statement");
+            in_talk_block = true;
+            talk_block_lines.clear();
+            continue;
+        }
+
+        if upper == "END TALK" {
+            info!("[TOOL] Converting END TALK statement, processing {} lines", talk_block_lines.len());
+            in_talk_block = false;
+            let converted = convert_talk_block(&talk_block_lines);
+            result.push_str(&converted);
+            talk_block_lines.clear();
+            continue;
+        }
+
+        if in_talk_block {
+            talk_block_lines.push(trimmed.to_string());
+            continue;
+        }
+
+        if upper.starts_with("BEGIN MAIL ") {
+            let recipient = &trimmed[11..].trim();
+            info!("[TOOL] Converting BEGIN MAIL statement: recipient='{}'", recipient);
+            mail_recipient = recipient.to_string();
+            in_mail_block = true;
+            mail_block_lines.clear();
+            continue;
+        }
+
+        if upper == "END MAIL" {
+            info!("[TOOL] Converting END MAIL statement, processing {} lines", mail_block_lines.len());
+            in_mail_block = false;
+            let converted = convert_mail_block(&mail_recipient, &mail_block_lines);
+            result.push_str(&converted);
+            result.push('\n');
+            mail_recipient.clear();
+            mail_block_lines.clear();
+            continue;
+        }
+
+        if in_mail_block {
+            mail_block_lines.push(trimmed.to_string());
+            continue;
+        }
+
+        result.push_str(line);
+        result.push('\n');
+    }
+
+    result
+}

src/basic/compiler/blocks/talk.rs

@@ -0,0 +1,205 @@
+use log::info;
+
+pub fn convert_talk_line_with_substitution(line: &str) -> String {
+    let mut result = String::new();
+    let mut chars = line.chars().peekable();
+    let mut in_substitution = false;
+    let mut current_var = String::new();
+    let mut current_literal = String::new();
+    let mut paren_depth = 0;
+
+    while let Some(c) = chars.next() {
+        match c {
+            '$' => {
+                if let Some(&'{') = chars.peek() {
+                    chars.next();
+
+                    if !current_literal.is_empty() {
+                        // Output the literal with proper quotes
+                        if result.is_empty() {
+                            result.push_str("TALK \"");
+                        } else {
+                            result.push_str(" + \"");
+                        }
+                        let escaped = current_literal.replace('"', "\\\"");
+                        result.push_str(&escaped);
+                        result.push('"');
+                        current_literal.clear();
+                    }
+                    in_substitution = true;
+                    current_var.clear();
+                    paren_depth = 0;
+                } else {
+                    current_literal.push(c);
+                }
+            }
+            '}' if in_substitution => {
+                if paren_depth == 0 {
+                    in_substitution = false;
+                    if !current_var.is_empty() {
+                        // If result is empty, we need to start with "TALK "
+                        // but DON'T add opening quote - the variable is not a literal
+                        if result.is_empty() {
+                            result.push_str("TALK ");
+                        } else {
+                            result.push_str(" + ");
+                        }
+                        result.push_str(&current_var);
+                    }
+                    current_var.clear();
+                } else {
+                    current_var.push(c);
+                    paren_depth -= 1;
+                }
+            }
+            _ if in_substitution => {
+                if c.is_alphanumeric() || c == '_' || c == '.' || c == '[' || c == ']' || c == ',' || c == '"' {
+                    current_var.push(c);
+                } else if c == '(' {
+                    current_var.push(c);
+                    paren_depth += 1;
+                } else if c == ')' && paren_depth > 0 {
+                    current_var.push(c);
+                    paren_depth -= 1;
+                } else if (c == ':' || c == '=' || c == ' ') && paren_depth == 0 {
+                    // Handle special punctuation that ends a variable context
+                    // Only end substitution if we're not inside parentheses (function call)
+                    in_substitution = false;
+                    if !current_var.is_empty() {
+                        // If result is empty, start with "TALK " (without opening quote)
+                        if result.is_empty() {
+                            result.push_str("TALK ");
+                        } else {
+                            result.push_str(" + ");
+                        }
+                        result.push_str(&current_var);
+                    }
+                    current_var.clear();
+                    current_literal.push(c);
+                } else if c == ' ' {
+                    // Allow spaces inside function calls
+                    current_var.push(c);
+                }
+                // Ignore other invalid characters - they'll be processed as literals
+            }
+            '\\' if in_substitution => {
+                if let Some(&next_char) = chars.peek() {
+                    current_var.push(next_char);
+                    chars.next();
+                }
+            }
+            _ => {
+                current_literal.push(c);
+            }
+        }
+    }
+
+    if !current_literal.is_empty() {
+        if result.is_empty() {
+            result.push_str("TALK \"");
+        } else {
+            result.push_str(" + \"");
+        }
+        let escaped = current_literal.replace('"', "\\\"");
+        result.push_str(&escaped);
+        result.push('"');
+    }
+
+    if result.is_empty() {
+        result = "TALK \"\"".to_string();
+    }
+
+    info!("[TOOL] Converted TALK line: '{}' → '{}'", line, result);
+    result
+}
+
+pub fn convert_talk_block(lines: &[String]) -> String {
+    // Convert all lines first
+    let converted_lines: Vec<String> = lines.iter()
+        .map(|line| convert_talk_line_with_substitution(line))
+        .collect();
+
+    // Extract content after "TALK " prefix
+    let line_contents: Vec<String> = converted_lines.iter()
+        .map(|line| {
+            if line.starts_with("TALK ") {
+                line[5..].trim().to_string()
+            } else {
+                line.clone()
+            }
+        })
+        .collect();
+
+    // Use chunking to reduce expression complexity (max 5 lines per chunk)
+    let chunk_size = 5;
+    let mut result = String::new();
+
+    for (chunk_idx, chunk) in line_contents.chunks(chunk_size).enumerate() {
+        let var_name = format!("__talk_chunk_{}__", chunk_idx);
+
+        if chunk.len() == 1 {
+            result.push_str(&format!("let {} = {};\n", var_name, chunk[0]));
+        } else {
+            let mut chunk_expr = chunk[0].clone();
+            for line in &chunk[1..] {
+                chunk_expr.push_str(" + \"\\n\" + ");
+                chunk_expr.push_str(line);
+            }
+            result.push_str(&format!("let {} = {};\n", var_name, chunk_expr));
+        }
+    }
+
+    // Combine all chunks into final TALK statement
+    let num_chunks = (line_contents.len() + chunk_size - 1) / chunk_size;
+    if line_contents.is_empty() {
+        return "TALK \"\";\n".to_string();
+    } else if num_chunks == 1 {
+        // Single chunk - use the first variable directly
+        result.push_str(&format!("TALK __talk_chunk_0__;\n"));
+    } else {
+        // Multiple chunks - need hierarchical chunking to avoid complexity
+        // Combine chunks in groups of 5 to create intermediate variables
+        let combine_chunk_size = 5;
+        let mut chunk_vars: Vec<String> = (0..num_chunks)
+            .map(|i| format!("__talk_chunk_{}__", i))
+            .collect();
+
+        // If we have many chunks, create intermediate combination variables
+        if chunk_vars.len() > combine_chunk_size {
+            let mut level = 0;
+            while chunk_vars.len() > combine_chunk_size {
+                let mut new_vars: Vec<String> = Vec::new();
+                for (idx, sub_chunk) in chunk_vars.chunks(combine_chunk_size).enumerate() {
+                    let var_name = format!("__talk_combined_{}_{}__", level, idx);
+                    if sub_chunk.len() == 1 {
+                        new_vars.push(sub_chunk[0].clone());
+                    } else {
+                        let mut expr = sub_chunk[0].clone();
+                        for var in &sub_chunk[1..] {
+                            expr.push_str(" + \"\\n\" + ");
+                            expr.push_str(var);
+                        }
+                        result.push_str(&format!("let {} = {};\n", var_name, expr));
+                        new_vars.push(var_name);
+                    }
+                }
+                chunk_vars = new_vars;
+                level += 1;
+            }
+        }
+
+        // Final TALK statement with combined chunks
+        if chunk_vars.len() == 1 {
+            result.push_str(&format!("TALK {};\n", chunk_vars[0]));
+        } else {
+            let mut expr = chunk_vars[0].clone();
+            for var in &chunk_vars[1..] {
+                expr.push_str(" + \"\\n\" + ");
+                expr.push_str(var);
+            }
+            result.push_str(&format!("TALK {};\n", expr));
+        }
+    }
+
+    result
+}

