The Relevance MCP is now available to users with the Member role in RBAC-enabled organizations.The role gate that previously required Editor or above has been lowered to Member for configuring, deleting, and authenticating MCP servers; viewing them is unchanged. Requests run with the user’s own auth, so a Member only sees and operates on Agents they already have access to.See the MCP server documentation.
Fixes
Agent chat Fixed highlights on the Agent chat buttons.
MCP Fixed an issue where reconnecting an MCP server lost previously authorized scopes.
Triggers Resolved a race where webhook triggers could fire twice if the request landed during a deploy.
Improvements
Tools Reduced cold-start time on the first Tool execution after a deploy.
MCP Clearer error messages when an MCP request hits a permission boundary.
Extras
API Added a cursor query parameter to the Agent list endpoint for stable pagination.
CLI New --dry-run flag on the relevance deploy command.
Confluence is now available as a source for Knowledge tables.Pages are polled every 5 minutes; changes flow back into the connected table. Same sync model as the existing Google Drive, Notion, and SharePoint integrations.Connect in Knowledge → Add Source → Confluence.
Admins can now disable public cloning and embedding at the project level for every Agent, Workforce, and Tool.Toggling these settings off at the project level overrides individual asset settings and prevents users from re-enabling them on specific assets.Manage in Project Settings → Manage Access.
The Relevance AI Android app is live on the Google Play Store.Trigger Agents, view runs, and receive push notifications from your phone. Sign in with the same credentials as the web app.Get it on Google Play.
Fixes
Knowledge Fixed a case where large Notion pages were truncated mid-sync.
Workforce Resolved a UI flicker when switching between Workforce tabs.
Improvements
Permissions Faster permission checks on projects with large asset counts.
Mobile Smoother scroll performance on long Agent run histories.
Extras
API Added a source_type field to the Knowledge sync webhook payload.
Agent chat New keyboard shortcut ⌘/ to focus the message input.
Production monitoring for Agents using LLM judges. Define evaluation criteria (summary quality, hallucination detection, workflow execution), pick a sample rate, and watch scores trend over time. Drill into failed runs to see what went wrong.
Agent triggers now pause without being deleted — same controls that exist on Workforces. Toggle individual triggers, or pause every trigger on an Agent at once. Paused triggers keep their configuration.Manage from Agent → Triggers.
Chat now includes people search powered by Exa. Ask Chat to find people by role, location, or skills and it returns profiles with current roles and companies.Example: “Find me product designers in Sydney with SaaS experience.”
The Analytics dashboard now plots your org’s concurrent operations as a timeseries over the last 7 days, with your plan limit drawn on the chart so you can see how close you’ve been running to the ceiling.
The Relevance AI app is live in the Microsoft Teams App Store. Trigger Agents from Teams channels and chats, and receive run results back in the same thread.Install from the Teams App Store and connect your workspace.
Project invitation emails now show delivery, bounce, and open status — and let you resend invites that bounced or got missed. No more guessing whether a teammate’s invite landed.See it in Project Settings → Pending Invites.
The Analytics dashboard now breaks credit usage down by individual user, with a per-Agent breakdown showing which Agents each user runs most.Find it at Analytics Dashboard → Users.
Agents can now be triggered by Microsoft Teams calendar events — meeting start, end, etc. Parity with the existing Google Calendar trigger, so Microsoft 365 teams can wire up the same meeting-prep, follow-up, and summary workflows.Add it from Agent → Triggers → Teams Calendar.
.pptx files can be attached in Chat. Agents read the text, images, and slide layouts.For decks with 100+ slides the file may exceed model limits — Gemini and OpenAI models handle the largest decks best.
Anthropic’s flagship Opus 4.6 is now available in the model picker.
Context window: 1M tokens (both the base entry and the “1M context” entry — the second one exists for pricing reasons, with separate tiers for ≤200K and >200K input)
You can now edit a bulk run after it’s been created — change daily task volume, change time zone, change other parameters in the bulk run modal — instead of cancelling and starting over.Open your Agent, select the active bulk run, and click Edit.
Structured testing for Agents. Create test suites with multiple scenarios, generate user inputs, and evaluate runs with LLM-based judges against criteria you define. Track scores across test runs so regressions show up.Tool-call simulation (run an Agent through a flow without firing real Tool side effects) is coming next.Find it under your Agent’s Monitor → Test Suites.
Enterprise only — talk to your Relevance AI rep for access.
If you’re running Anthropic models through your own AWS Bedrock account (BYO), Relevance now forwards a guardrail identifier and version on every Bedrock invoke. The actual filters — prompt-injection blocking, PII masking, custom topic restrictions, profanity controls — are configured AWS-side in your Bedrock console; Relevance passes the ARN through.Set the guardrail ID/version in your project’s Anthropic integration settings.Setup guide →
Audit logs and execution traces can now be streamed to a customer-controlled destination in OpenTelemetry format (OTel spans + logs). Plug it into your existing observability stack — Datadog, Honeycomb, Grafana, whatever — and run governance and compliance queries alongside the rest of your telemetry.
Enterprise with RBAC enabled. RBAC is rolling out progressively — talk to your Relevance AI rep if you don’t see it.
Workforce triggers now support pause, cancel, and resume — both per-trigger and at the whole-Workforce level. Pause an entire Workforce during maintenance, then resume it without recreating anything.Manage from the Workforce dashboard, or use the Pause all triggers button in the top menu.
Approve or reject pending Tool calls from your Agents directly in Slack, without switching contexts. The notification shows the Tool name and parameters; one click sends the approval back.Enable per Slack trigger: Triggers → ⋯ → Edit trigger → Advanced settings → Tool approval from Slack → Yes.
The Plan & Billing page now supports filtering credit usage by date range and by project, with side-by-side comparison and export. Useful for chargeback reporting and spotting which project is driving spend.Profile → Manage organization → Plan and billing.
Validate a Phone Agent’s prompt, voice, and conversation flow from the dashboard without placing a real call. Microphone permission required.On the Phone Agent’s build page, hit Test in Browser under Call Agent.
Type a trigger word in the meeting chat and your Meeting Agent acts on it mid-meeting — run a search, pause or resume the recording, build slides, post to Notion or Slack — without breaking the conversation.Build a new Agent, add the Relevance AI Meeting Bot trigger, attach the Tools you want (e.g. Pause Recording, Resume Recording, Google Search, Slide Builder) and define each Tool’s trigger word.
