fix(postgres): sanitize all non-identifier chars in table names

rag/engine/postgres.go

@@ -97,10 +97,21 @@ func NewPostgresDBCollection(collectionName, databaseURL string, openaiClient *o
 }
 
 func sanitizeTableName(name string) string {
-	// Replace invalid characters with underscores
-	name = strings.ReplaceAll(name, "-", "_")
-	name = strings.ReplaceAll(name, ".", "_")
-	name = strings.ReplaceAll(name, " ", "_")
+	// Replace every character that is not a valid PostgreSQL identifier
+	// character with an underscore. Allowlisting (rather than stripping a
+	// hardcoded set such as '-', '.', ' ') guarantees the result is a legal
+	// unquoted identifier whatever the collection name contains: the ':'
+	// namespace separator used for per-user collections (e.g. the synthetic
+	// "legacy-api-key:<agent>" name) previously slipped through and produced
+	// "syntax error at or near ':'" on CREATE TABLE.
+	name = strings.Map(func(r rune) rune {
+		switch {
+		case r >= 'a' && r <= 'z', r >= 'A' && r <= 'Z', r >= '0' && r <= '9', r == '_':
+			return r
+		default:
+			return '_'
+		}
+	}, name)
 	// Ensure it starts with a letter
 	if len(name) > 0 && (name[0] < 'a' || name[0] > 'z') && (name[0] < 'A' || name[0] > 'Z') {
 		name = "col_" + name

rag/engine/sanitize_test.go

@@ -0,0 +1,42 @@
+package engine
+
+import (
+	"regexp"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+// validPostgresIdentifier matches an unquoted PostgreSQL identifier: it must
+// start with a letter or underscore and contain only letters, digits and
+// underscores. Anything else triggers a SQL syntax error when interpolated
+// into DDL such as CREATE TABLE.
+var validPostgresIdentifier = regexp.MustCompile(`^[A-Za-z_][A-Za-z0-9_]*$`)
+
+var _ = Describe("sanitizeTableName", func() {
+	It("strips the ':' namespace separator used for per-user collections", func() {
+		// Regression for LocalAI #10375: agents created under a legacy API key
+		// get the collection name "legacy-api-key:<agent>", and the colon broke
+		// PostgreSQL table creation (ERROR: syntax error at or near ":").
+		got := sanitizeTableName("legacy-api-key:LiteraryResearch")
+		Expect(got).NotTo(ContainSubstring(":"))
+		Expect(got).To(MatchRegexp(validPostgresIdentifier.String()))
+	})
+
+	It("produces a valid identifier for any input", func() {
+		for _, in := range []string{
+			"legacy-api-key:LiteraryResearch",
+			"user@example.com/notes",
+			"a b.c-d:e",
+			"123starts-with-digit",
+			"with$pecial%chars!",
+		} {
+			Expect(sanitizeTableName(in)).To(MatchRegexp(validPostgresIdentifier.String()),
+				"input %q should sanitize to a valid PostgreSQL identifier", in)
+		}
+	})
+
+	It("still maps the previously-handled characters to underscores", func() {
+		Expect(sanitizeTableName("my-collection.name here")).To(Equal("documents_my_collection_name_here"))
+	})
+})