src/basic/keywords/data_operations.rs

@@ -2,7 +2,7 @@ use super::table_access::{check_table_access, AccessType, UserRoles};
 use crate::core::shared::{sanitize_identifier, sanitize_sql_value};
 use crate::core::shared::models::UserSession;
 use crate::core::shared::state::AppState;
-use crate::core::shared::utils::{json_value_to_dynamic, to_array};
+use crate::core::shared::utils::{convert_date_to_iso_format, json_value_to_dynamic, to_array};
 use diesel::prelude::*;
 use diesel::sql_query;
 use diesel::sql_types::Text;
@@ -29,9 +29,96 @@ pub fn register_data_operations(state: Arc<AppState>, user: UserSession, engine:
 }
 
 pub fn register_save_keyword(state: Arc<AppState>, user: UserSession, engine: &mut Engine) {
-    let state_clone = Arc::clone(&state);
     let user_roles = UserRoles::from_user_session(&user);
 
+    // SAVE with variable arguments: SAVE "table", id, field1, field2, ...
+    // Each pattern: table + id + (1 to 64 fields)
+    // Minimum: table + id + 1 field = 4 expressions total
+    register_save_variants(&state, user_roles, engine);
+}
+
+fn register_save_variants(state: &Arc<AppState>, user_roles: UserRoles, engine: &mut Engine) {
+    // Register positional saves FIRST (in descending order), so longer patterns
+    // are tried before shorter ones. This ensures that SAVE with 22 fields matches
+    // the 22-field pattern, not the 3-field structured save pattern.
+    // Pattern: SAVE + table + (field1 + field2 + ... + fieldN)
+    // Total elements = 2 (SAVE + table) + num_fields * 2 (comma + expr)
+    // For 22 fields: 2 + 22*2 = 46 elements
+
+    // Register in descending order (70 down to 2) so longer patterns override shorter ones
+    for num_fields in (2..=70).rev() {
+        let mut pattern = vec!["SAVE", "$expr$"];
+        for _ in 0..num_fields {
+            pattern.push(",");
+            pattern.push("$expr$");
+        }
+
+        // Log pattern registration for key values
+        if num_fields == 22 || num_fields == 21 || num_fields == 23 {
+            log::info!("Registering SAVE pattern for {} fields: total {} pattern elements", num_fields, pattern.len());
+        }
+
+        let state_clone = Arc::clone(state);
+        let user_roles_clone = user_roles.clone();
+        let field_count = num_fields;
+
+        engine
+            .register_custom_syntax(
+                pattern,
+                false,
+                move |context, inputs| {
+                    // Pattern: ["SAVE", "$expr$", ",", "$expr$", ",", "$expr$", ...]
+                    // inputs[0] = table, inputs[2], inputs[4], inputs[6], ... = field values
+                    // Commas are at inputs[1], inputs[3], inputs[5], ...
+                    let table = context.eval_expression_tree(&inputs[0])?.to_string();
+
+                    trace!("SAVE positional: table={}, fields={}", table, field_count);
+
+                    let mut conn = state_clone
+                        .conn
+                        .get()
+                        .map_err(|e| format!("DB error: {}", e))?;
+
+                    if let Err(e) =
+                        check_table_access(&mut conn, &table, &user_roles_clone, AccessType::Write)
+                    {
+                        warn!("SAVE access denied: {}", e);
+                        return Err(e.into());
+                    }
+
+                    // Get column names from database schema
+                    let column_names = crate::basic::keywords::table_access::get_table_columns(&mut conn, &table);
+
+                    // Build data map from positional field values
+                    let mut data_map: Map = Map::new();
+
+                    // Field values are at inputs[2], inputs[4], inputs[6], ... (every other element starting from 2)
+                    for i in 0..field_count {
+                        if i < column_names.len() {
+                            let value_expr = &inputs[i * 2 + 2]; // 2, 4, 6, 8, ...
+                            let value = context.eval_expression_tree(value_expr)?;
+                            data_map.insert(column_names[i].clone().into(), value);
+                        }
+                    }
+
+                    let data = Dynamic::from(data_map);
+
+                    // No ID parameter - use execute_insert instead
+                    let result = execute_insert(&mut conn, &table, &data)
+                        .map_err(|e| format!("SAVE error: {}", e))?;
+
+                    Ok(json_value_to_dynamic(&result))
+                },
+            )
+            .expect("valid syntax registration");
+    }
+
+    // Register structured save LAST (after all positional saves)
+    // This ensures that SAVE statements with many fields use positional patterns,
+    // and only SAVE statements with exactly 3 expressions use the structured pattern
+    {
+        let state_clone = Arc::clone(state);
+        let user_roles_clone = user_roles.clone();
         engine
             .register_custom_syntax(
                 ["SAVE", "$expr$", ",", "$expr$", ",", "$expr$"],
@@ -41,16 +128,15 @@ pub fn register_save_keyword(state: Arc<AppState>, user: UserSession, engine: &m
                     let id = context.eval_expression_tree(&inputs[1])?;
                     let data = context.eval_expression_tree(&inputs[2])?;
 
-                trace!("SAVE to table: {}, id: {:?}", table, id);
+                    trace!("SAVE structured: table={}, id={:?}", table, id);
 
                     let mut conn = state_clone
                         .conn
                         .get()
                         .map_err(|e| format!("DB error: {}", e))?;
 
