Bugs Fixed:

1) Flows not starting on a keyword :

If a keyword or periodic trigger pointed to a flow that had no published version, the incoming message failed – the contact/user received no reply and the flow never started.
Glific now skips that trigger and continues processing the message normally, so the rest of the flows are unaffected.

2) Flow messages breaking on a mistyped variable :

When a message in a flow referred to a variable one level deeper than it existed – for example @results.age.value where age was already a plain value – the flow stopped mid-conversation.
These references now resolve to nothing and the flow continues instead of erroring out.

3) Duplicate messages after a wait for response :

Under heavy load, a flow paused at a “wait for response” step could occasionally be resumed twice for the same contact, which meant the contact received the next message twice.
Flow wake-ups are now processed one at a time per organisation, so this can no longer happen.

4) Duplicate rows in BigQuery exports :

The clean-up that removes duplicate rows from BigQuery had stopped running because of an invalid query, which meant duplicates were building up in the messages, flow_contexts, flow_results, wa_messages and messages_media tables.
If you have dashboards or reports built on these tables, message counts and flow results may have been inflated during this period. The clean-up is now working again.

5) WhatsApp group messages being missed :

In groups where members’ phone numbers are not visible to Glific, incoming messages could be dropped or retried repeatedly.
Glific now recovers the sender’s number where it can, and skips it quietly where it cannot, so group message capture is more reliable.

6) Improving message reliability:

In early September, we experienced a messaging incident that affected multiple organisations.
We identified gaps in message processing, monitoring, database performance, and incident communication.

Since then, we have:

• Moved message processing to organisation-partitioned queues to reduce cross-org impact.
• Improved database and contact import performance.
• Added plans for better error monitoring, alerting, and org-level visibility.
• Started improving how we handle long-running flows and communicate during incidents.

We’re continuing to work on the remaining corrective actions to make message delivery more reliable and resilient.

Read the detailed incident report for the full timeline, impact, and corrective actions.

Enhancement Done:


1) AI Evaluations experience:

Glific’s AI Evaluations moved beyond simple similarity matching. Evaluations now use a built-in AI judge that scores answers against three measures, and the Evaluations tab was rebuilt with a fuller set of tools for reviewing and improving results.

Highlights:

• Introduced a native AI judge that scores answers on adherence to ground truth, knowledge base, and prompt.

• Rebuilt the Evaluations tab with Run/History views, a chat sandbox for trying an assistant directly, and a Knowledge Base view.

• Added CSV export for evaluation run history.

• Added a tooltip showing the judge’s reasoning behind each score.

• Let evaluations run independently per assistant version, instead of one run blocking all others.

• Raised the limit on Golden Q&A uploads to support longer, multi-line answers.
  

2) WhatsApp templates with the new Template Library:

Creating a new HSM template is now easier with a library of Meta’s pre-approved templates to start from, rather than writing every template from scratch.

Highlights:

• Added a Template Library to browse pre-approved WhatsApp templates grouped by use case.

• Added a preview and a “Create from template” option that pre-fills the template form.

• We have detailed documentation on this please do give it a read here. 

Documentation Update:

New Documentation Added:

1) WhatsApp Business Profile Info:

A new detailed explanation has been added here covering how to update the business profile shown to contacts, including description, address, email, and website, through the Gupshup dashboard.

Revamped Existing Documentation:

1) Label the Incoming Message:

Rewritten with clearer guidance on how the node works and best practices, including placing it after a Wait for Response node so it labels the reply you intend.

2) Wait for the contact to respond :

We have made a few improvements to make text-based conditions easier to understand and use. You can now use has all of the words  to match responses that contain multiple specific words. We have also added clearer guidance for using has multiple values, including validation rules, comparison examples, and a visual reference to help you set up these conditions correctly.Please do give it a read here.