Skip to main content
Auto-apply rules let a label tag itself. Instead of picking labels by hand, you attach one or more regex rules to a label; when a message you send matches, Synti applies the label — and, for typed labels, extracts a value from the match. It’s the fastest way to keep sessions organized as you work, especially for recurring identifiers like ticket numbers, order IDs, or amounts.

​
Defining rules

Rules live in the autoRules array on a label inside your workspace’s label config:
Each rule has these fields:
Here a message containing JIRA-1234 attaches the label as ticket::JIRA-1234.

​
When rules run

Synti evaluates rules automatically whenever you send a message, including messages that were queued and sent later. It only scans your messages — assistant replies and tool output are never matched, so the agent’s own text can’t trigger a label. Before matching, Synti strips fenced code blocks (```) and inline code (`) from the message. That keeps sample data, logs, and snippets from tripping rules they weren’t meant to.

​
How matches are processed

All rules across all labels are evaluated together against the cleaned message. A few guardrails shape the result:
  • Match cap. A single message produces at most 10 matches, which prevents a bulk paste from exploding into dozens of labels.
  • Deduplication. If a session already carries a given label + value pair, a repeat match won’t add it again.
  • Aggregation. Multiple rules on one label all contribute; their matches combine.

​
Value normalization

Extracted values are normalized according to the label’s valueType:
  • string — passed through unchanged.
  • number — currency symbols and commas are stripped, and shorthand suffixes expand, so 1.5M becomes 1500000 and 50k becomes 50000.
  • date — passed through unchanged (author it in ISO YYYY-MM-DD form).

​
Validation and safety

Synti validates patterns both when you save the config and again at runtime. An invalid regex is logged and skipped rather than breaking evaluation, and save-time validation additionally flags patterns vulnerable to catastrophic backtracking (ReDoS) so a slow expression can’t stall matching.
Write specific patterns with word boundaries (\b) and anchored formats. A rule like \b(ORD-\d{6})\b is both faster and less likely to produce false positives than a loose one.
Auto-rules only apply labels — they never remove them. To take a label off a session, remove it manually or ask the agent.