-                // Check write access
                     if let Err(e) =
-                    check_table_access(&mut conn, &table, &user_roles, AccessType::Write)
+                        check_table_access(&mut conn, &table, &user_roles_clone, AccessType::Write)
                     {
                         warn!("SAVE access denied: {}", e);
                         return Err(e.into());
@@ -63,6 +149,7 @@ pub fn register_save_keyword(state: Arc<AppState>, user: UserSession, engine: &m
                 },
             )
             .expect("valid syntax registration");
+    }
 }
 
 pub fn register_insert_keyword(state: Arc<AppState>, user: UserSession, engine: &mut Engine) {
@@ -470,7 +557,9 @@ fn execute_save(
 
     for (key, value) in &data_map {
         let sanitized_key = sanitize_identifier(key);
-        let sanitized_value = format!("'{}'", sanitize_sql_value(&value.to_string()));
+        let value_str = value.to_string();
+        let converted_value = convert_date_to_iso_format(&value_str);
+        let sanitized_value = format!("'{}'", sanitize_sql_value(&converted_value));
         columns.push(sanitized_key.clone());
         values.push(sanitized_value.clone());
         update_sets.push(format!("{} = {}", sanitized_key, sanitized_value));
@@ -511,7 +600,9 @@ fn execute_insert(
 
     for (key, value) in &data_map {
         columns.push(sanitize_identifier(key));
-        values.push(format!("'{}'", sanitize_sql_value(&value.to_string())));
+        let value_str = value.to_string();
+        let converted_value = convert_date_to_iso_format(&value_str);
+        values.push(format!("'{}'", sanitize_sql_value(&converted_value)));
     }
 
     let query = format!(
@@ -564,10 +655,12 @@ fn execute_update(
 
     let mut update_sets: Vec<String> = Vec::new();
     for (key, value) in &data_map {
+        let value_str = value.to_string();
+        let converted_value = convert_date_to_iso_format(&value_str);
         update_sets.push(format!(
             "{} = '{}'",
             sanitize_identifier(key),
-            sanitize_sql_value(&value.to_string())
+            sanitize_sql_value(&converted_value)
         ));
     }
 

src/basic/keywords/mcp_client.rs

@@ -120,7 +120,7 @@ impl Default for McpConnection {
     fn default() -> Self {
         Self {
             connection_type: ConnectionType::Http,
-            url: "http://localhost:8080".to_string(),
+            url: "http://localhost:9000".to_string(),
             port: None,
             timeout_seconds: 30,
             max_retries: 3,

src/basic/keywords/mod.rs

@@ -69,6 +69,7 @@ pub mod string_functions;
 pub mod switch_case;
 pub mod table_access;
 pub mod table_definition;
+pub mod table_migration;
 pub mod universal_messaging;
 pub mod use_tool;
 pub mod use_website;

src/basic/keywords/table_access.rs

@@ -405,6 +405,30 @@ pub fn filter_write_fields(
     }
 }
 
+/// Get column names for a table from the database schema
+pub fn get_table_columns(conn: &mut PgConnection, table_name: &str) -> Vec<String> {
+    use diesel::prelude::*;
+    use diesel::sql_types::Text;
+
+    // Define a struct for the query result
+    #[derive(diesel::QueryableByName)]
+    struct ColumnName {
+        #[diesel(sql_type = Text)]
+        column_name: String,
+    }
+
+    // Query information_schema to get column names
+    diesel::sql_query(
+        "SELECT column_name FROM information_schema.columns WHERE table_name = $1 ORDER BY ordinal_position"
+    )
+    .bind::<Text, _>(table_name)
+    .load::<ColumnName>(conn)
+    .unwrap_or_default()
+    .into_iter()
+    .map(|c| c.column_name)
+    .collect()
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

src/basic/keywords/table_definition.rs

@@ -325,7 +325,7 @@ fn parse_field_definition(
     })
 }
 
-fn map_type_to_sql(field: &FieldDefinition, driver: &str) -> String {
+pub fn map_type_to_sql(field: &FieldDefinition, driver: &str) -> String {
     let base_type = match field.field_type.as_str() {
         "string" => {
             let len = field.length.unwrap_or(255);
@@ -630,6 +630,28 @@ pub fn process_table_definitions(
         return Ok(tables);
     }
 
+    // Use schema sync for both debug and release builds (non-destructive)
+    use super::table_migration::sync_bot_tables;
+
+    info!("Running schema migration sync (non-destructive)");
+
+    match sync_bot_tables(&state, bot_id, source) {
+        Ok(result) => {
+            info!("Schema sync completed: {} created, {} altered, {} columns added",
+                result.tables_created, result.tables_altered, result.columns_added);
+
+            // If sync was successful, skip standard table creation
+            if result.tables_created > 0 || result.tables_altered > 0 {
+                return Ok(tables);
+            }
+        }
+        Err(e) => {
+            error!("Schema sync failed: {}", e);
+            // Fall through to standard table creation
+        }
+    }
+
+    // Standard table creation (for release builds or as fallback)
     for table in &tables {
         info!(
             "Processing TABLE {} ON {}",

