6.2.0
📝 Release Information
- Release Date: 02/11/2026
- End of Sale: November 2027
- End of Support: November 2029
- Runtime Used: .NET 10 (End of Support : November 2028)
- Dependencies:
- Secure Work From Home Server (>= 6.0.0)
- Neuron Edge - Server (>= 1.3.6) (if server side STT is enabled)
- Cortex AI (>= 1.0.0) : If Summary, Sentiment, AQM integration is enabled in Secure Assist.
- IAM (>= 5.0.0)
- TetherfiAPI (>=2.0.2)
🚀 What's New
SWFH-1954, SWFH-1951 - Neuron Edge is now part of Secure Assist: the SWFH worker service runs it alongside the SWFH UI, and it shows the live call transcript with profanity detection, the call summary and the Agent Sensei scores.
SWFH-2095, SWFH-2104 - New Neuron Edge call assist window, with call summaries created locally on the agent's PC. Neuron Edge starts in the background with the SWFH UI.
SWFH-2105 - The local summarization model is downloaded from the SWFH Server instead of being installed with the MSI. The microphone stays disabled until the model is ready.
SWFH-2053 - Call summaries are shown as key points, with the call intent and the action points or promises made during the call.
SWFH-2055 - Neuron Edge button on the SWFH UI toolbar to start and end call capture and to open Neuron Edge in front of other windows. The button stays disabled until the agent has logged in.
SWFH-2056 - Neuron Edge chat view shows customer messages on the left and agent messages on the right, with the speaker and the message time.
SWFH-2054 - Real-time supervisor alerts for non-compliant words spoken during a call.
⭐ Improvements
SWFH-2120 - Reduced CPU and memory usage of the SWFH UI, fixed memory leaks, and fixed chat and screen glance issues.
SWFH-2052 - Automatic call triggers from the agent's call page are now configured with the SA_PLAYWRIGHT_CONFIG OCM key, and the call end is no longer delayed when the call's active time cannot be read from the page.
🐞 Bug Fixes
SWFH-2116 - Violations stored offline were not sent to the server, and the offline cache kept growing with high CPU usage before login. Neuron Edge call summaries and transcripts now reach the server.
SWFH-2119 - Login and Switch User issues when switching between vision and no-vision modes.
PRODQA-4574 - Neuron Edge showed an incorrect Secure Assist version.
SWFH-2058 - The Neuron Edge call history froze when a call changed from Processing to Completed.
🔧 Application Configuration Changes
SA_LOCAL_CALL_SUMMARIZATION in OCM : Settings for the call summaries created on the agent's PC. Summaries are always created locally, so the feature only needs a value to change the defaults. The optional value is a JSON object with any of these properties: Sample value (use the GPU, 4 threads, keep the model loaded):
- ModelFile : Model file name (default LocalSummarizationModel.gguf).
- Threads : CPU threads used for summarization (default 2; 0 = automatic, between 2 and 4).
- UseGpu : Use the GPU for summarization, falling back to the CPU when the PC has no Vulkan driver (default false).
- MaxOutputTokens : Maximum length of the generated summary (default 600).
- TimeoutSeconds : Time limit for one summary (default 300).
- IdleUnloadMinutes : Minutes without a summary after which the model is unloaded to free memory (default 10; 0 = keep it loaded).
- UseCommitmentHints : Help the model find the promises the agent made during the call (default true).
- DisableThinking : Turn off the reasoning mode of models that have one (default true).
- PauseAboveWords : Pause the live transcript once a call goes over this many words, so the summary does not take too long (default 1400; 0 = never pause).
or, as plain text: Balance Enquiry, Card Replacement, Loan Application, Address Change, Complaint
SA_CALL_INTENTS in OCM (Team Template) : The team's predefined call intents. When enabled with a value, the summary's intent is the matching one, or blank when none matches; otherwise the model describes the intent in its own words. Value: a JSON array of strings, or one intent per line (comma-separated when on a single line). Sample value:
SA_NON_COMPLIANT_WORDS in OCM (Team Template) : Words and phrases that raise a real-time supervisor alert when spoken during a call. Value: same format as SA_CALL_INTENTS. Sample value:
SA_PROFANITY_DETECTION in OCM : Turns profanity detection in Neuron Edge on or off. Detection is on when the key is not configured. Sample: enable or disable the feature; no value is needed.
SA_PROFANITY_WORDS in OCM : Extra words for the profanity filter, on top of its built-in list. Value: same format as SA_CALL_INTENTS. Sample value: ["idiot", "stupid", "shut up"]
SA_CALL_SUMMARIZATION in OCM (changed) : Agent Sensei is no longer used for call summaries, only for scores. The value holds the Agent Sensei endpoint settings (QueueUrl to submit a call, PollUrl to fetch its result). Sample value:
SA_PLAYWRIGHT_CONFIG in OCM (replaces SA_DOM_DETAILS) : JSON that describes the agent's call page for call page monitoring: the page URL, the CSS selectors of the End Call button and of the call details on the page, and the button's label. Call page monitoring does not start when the key is missing or its value is not valid JSON. SA_DOM_DETAILS is no longer read: move its value to SA_PLAYWRIGHT_CONFIG and remove the outer { "DomDetails": ... } wrapper, so the value is the inner object itself. Sample value:
Local summarization model : Every agent with Neuron Edge enabled (TRS_MANUAL_AGENT_CALL_CONTROL and Secure Assist) downloads the model from the SWFH Server to %APPDATA%\RemoteSecurity\Models\LocalSummarizer\. The model, its download link and where to place it on the server are described in the Secure Assist Server 6.2.0-alpha.1 release notes.