Shipping update - 2026-03-16

Highlights

• AI-assisted workflows now recover better when very large tool responses exceed preflight token-count limits. Instead of failing immediately, the system applies controlled truncation and retries, then surfaces clearer diagnostics if limits are still exceeded.
• Query generation now respects user-specified log index targets more precisely and avoids inventing index values. This reduces wrong-target queries when users explicitly choose a data slice.
• Natural-language time parsing has been strengthened across chat and telemetry flows, including clearer precedence rules between absolute timestamps and relative windows.
• Post-login routing for authorize flows now detects safe direct endpoints and bypasses wrapper redirects when appropriate, reducing failure-prone iframe-style handoffs.
• Chat error handling now treats unresolvable time expressions as a first-class recovery path and presents clearer next-step actions when a conversation cannot continue safely.

What this enables

• Teams can run heavier investigations without abrupt AI-session failures when responses are large.
• Operators can ask for logs with explicit index intent and get results aligned with that intent.
• Time-bound questions produce more consistent outcomes, especially when users mix natural language and explicit timestamps.
• Sign-in to connected workflows completes with fewer redirect edge cases.
• Users get clearer remediation paths in chat instead of generic dead-end failures.

Technical wins

• Added bounded fallback logic around token preflight failures, including retry paths and structured diagnostics for oversized requests.
• Threaded explicit index handling through query tooling, prompt templates, and deep-link resolution while preserving backward-compatible fallback behavior.
• Hardened temporal parsing and validation rules, including better bounds checking and clearer error semantics across multiple surfaces.
• Refactored post-login redirect logic into tested utility code so same-origin authorize paths can be handled directly and safely.
• Expanded UI and chat-state error categorization so recovery actions can be triggered for more than one failure class, not only context-limit cases.