src/basic/keywords/table_migration.rs

@@ -0,0 +1,249 @@
+/*****************************************************************************\
+|  Table Schema Migration Module
+|  Automatically syncs table.bas definitions with database schema
+\*****************************************************************************/
+
+use crate::core::shared::sanitize_identifier;
+use crate::core::shared::state::AppState;
+use diesel::prelude::*;
+use diesel::sql_query;
+use diesel::sql_types::{Text, Nullable};
+use log::{error, info, warn};
+use std::error::Error;
+use std::sync::Arc;
+use uuid::Uuid;
+
+use super::table_definition::{FieldDefinition, TableDefinition, map_type_to_sql, parse_table_definition};
+
+/// Schema migration result
+#[derive(Debug, Default)]
+pub struct MigrationResult {
+    pub tables_created: usize,
+    pub tables_altered: usize,
+    pub columns_added: usize,
+    pub errors: Vec<String>,
+}
+
+/// Column metadata from database
+#[derive(Debug, Clone)]
+struct DbColumn {
+    name: String,
+    data_type: String,
+    is_nullable: bool,
+}
+
+/// Compare and sync table schema with definition
+pub fn sync_table_schema(
+    table: &TableDefinition,
+    existing_columns: &[DbColumn],
+    create_sql: &str,
+    conn: &mut diesel::PgConnection,
+) -> Result<MigrationResult, Box<dyn Error + Send + Sync>> {
+    let mut result = MigrationResult::default();
+
+    // If no columns exist, create the table
+    if existing_columns.is_empty() {
+        info!("Creating new table: {}", table.name);
+        sql_query(create_sql).execute(conn)
+            .map_err(|e| format!("Failed to create table {}: {}", table.name, e))?;
+        result.tables_created += 1;
+        return Ok(result);
+    }
+
+    // Check for schema drift
+    let existing_col_names: std::collections::HashSet<String> =
+        existing_columns.iter().map(|c| c.name.clone()).collect();
+
+    let mut missing_columns: Vec<&FieldDefinition> = Vec::new();
+    for field in &table.fields {
+        if !existing_col_names.contains(&field.name) {
+            missing_columns.push(field);
+        }
+    }
+
+    // Add missing columns
+    if !missing_columns.is_empty() {
+        info!("Table {} is missing {} columns, adding them", table.name, missing_columns.len());
+
+        for field in &missing_columns {
+            let sql_type = map_type_to_sql(field, "postgres");
+            let column_sql = if field.is_nullable {
+                format!("ALTER TABLE {} ADD COLUMN IF NOT EXISTS {} {}",
+                    sanitize_identifier(&table.name),
+                    sanitize_identifier(&field.name),
+                    sql_type)
+            } else {
+                // For NOT NULL columns, add as nullable first then set default
+                format!("ALTER TABLE {} ADD COLUMN IF NOT EXISTS {} {}",
+                    sanitize_identifier(&table.name),
+                    sanitize_identifier(&field.name),
+                    sql_type)
+            };
+
+            info!("Adding column: {}.{} ({})", table.name, field.name, sql_type);
+            match sql_query(&column_sql).execute(conn) {
+                Ok(_) => {
+                    result.columns_added += 1;
+                    info!("Successfully added column {}.{}", table.name, field.name);
+                }
+                Err(e) => {
+                    // Check if column already exists (ignore error)
+                    let err_str = e.to_string();
+                    if !err_str.contains("already exists") && !err_str.contains("duplicate column") {
+                        let error_msg = format!("Failed to add column {}.{}: {}", table.name, field.name, e);
+                        error!("{}", error_msg);
+                        result.errors.push(error_msg);
+                    } else {
+                        info!("Column {}.{} already exists, skipping", table.name, field.name);
+                    }
+                }
+            }
+        }
+        result.tables_altered += 1;
+    }
+
+    Ok(result)
+}
+
+/// Get existing columns from a table
+pub fn get_table_columns(
+    table_name: &str,
+    conn: &mut diesel::PgConnection,
+) -> Result<Vec<DbColumn>, Box<dyn Error + Send + Sync>> {
+    let query = format!(
+        "SELECT column_name, data_type, is_nullable
+         FROM information_schema.columns
+         WHERE table_name = '{}' AND table_schema = 'public'
+         ORDER BY ordinal_position",
+        sanitize_identifier(table_name)
+    );
+
+    #[derive(QueryableByName)]
+    struct ColumnRow {
+        #[diesel(sql_type = Text)]
+        column_name: String,
+        #[diesel(sql_type = Text)]
+        data_type: String,
+        #[diesel(sql_type = Text)]
+        is_nullable: String,
+    }
+
+    let rows: Vec<ColumnRow> = match sql_query(&query).load(conn) {
+        Ok(r) => r,
+        Err(e) => {
+            // Table doesn't exist
+            return Err(format!("Table {} does not exist: {}", table_name, e).into());
+        }
+    };
+
+    Ok(rows.into_iter().map(|row| DbColumn {
+        name: row.column_name,
+        data_type: row.data_type,
+        is_nullable: row.is_nullable == "YES",
+    }).collect())
+}
+
+/// Process table definitions with schema sync for a specific bot
+pub fn sync_bot_tables(
+    state: &Arc<AppState>,
+    bot_id: Uuid,
+    source: &str,
+) -> Result<MigrationResult, Box<dyn Error + Send + Sync>> {
+    let tables = parse_table_definition(source)?;
+    let mut result = MigrationResult::default();
+
+    info!("Processing {} table definitions with schema sync for bot {}", tables.len(), bot_id);
+
+    // Get bot's database connection
+    let pool = state.bot_database_manager.get_bot_pool(bot_id)?;
+    let mut conn = pool.get()?;
+
+    for table in &tables {
+        if table.connection_name != "default" {
+            continue; // Skip external connections for now
+        }
+
+        info!("Syncing table: {}", table.name);
+
+        // Get existing columns
+        let existing_columns = match get_table_columns(&table.name, &mut conn) {
+            Ok(cols) => cols,
+            Err(_) => {
+                // Table doesn't exist yet
+                vec![]
+            }
+        };
+
+        // Generate CREATE TABLE SQL
+        let create_sql = super::table_definition::generate_create_table_sql(table, "postgres");
+
+        // Sync schema
+        match sync_table_schema(table, &existing_columns, &create_sql, &mut conn) {
+            Ok(sync_result) => {
+                result.tables_created += sync_result.tables_created;
+                result.tables_altered += sync_result.tables_altered;
+                result.columns_added += sync_result.columns_added;
+                result.errors.extend(sync_result.errors);
+            }
+            Err(e) => {
+                let error_msg = format!("Failed to sync table {}: {}", table.name, e);
+                error!("{}", error_msg);
+                result.errors.push(error_msg);
+            }
+        }
+    }
+
+    // Log summary
+    info!("Schema sync summary for bot {}: {} tables created, {} altered, {} columns added, {} errors",
+        bot_id, result.tables_created, result.tables_altered, result.columns_added, result.errors.len());
+
+    if !result.errors.is_empty() {
+        warn!("Schema sync completed with {} errors:", result.errors.len());
+        for error in &result.errors {
+            warn!("  - {}", error);
+        }
+    }
+
+    Ok(result)
+}
+
+/// Validate that all required columns exist
+pub fn validate_table_schema(
+    table_name: &str,
+    required_fields: &[FieldDefinition],
+    conn: &mut diesel::PgConnection,
+) -> Result<bool, Box<dyn Error + Send + Sync>> {
+    let existing_columns = get_table_columns(table_name, conn)?;
+    let existing_col_names: std::collections::HashSet<String> =
+        existing_columns.iter().map(|c| c.name.clone()).collect();
+
+    let mut missing = Vec::new();
+    for field in required_fields {
+        if !existing_col_names.contains(&field.name) {
+            missing.push(field.name.clone());
+        }
+    }
+
+    if !missing.is_empty() {
+        warn!("Table {} is missing columns: {:?}", table_name, missing);
+        return Ok(false);
+    }
+
+    Ok(true)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_db_column_creation() {
+        let col = DbColumn {
+            name: "test_col".to_string(),
+            data_type: "character varying".to_string(),
+            is_nullable: true,
+        };
+        assert_eq!(col.name, "test_col");
+        assert_eq!(col.is_nullable, true);
+    }
+}

src/basic/keywords/use_website.rs

@@ -826,7 +826,7 @@ mod tests {
             "docs_example_com_path"
         );
         assert_eq!(
-            sanitize_url_for_collection("http://test.site:8080"),
+            sanitize_url_for_collection("http://test.site:9000"),
             "test_site_8080"
         );
     }

src/core/bootstrap/bootstrap_manager.rs

@@ -106,9 +106,9 @@ impl BootstrapManager {
             }
         }
 
-        if pm.is_installed("postgres") {
+        if pm.is_installed("tables") {
             info!("Starting PostgreSQL...");
-            match pm.start("postgres") {
+            match pm.start("tables") {
                 Ok(_child) => {
                     info!("PostgreSQL started");
                 }

src/core/bot/kb_context.rs

@@ -436,9 +436,22 @@ pub async fn inject_kb_context(
         return Ok(());
     }
 
