May 8, 2026

Changelog — Weeks of Apr 27 & May 4, 2026

New features and improvements

Added a Microsoft Graph TEAMS_AND_CHATS permission group
Calendar tools now let agents set a custom description on service calls, schedule entries, and appointments instead of only the ticket subject.
Expanded ConnectWise Automate API coverage to v2.5.7 — 96 operations across 68 endpoints.
Adaptive rate limiter on ConnectWise Asio RMM — learns from response headers and downgrades automatically on 429 responses.
Customer replies on a ticket now bypass the 180-second callback dedup window across ConnectWise, Autotask, and Halo, so agents react immediately when a customer responds.
Event history rows include a deep link to the ticket in your PSA.

Added a Stop Execution button on running workflows — cancel a misfiring run from the dashboard with a confirmation prompt.
Workflow update triggers now support ticket type as a watched field.
Workflow filter values are validated before save so you can't accidentally ship a workflow that references missing values.

Redesigned Create-from-template dialog — every workflow and agent has a plain-English name, scope meter, and integration requirements grouped by category so you can satisfy them with whichever vendor you have.
Analytics page now shows the in-progress (current) bucket and week-range labels for clearer trend reading.
"Enabled only" filter on the permission matrix now applies to integration permissions too, hiding noise from disabled integrations.

Sandbox scripts can now generate PDFs via reportlab + matplotlib (replacing fpdf2) — better fonts, charts, and layout control.
Sandbox emails sent via neo.client support file attachments end-to-end.

Agentic workflows are now billed on the resolved toolbox — auto-injected tools count toward the max tier, matching what's actually executed.

LLM prompt cache retention extended to 24 hours on supported models, dramatically reducing cold-start cache misses.
Compacted the Dispatch agenda table to reduce prompt tokens per run.
Capped PSA, RMM, and External API tool response sizes to protect agent input budget on large datasets.

Suppressed transient ITGlue auth-failure emails — only final-attempt failures notify, eliminating false positives during brief outages.
Per-tenant deduplication on concurrency-cap Teams notifications so one busy tenant doesn't suppress alerts for unrelated tenants.

Fixed RMM find_configurations not tagging devices with their source system, causing ambiguous matches across multiple RMMs.
Fixed empty-string ticket IDs bypassing the entityless TIL lookup path.
Fixed master-agent stalling on COMMENTED TIL approvals — added a code-level safety net.
Fixed agents accepting string query_parameters instead of objects in generic API tools — they now get a clear validation error.
Fixed the time-entry tool falling back to a "Time logged" placeholder when no summary was provided.
Fixed RMM dynamic-script execution running multiple times within the dedup TTL.

Fixed CW PATCH failures when Service Item/Type/SubType IDs didn't match the destination board — values now carry over by name.
Fixed assigner using the pre-update board ID after a board change in the same workflow.
Fixed CW custom-fields JSON Patch guidance and added CW_ITEM validation.
Fixed triage matching CW predefined options by board-scoped name instead of global ID.

Fixed Autotask "script upgrade mode" 500 errors — now treated as transient and retried.
Fixed Autotask billable-cycle PSA patch-retry recursion that could spike API calls.
Fixed dispatch including cancelled/rejected Autotask PTO entries when calculating availability.

Fixed Halo dispatch using dateclosed IS_NULL instead of the open_only parameter, missing some open tickets.
Fixed Halo default-outcome behavior — now uses the first allowed outcome when none is configured.
Fixed Halo treating the "Unassigned" pseudo-agent (id=1) as a real owner.

Fixed dispatch losing the agenda "Now:" header when a tech had no appointments.
Fixed CW holiday matching and rejected past-dated schedule entries.
Added optimistic locking so parallel dispatch workflows can't double-book the same slot.

Fixed RMM components batch upsert using :name::type casts that failed on certain types — now uses CAST().
Fixed RMM device-sync heartbeats timing out at 5 minutes — bumped to 30 minutes for large fleets.
Fixed CW Asio RMM 404s on orphan companies — they're now skipped instead of failing the sync.

Fixed Exchange Online connection failures going undetected — EXO session is now verified after Connect-ExchangeOnline and retried on transient errors.
Fixed a trailing comma in allowed_cmdlets.ps1 that broke every Exchange Online run.
Improved ITGlue test_connection — 5 attempts with exponential backoff to handle brief outages.

Fixed sandbox proxy JWT expiring mid-run — now refreshed automatically and fails loudly on expiry.
Fixed sandbox stderr overwhelming the agent's input budget on noisy commands.

Fixed in-flight chained-action runs ignoring stop signals.
Fixed test-mode notes not preserving the agent-authored body in MSP-visible logs.

Fixed analytics rolling 7-day buckets — they now end at the requested to_date instead of "now".
Fixed OpenAI client not retrying on aiohttp errors and asyncio.TimeoutError — these now engage the OpenAI-direct fallback.

Renamed BI templates category to "For vCIOs" and restored ScalePad in Integrations Needed.