+    // Sanitize context to remove UTF-16 surrogate characters that can't be encoded in UTF-8
+    let sanitized_context = context_string
+        .chars()
+        .filter(|c| {
+            let cp = *c as u32;
+            !(0xD800..=0xDBFF).contains(&cp) && !(0xDC00..=0xDFFF).contains(&cp)
+        })
+        .collect::<String>();
+
+    if sanitized_context.is_empty() {
+        return Ok(());
+    }
+
     info!(
         "Injecting {} characters of KB/website context into prompt for session {}",
-        context_string.len(),
+        sanitized_context.len(),
         session_id
     );
 
@@ -447,7 +460,7 @@ pub async fn inject_kb_context(
 
         if let Some(idx) = system_msg_idx {
             if let Some(content) = messages_array[idx]["content"].as_str() {
-                let new_content = format!("{}\n{}", content, context_string);
+                let new_content = format!("{}\n{}", content, sanitized_context);
                 messages_array[idx]["content"] = serde_json::Value::String(new_content);
             }
         } else {
@@ -455,7 +468,7 @@ pub async fn inject_kb_context(
                 0,
                 serde_json::json!({
                     "role": "system",
-                    "content": context_string
+                    "content": sanitized_context
                 }),
             );
         }

src/core/bot/mod.rs

@@ -580,11 +580,20 @@ impl BotOrchestrator {
             }
         }
 
+        // Sanitize user message to remove any UTF-16 surrogate characters
+        let sanitized_message_content = message_content
+            .chars()
+            .filter(|c| {
+                let cp = *c as u32;
+                !(0xD800..=0xDBFF).contains(&cp) && !(0xDC00..=0xDFFF).contains(&cp)
+            })
+            .collect::<String>();
+
         // Add the current user message to the messages array
         if let Some(msgs_array) = messages.as_array_mut() {
             msgs_array.push(serde_json::json!({
                 "role": "user",
-                "content": message_content
+                "content": sanitized_message_content
             }));
         }
 
@@ -644,6 +653,7 @@ impl BotOrchestrator {
         let mut analysis_buffer = String::new();
         let mut in_analysis = false;
         let mut tool_call_buffer = String::new(); // Accumulate potential tool call JSON chunks
+        let mut accumulating_tool_call = false; // Track if we're currently accumulating a tool call
         let handler = llm_models::get_handler(&model);
 
         info!("[STREAM_START] Entering stream processing loop for model: {}", model);
@@ -679,12 +689,62 @@ impl BotOrchestrator {
             // ===== GENERIC TOOL EXECUTION =====
             // Add chunk to tool_call_buffer and try to parse
             // Tool calls arrive as JSON that can span multiple chunks
-            let looks_like_json = chunk.trim().starts_with('{') || chunk.trim().starts_with('[') ||
-                                   tool_call_buffer.contains('{') || tool_call_buffer.contains('[');
 
-            let chunk_in_tool_buffer = if looks_like_json {
+            // Check if this chunk contains JSON (either starts with {/[ or contains {/[)
+            let chunk_contains_json = chunk.trim().starts_with('{') || chunk.trim().starts_with('[') ||
+                                        chunk.contains('{') || chunk.contains('[');
+
+            let chunk_in_tool_buffer = if accumulating_tool_call {
+                // Already accumulating - add entire chunk to buffer
                 tool_call_buffer.push_str(&chunk);
                 true
+            } else if chunk_contains_json {
+                // Check if { appears in the middle of the chunk (mixed text + JSON)
+                let json_start = chunk.find('{').or_else(|| chunk.find('['));
+
+                if let Some(pos) = json_start {
+                    if pos > 0 {
+                        // Send the part before { as regular content
+                        let regular_part = &chunk[..pos];
+                        if !regular_part.trim().is_empty() {
+                            info!("[STREAM_CONTENT] Sending regular part before JSON: '{}', len: {}", regular_part, regular_part.len());
+                            full_response.push_str(regular_part);
+
+                            let response = BotResponse {
+                                bot_id: message.bot_id.clone(),
+                                user_id: message.user_id.clone(),
+                                session_id: message.session_id.clone(),
+                                channel: message.channel.clone(),
+                                content: regular_part.to_string(),
+                                message_type: MessageType::BOT_RESPONSE,
+                                stream_token: None,
+                                is_complete: false,
+                                suggestions: Vec::new(),
+                                context_name: None,
+                                context_length: 0,
+                                context_max_length: 0,
+                            };
+
+                            if response_tx.send(response).await.is_err() {
+                                warn!("Response channel closed");
+                                break;
+                            }
+                        }
+
+                        // Start accumulating from { onwards
+                        accumulating_tool_call = true;
+                        tool_call_buffer.push_str(&chunk[pos..]);
+                        true
+                    } else {
+                        // Chunk starts with { or [
+                        accumulating_tool_call = true;
+                        tool_call_buffer.push_str(&chunk);
+                        true
+                    }
+                } else {
+                    // Contains {/[ but find() failed - shouldn't happen, but send as regular content
+                    false
+                }
             } else {
                 false
             };
@@ -774,13 +834,15 @@ impl BotOrchestrator {
                 // Don't add tool_call JSON to full_response or analysis_buffer
                 // Clear the tool_call_buffer since we found and executed a tool call
                 tool_call_buffer.clear();
+                accumulating_tool_call = false; // Reset accumulation flag
                 // Continue to next chunk
                 continue;
             }
 
             // Clear tool_call_buffer if it's getting too large and no tool call was found
             // This prevents memory issues from accumulating JSON fragments
-            if tool_call_buffer.len() > 10000 {
+            // Increased limit to 50000 to handle large tool calls with many parameters
+            if tool_call_buffer.len() > 50000 {
                 // Flush accumulated content to client since it's too large to be a tool call
                 info!("[TOOL_EXEC] Flushing tool_call_buffer (too large, assuming not a tool call)");
                 full_response.push_str(&tool_call_buffer);
@@ -801,6 +863,7 @@ impl BotOrchestrator {
                 };
 
                 tool_call_buffer.clear();
+                accumulating_tool_call = false; // Reset accumulation flag after flush
 
                 if response_tx.send(response).await.is_err() {
                     warn!("Response channel closed");
@@ -810,7 +873,7 @@ impl BotOrchestrator {
 
             // If this chunk was added to tool_call_buffer and no tool call was found yet,
             // skip processing (it's part of an incomplete tool call JSON)
-            if chunk_in_tool_buffer && tool_call_buffer.len() <= 10000 {
+            if chunk_in_tool_buffer {
                 continue;
             }
             // ===== END TOOL EXECUTION =====

src/core/bot/tool_executor.rs

@@ -2,6 +2,7 @@
 /// Works across all LLM providers (GLM, OpenAI, Claude, etc.)
 use log::{error, info, warn};
 use serde_json::Value;
+use std::collections::HashMap;
 use std::fs::OpenOptions;
 use std::io::Write;
 use std::path::Path;
@@ -264,6 +265,8 @@ impl ToolExecutor {
         script_service.load_bot_config_params(state, bot_id);
 
         // Set tool parameters as variables in the engine scope
+        // Note: DATE parameters are now sent by LLM in ISO 8601 format (YYYY-MM-DD)
+        // The tool schema with format="date" tells the LLM to use this agnostic format
         if let Some(obj) = arguments.as_object() {
             for (key, value) in obj {
                 let value_str = match value {

src/core/package_manager/facade.rs

@@ -553,7 +553,7 @@ Store credentials in Vault:
                     r"Email Server (Stalwart):
   SMTP: {}:25
   IMAP: {}:143
-  Web:  http://{}:8080
+  Web:  http://{}:9000
 
 Store credentials in Vault:
   botserver vault put gbo/email server={} port=25 username=admin password=<your-password>",
@@ -563,11 +563,11 @@ Store credentials in Vault:
             "directory" => {
                 format!(
                     r"Zitadel Identity Provider:
-  URL: http://{}:8080
+  URL: http://{}:9000
-  Console: http://{}:8080/ui/console
+  Console: http://{}:9000/ui/console
 
 Store credentials in Vault:
-  botserver vault put gbo/directory url=http://{}:8080 client_id=<client-id> client_secret=<client-secret>",
+  botserver vault put gbo/directory url=http://{}:9000 client_id=<client-id> client_secret=<client-secret>",
                     ip, ip, ip
                 )
             }

src/core/package_manager/installer.rs

@@ -602,7 +602,7 @@ impl PackageManager {
                 post_install_cmds_windows: vec![],
                 env_vars: HashMap::new(),
                 data_download_list: Vec::new(),
-                exec_cmd: "php -S 0.0.0.0:8080 -t {{DATA_PATH}}/roundcubemail".to_string(),
+                exec_cmd: "php -S 0.0.0.0:9000 -t {{DATA_PATH}}/roundcubemail".to_string(),
                 check_cmd:
                     "curl -f -k --connect-timeout 2 -m 5 https://localhost:8300 >/dev/null 2>&1"
                         .to_string(),

src/core/package_manager/setup/email_setup.rs

@@ -203,7 +203,7 @@ impl EmailSetup {
 
         let issuer_url = dir_config["base_url"]
             .as_str()
-            .unwrap_or("http://localhost:8080");
+            .unwrap_or("http://localhost:9000");
 
         log::info!("Setting up OIDC authentication with Directory...");
         log::info!("Issuer URL: {}", issuer_url);
@@ -289,7 +289,7 @@ protocol = "imap"
 tls.implicit = true
 
 [server.listener."http"]
-bind = ["0.0.0.0:8080"]
+bind = ["0.0.0.0:9000"]
 protocol = "http"
 
 [storage]
@@ -315,7 +315,7 @@ store = "sqlite"
             r#"
 [directory."oidc"]
 type = "oidc"
-issuer = "http://localhost:8080"
+issuer = "http://localhost:9000"
 client-id = "{{CLIENT_ID}}"
 client-secret = "{{CLIENT_SECRET}}"
 

src/core/secrets/mod.rs

@@ -381,7 +381,7 @@ impl SecretsManager {
                 secrets.insert("token".into(), String::new());
             }
             SecretPaths::ALM => {
-                secrets.insert("url".into(), "http://localhost:8080".into());
+                secrets.insert("url".into(), "http://localhost:9000".into());
                 secrets.insert("username".into(), String::new());
                 secrets.insert("password".into(), String::new());
             }

src/core/shared/test_utils.rs

@@ -249,13 +249,13 @@ impl Default for TestAppStateBuilder {
 #[cfg(feature = "directory")]
 pub fn create_mock_auth_service() -> AuthService {
     let config = ZitadelConfig {
-        issuer_url: "http://localhost:8080".to_string(),
+        issuer_url: "http://localhost:9000".to_string(),
-        issuer: "http://localhost:8080".to_string(),
+        issuer: "http://localhost:9000".to_string(),
         client_id: "mock_client_id".to_string(),
         client_secret: "mock_client_secret".to_string(),
         redirect_uri: "http://localhost:3000/callback".to_string(),
         project_id: "mock_project_id".to_string(),
-        api_url: "http://localhost:8080".to_string(),
+        api_url: "http://localhost:9000".to_string(),
         service_account_key: None,
     };
 

src/core/shared/utils.rs

@@ -556,3 +556,96 @@ fn estimate_chars_per_token(model: &str) -> usize {
         4 // Default conservative estimate
     }
 }
+
+/// Convert date string from user locale format to ISO format (YYYY-MM-DD) for PostgreSQL.
+///
+/// The LLM automatically formats dates according to the user's language/idiom based on:
+/// 1. The conversation context (user's language)
+/// 2. The PARAM LIKE example (e.g., "15/12/2026" for DD/MM/YYYY)
+///
+/// This function handles the most common formats:
+/// - ISO: YYYY-MM-DD (already in ISO, returned as-is)
+/// - Brazilian/Portuguese: DD/MM/YYYY or DD/MM/YY
+/// - US/English: MM/DD/YYYY or MM/DD/YY
+///
+/// If the value doesn't match any date pattern, returns it unchanged.
+///
+/// NOTE: This function does NOT try to guess ambiguous formats.
+/// The LLM is responsible for formatting dates correctly based on user language.
+/// The PARAM declaration's LIKE example tells the LLM the expected format.
+///
+/// # Arguments
+/// * `value` - The date string to convert (as provided by the LLM)
+///
+/// # Returns
+/// ISO formatted date string (YYYY-MM-DD) or original value if not a recognized date
+pub fn convert_date_to_iso_format(value: &str) -> String {
+    let value = value.trim();
+
+    // Already in ISO format (YYYY-MM-DD) - return as-is
+    if value.len() == 10 && value.chars().nth(4) == Some('-') && value.chars().nth(7) == Some('-') {
+        let parts: Vec<&str> = value.split('-').collect();
+        if parts.len() == 3
+            && parts[0].len() == 4
+            && parts[1].len() == 2
+            && parts[2].len() == 2
+            && parts[0].chars().all(|c| c.is_ascii_digit())
+            && parts[1].chars().all(|c| c.is_ascii_digit())
+            && parts[2].chars().all(|c| c.is_ascii_digit())
+        {
+            if let (Ok(year), Ok(month), Ok(day)) =
+                (parts[0].parse::<u32>(), parts[1].parse::<u32>(), parts[2].parse::<u32>())
+            {
+                if month >= 1 && month <= 12 && day >= 1 && day <= 31 && year >= 1900 && year <= 2100 {
+                    return value.to_string();
+                }
+            }
+        }
+    }
+
+    // Handle slash-separated formats: DD/MM/YYYY or MM/DD/YYYY
+    // We need to detect which format based on the PARAM declaration's LIKE example
+    // For now, default to DD/MM/YYYY (Brazilian format) as this is the most common for this bot
+    // TODO: Pass language/idiom from session to determine correct format
+    if value.len() >= 8 && value.len() <= 10 {
+        let parts: Vec<&str> = value.split('/').collect();
+        if parts.len() == 3 {
+            let all_numeric = parts[0].chars().all(|c| c.is_ascii_digit())
+                && parts[1].chars().all(|c| c.is_ascii_digit())
+                && parts[2].chars().all(|c| c.is_ascii_digit());
+
+            if all_numeric {
+                // Parse the three parts
+                let a = parts[0].parse::<u32>().ok();
+                let b = parts[1].parse::<u32>().ok();
+                let c = if parts[2].len() == 2 {
+                    // Convert 2-digit year to 4-digit
+                    parts[2].parse::<u32>().ok().map(|y| {
+                        if y < 50 {
+                            2000 + y
+                        } else {
+                            1900 + y
+                        }
+                    })
+                } else {
+                    parts[2].parse::<u32>().ok()
+                };
+
+                if let (Some(first), Some(second), Some(third)) = (a, b, c) {
+                    // Default: DD/MM/YYYY format (Brazilian/Portuguese)
+                    // The LLM should format dates according to the user's language
+                    // and the PARAM LIKE example (e.g., "15/12/2026" for DD/MM/YYYY)
+                    let (year, month, day) = (third, second, first);
+
+                    // Validate the determined date
+                    if day >= 1 && day <= 31 && month >= 1 && month <= 12 && year >= 1900 && year <= 2100 {
+                        return format!("{:04}-{:02}-{:02}", year, month, day);
+                    }
+                }
+            }
+        }
+    }
+
+    // Not a recognized date pattern, return unchanged
+    value.to_string()
+}

src/core/urls.rs

@@ -479,7 +479,7 @@ impl ApiUrls {
 pub struct InternalUrls;
 
 impl InternalUrls {
-    pub const DIRECTORY_BASE: &'static str = "http://localhost:8080";
+    pub const DIRECTORY_BASE: &'static str = "http://localhost:9000";
     pub const DATABASE: &'static str = "postgres://localhost:5432";
     pub const CACHE: &'static str = "redis://localhost:6379";
     pub const DRIVE: &'static str = "https://localhost:9000";

src/email/messages.rs

@@ -65,8 +65,8 @@ fn is_tracking_pixel_enabled(state: &Arc<AppState>, bot_id: Option<Uuid>) -> boo
 fn inject_tracking_pixel(html_body: &str, tracking_id: &str, state: &Arc<AppState>) -> String {
     let config_manager = crate::core::config::ConfigManager::new(state.conn.clone());
     let base_url = config_manager
-        .get_config(&Uuid::nil(), "server-url", Some("http://localhost:8080"))
+        .get_config(&Uuid::nil(), "server-url", Some("http://localhost:9000"))
-        .unwrap_or_else(|_| "http://localhost:8080".to_string());
+        .unwrap_or_else(|_| "http://localhost:9000".to_string());
 
     let pixel_url = format!("{}/api/email/tracking/pixel/{}", base_url, tracking_id);
     let pixel_html = format!(

src/email/tracking.rs

@@ -31,8 +31,8 @@ pub fn is_tracking_pixel_enabled(state: &Arc<AppState>, bot_id: Option<Uuid>) ->
 pub fn inject_tracking_pixel(html_body: &str, tracking_id: &str, state: &Arc<AppState>) -> String {
     let config_manager = crate::core::config::ConfigManager::new(state.conn.clone());
     let base_url = config_manager
-        .get_config(&Uuid::nil(), "server-url", Some("http://localhost:8080"))
+        .get_config(&Uuid::nil(), "server-url", Some("http://localhost:9000"))
-        .unwrap_or_else(|_| "http://localhost:8080".to_string());
+        .unwrap_or_else(|_| "http://localhost:9000".to_string());
 
     let pixel_url = format!("{}/api/email/tracking/pixel/{}", base_url, tracking_id);
     let pixel_html = format!(

src/llm/claude.rs

@@ -232,6 +232,16 @@ impl ClaudeClient {
         (system_prompt, claude_messages)
     }
 
+    /// Sanitizes a string by removing invalid UTF-8 surrogate characters
+    fn sanitize_utf8(input: &str) -> String {
+        input.chars()
+            .filter(|c| {
+                let cp = *c as u32;
+                !(0xD800..=0xDBFF).contains(&cp) && !(0xDC00..=0xDFFF).contains(&cp)
+            })
+            .collect()
+    }
+
     pub fn build_messages(
         system_prompt: &str,
         context_data: &str,
@@ -241,15 +251,15 @@ impl ClaudeClient {
         let mut system_parts = Vec::new();
 
         if !system_prompt.is_empty() {
-            system_parts.push(system_prompt.to_string());
+            system_parts.push(Self::sanitize_utf8(system_prompt));
         }
         if !context_data.is_empty() {
-            system_parts.push(context_data.to_string());
+            system_parts.push(Self::sanitize_utf8(context_data));
         }
 
         for (role, content) in history {
             if role == "episodic" || role == "compact" {
-                system_parts.push(format!("[Previous conversation summary]: {content}"));
+                system_parts.push(format!("[Previous conversation summary]: {}", Self::sanitize_utf8(content)));
             }
         }
 
@@ -270,7 +280,8 @@ impl ClaudeClient {
             };
 
             if let Some(norm_role) = normalized_role {
-                if content.is_empty() {
+                let sanitized_content = Self::sanitize_utf8(content);
+                if sanitized_content.is_empty() {
                     continue;
                 }
 
@@ -278,14 +289,14 @@ impl ClaudeClient {
                     if let Some(last_msg) = messages.last_mut() {
                         let last_msg: &mut ClaudeMessage = last_msg;
                         last_msg.content.push_str("\n\n");
-                        last_msg.content.push_str(content);
+                        last_msg.content.push_str(&sanitized_content);
                         continue;
                     }
                 }
 
                 messages.push(ClaudeMessage {
                     role: norm_role.clone(),
-                    content: content.clone(),
+                    content: sanitized_content,
                 });
                 last_role = Some(norm_role);
             }

src/llm/glm.rs

@@ -116,6 +116,16 @@ impl GLMClient {
         // GLM/z.ai uses /chat/completions (not /v1/chat/completions)
         format!("{}/chat/completions", self.base_url)
     }
+
+    /// Sanitizes a string by removing invalid UTF-8 surrogate characters
+    fn sanitize_utf8(input: &str) -> String {
+        input.chars()
+            .filter(|c| {
+                let cp = *c as u32;
+                !(0xD800..=0xDBFF).contains(&cp) && !(0xDC00..=0xDFFF).contains(&cp)
+            })
+            .collect()
+    }
 }
 
 #[async_trait]
@@ -183,11 +193,6 @@ impl LLMProvider for GLMClient {
         key: &str,
         tools: Option<&Vec<Value>>,
     ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-        // DEBUG: Log what we received
-        info!("[GLM_DEBUG] config type: {}", config);
-        info!("[GLM_DEBUG] prompt: '{}'", prompt);
-        info!("[GLM_DEBUG] config as JSON: {}", serde_json::to_string_pretty(config).unwrap_or_default());
-
         // config IS the messages array directly, not nested
         let messages = if let Some(msgs) = config.as_array() {
             // Convert messages from config format to GLM format
@@ -195,25 +200,23 @@ impl LLMProvider for GLMClient {
                 .filter_map(|m| {
                     let role = m.get("role")?.as_str()?;
                     let content = m.get("content")?.as_str()?;
-                    info!("[GLM_DEBUG] Processing message - role: {}, content: '{}'", role, content);
+                    let sanitized = Self::sanitize_utf8(content);
-                    if !content.is_empty() {
+                    if !sanitized.is_empty() {
                         Some(GLMMessage {
                             role: role.to_string(),
-                            content: Some(content.to_string()),
+                            content: Some(sanitized),
                             tool_calls: None,
                         })
                     } else {
-                        info!("[GLM_DEBUG] Skipping empty content message");
                         None
                     }
                 })
                 .collect::<Vec<_>>()
         } else {
             // Fallback to building from prompt
-            info!("[GLM_DEBUG] No array found, using prompt: '{}'", prompt);
             vec![GLMMessage {
                 role: "user".to_string(),
-                content: Some(prompt.to_string()),
+                content: Some(Self::sanitize_utf8(prompt)),
                 tool_calls: None,
             }]
         };
@@ -223,8 +226,6 @@ impl LLMProvider for GLMClient {
             return Err("No valid messages in request".into());
         }
 
-        info!("[GLM_DEBUG] Final GLM messages count: {}", messages.len());
-
         // Use glm-4.7 for tool calling support
         // GLM-4.7 supports standard OpenAI-compatible function calling
         let model_name = if model == "glm-4" { "glm-4.7" } else { model };
@@ -249,10 +250,6 @@ impl LLMProvider for GLMClient {
         let url = self.build_url();
         info!("GLM streaming request to: {}", url);
 
-        // Log the exact request being sent
-        let request_json = serde_json::to_string_pretty(&request).unwrap_or_default();
-        info!("GLM request body: {}", request_json);
-
         let response = self
             .client
             .post(&url)
@@ -292,18 +289,13 @@ impl LLMProvider for GLMClient {
 
                 if line.starts_with("data: ") {
                     let json_str = line[6..].trim();
-                    info!("[GLM_SSE] Received SSE line ({} chars): {}", json_str.len(), json_str);
                     if let Ok(chunk_data) = serde_json::from_str::<Value>(json_str) {
                         if let Some(choices) = chunk_data.get("choices").and_then(|c| c.as_array()) {
                             for choice in choices {
-                                info!("[GLM_SSE] Processing choice");
                                 if let Some(delta) = choice.get("delta") {
-                                    info!("[GLM_SSE] Delta: {}", serde_json::to_string(delta).unwrap_or_default());
-
                                     // Handle tool_calls (GLM-4.7 standard function calling)
                                     if let Some(tool_calls) = delta.get("tool_calls").and_then(|t| t.as_array()) {
                                         for tool_call in tool_calls {
-                                            info!("[GLM_SSE] Tool call detected: {}", serde_json::to_string(tool_call).unwrap_or_default());
                                             // Send tool_calls as JSON for the calling code to process
                                             let tool_call_json = serde_json::json!({
                                                 "type": "tool_call",
@@ -323,7 +315,6 @@ impl LLMProvider for GLMClient {
                                     // This makes GLM behave like OpenAI-compatible APIs
                                     if let Some(content) = delta.get("content").and_then(|c| c.as_str()) {
                                         if !content.is_empty() {
-                                            info!("[GLM_TX] Sending to channel: '{}'", content);
                                             match tx.send(content.to_string()).await {
                                                 Ok(_) => {},
                                                 Err(e) => {
@@ -331,11 +322,9 @@ impl LLMProvider for GLMClient {
                                                 }
                                             }
                                         }
-                                    } else {
-                                        info!("[GLM_SSE] No content field in delta");
                                     }
                                 } else {
-                                    info!("[GLM_SSE] No delta in choice");
+                                    // No delta in choice
                                 }
                                 if let Some(reason) = choice.get("finish_reason").and_then(|r| r.as_str()) {
                                     if !reason.is_empty() {

src/llm/mod.rs

@@ -185,6 +185,17 @@ impl OpenAIClient {
         }
     }
 
+    /// Sanitizes a string by removing invalid UTF-8 surrogate characters
+    /// that cannot be encoded in valid UTF-8 (surrogates are only valid in UTF-16)
+    fn sanitize_utf8(input: &str) -> String {
+        input.chars()
+            .filter(|c| {
+                let cp = *c as u32;
+                !(0xD800..=0xDBFF).contains(&cp) && !(0xDC00..=0xDFFF).contains(&cp)
+            })
+            .collect()
+    }
+
     pub fn build_messages(
         system_prompt: &str,
         context_data: &str,
@@ -194,19 +205,19 @@ impl OpenAIClient {
         if !system_prompt.is_empty() {
             messages.push(serde_json::json!({
                 "role": "system",
-                "content": system_prompt
+                "content": Self::sanitize_utf8(system_prompt)
             }));
         }
         if !context_data.is_empty() {
             messages.push(serde_json::json!({
                 "role": "system",
-                "content": context_data
+                "content": Self::sanitize_utf8(context_data)
             }));
         }
         for (role, content) in history {
             messages.push(serde_json::json!({
                 "role": role,
-                "content": content
+                "content": Self::sanitize_utf8(content)
             }));
         }
         serde_json::Value::Array(messages)
@@ -747,10 +758,10 @@ mod tests {
     fn test_openai_client_new_custom_url() {
         let client = OpenAIClient::new(
             "test_key".to_string(),
-            Some("http://localhost:8080".to_string()),
+            Some("http://localhost:9000".to_string()),
             None,
         );
-        assert_eq!(client.base_url, "http://localhost:8080");
+        assert_eq!(client.base_url, "http://localhost:9000");
     }
 
     #[test]

src/security/cors.rs

@@ -101,13 +101,13 @@ impl CorsConfig {
         Self {
             allowed_origins: vec![
                 "http://localhost:3000".to_string(),
-                "http://localhost:8080".to_string(),
+                "http://localhost:9000".to_string(),
                 "http://localhost:8300".to_string(),
                 "http://127.0.0.1:3000".to_string(),
-                "http://127.0.0.1:8080".to_string(),
+                "http://127.0.0.1:9000".to_string(),
                 "http://127.0.0.1:8300".to_string(),
                 "https://localhost:3000".to_string(),
-                "https://localhost:8080".to_string(),
+                "https://localhost:9000".to_string(),
                 "https://localhost:8300".to_string(),
             ],
             allowed_methods: vec![
@@ -576,7 +576,7 @@ mod tests {
         assert!(is_localhost_origin("http://localhost:3000"));
         assert!(is_localhost_origin("https://localhost:8443"));
         assert!(is_localhost_origin("http://127.0.0.1"));
-        assert!(is_localhost_origin("http://127.0.0.1:8080"));
+        assert!(is_localhost_origin("http://127.0.0.1:9000"));
         assert!(!is_localhost_origin("http://example.com"));
     }