This month we launched Agent OS in beta, bringing a more capable Enterpret Agent and proactive Agent Automations to customers, made Enterpret available inside ChatGPT, shipped four major feedback integrations plus new Agent connectors, introduced live sales deal-risk intelligence, and webhook delivery for Escalation Shield Agent, and made broad improvements across speed, accuracy, ingestion, exports, and reliability. Here's what's new.

Agent OS: Proactive Customer Intelligence, Powered by Enterpret Agent

We launched Agent OS in beta: a new way to investigate customer feedback, work with connected business context, create shareable outputs, and trigger action across the tools your teams already use.

• Enterpret Agent — Wisdom, supercharged: Ask complex, multi-step questions and have the Agent investigate across customer feedback and connected systems, then turn the result into cited, shareable work.
• Reusable skills: Package repeatable instructions and best practices so the Agent can consistently execute workflows for your team.
• Connected context and actions: Native connectors include Salesforce, Jira, Linear, GitHub, Notion, Canva, Granola, Amplitude, PostHog, Slack, and now Zendesk MCP — so the Agent can bring customer evidence together with the systems where work happens.
• Shareable artifacts: Create files, dashboards, documents, and other polished outputs directly from an Agent session.
• A smoother Agent experience: We added inline session naming, clearer approval controls, improved readability for dense answers, better file previews, and a resizable workspace for Agent-assisted taxonomy work.
• Usage guardrails: Per-organization spend caps and internal monitoring provide a safer foundation for scaling Agent usage during beta.

Agent Automations: Customer Intelligence That Keeps Running

Agent Automations put Enterpret Agent to work in the background, on a schedule or in response to a signal, so teams can move from one-off analysis to repeatable operating workflows.

• Scheduled workflows: Run recurring analyses such as weekly Voice of Customer reports, launch monitors, support scorecards, help-center gap reviews, account-health summaries, and customer-story discovery.
• Signal-driven workflows: Trigger work when Enterpret detects a feedback spike, sentiment shift, urgent escalation, quality issue, or emerging account risk.
• Connected automation runs: Automations can use the same connector catalog and reusable skills as on-demand Agent sessions, including connected tools such as Linear, Notion, PostHog, and Slack.
• Fresh, audience-ready outputs: Each run produces a new synthesis of what changed, what is emerging, what resolved, and why it matters — delivered in the format and destination that fit the audience.

Enterpret is Now Available in ChatGPT

Enterpret is live in the ChatGPT App Store, bringing cited customer intelligence directly into ChatGPT and Codex workflows.

• Ask questions about customers without leaving ChatGPT.
• Get answers grounded in your Adaptive Taxonomy and feedback records, with citations back to Enterpret.
• Combine customer context with the documents and workflows already being used in ChatGPT.

New Feedback Integrations: Pylon, Decagon, Fathom, and HubSpot

Four new integrations went live in June, expanding the customer conversations, calls, support interactions, and CRM activity teams can analyze in Enterpret.

• Pylon: Ingest issues, conversations, tickets, call recordings, and survey responses as feedback, while syncing contacts and accounts for user- and account-level analysis.
• Decagon: Automatically sync AI-agent and support conversations into Enterpret for search, Reasons, Pulse, dashboards, and Agent analysis.
• Fathom: Pull meeting recordings and speaker-level transcripts on a recurring schedule, with speaker attribution that separates internal teams from customers.
• HubSpot Feedback: Ingest tickets, deals, leads, feedback submissions, notes, emails, calls, meetings, activities, and custom objects, with flexible field mapping and associated-object context.

Live Deal Intelligence: Find Risk Before a Deal Closes

An early access release of Live Deal Intelligence extends Sales Intelligence beyond explaining completed deals to identifying material risks while an opportunity is still open.

• Evidence-grounded deal risks: Risks are derived from source feedback and can include a category, severity, explanation, supporting evidence, and mitigation.
• Actionable by default: The initial experience prioritizes medium- and high-severity risks to reduce noise.
• Available across Sales Intelligence: Stored risk objects can be used in Deal Detail, Quantify, and Enterpret Agent workflows.
• Built for proactive selling: Teams can move from “Why did we win or lose?” to “What should we pay attention to right now?”

Escalation Shield Webhooks

Escalation Shield can now deliver support-escalation signals to a webhook, alongside Slack and email. This makes escalations programmatically consumable — for example, to prioritize work in Zendesk or Intercom, or to start an agentic workflow when an urgent issue is detected. Learn more.

MCP Reliability and Lifecycle Improvements

• Activity-based token expiry: MCP tokens now stay active when used at least once every 90 days, with a one-year absolute cap, instead of expiring after a fixed 180 days regardless of activity.
• Proactive expiry warnings: Token owners receive email notifications 14 days and 7 days before expiry, plus an expiry notification with a direct path to generate a replacement.
• Faster customer quotes: Quote generation across Enterpret MCP, Wisdom, and Enterpret Agent is now approximately 80% faster.
• Claude and Cursor connectivity: Fixed an MCP v2 regression that could block connections and authentication in Claude and Cursor.

Quality & Reliability Improvements

A broad set of improvements made Enterpret faster, more accurate, and more dependable across agents, analytics, ingestion, and exports.

• Escalation Agent accuracy and stability: Escalation intent and emotional intensity now use discrete, model-stable levels instead of drifting 0–100 scores. Alerting runs on a calibrated decision matrix, with upgraded models selected through production benchmarks.
• Sharper feedback summaries: Improved grounding, actor attribution, context retention, and hallucination resistance in the summary pipeline, reducing incorrect roles, unsupported details, and prescriptive next steps.
• Faster time to insight: For a high-volume Zendesk workflow, measured p90 end-to-end latency fell from roughly 105 minutes to 26 minutes by the end of June.

Bug Fixes & Product Polish

• Quantify: Restored the user-sentiment breakdown and fixed text-search filters that could reset or reload incorrectly.
• Enterpret Agent: Fixed file-preview regressions, session names that did not persist on blur, dense markdown readability, attachment-review cleanup, and Calendly authorization failures.
• Onboarding: Fixed OAuth redirect and post-connect drawer behavior, improved readiness-state handling, and preserved navigation state across the onboarding flow.
• Integration metadata: Cleaned up noisy internal metadata labels that surfaced instead of user-friendly field names for some webhook integrations.
• Feed dates: Removed historical cross-partition duplicates that could surface incorrect 1970 dates in filtered call feeds.
• Slack links: Fixed malformed dashboard URLs that could cause Slack preview rendering to time out.
• Data exports: Fixed timeout and empty-response cases in the Export API and repaired missing compacted data.
• Pipeline resilience: Hardened Sales Intelligence and Product Feedback pipelines against rate limits, dependency timeouts, and replay/backfill failures.

Thank you for pushing us forward. Lots more coming in July — stay tuned!

This month we shipped a major upgrade to the Enterpret MCP server — now with citations on every answer and support for any MCP host — launched Claude Opus 4.8 in Wisdom, brought in-product transparency to every integration, made taxonomy editing faster with new AI-assisted chat, validation, and node-level edits, upgraded our Escalation Agent, and landed a broad set of reliability and accuracy fixes. Here's what's new.

Enterpret MCP Server v2: Cited, Simpler, and Works Everywhere

A major upgrade to the Enterpret MCP server, which lets AI hosts query your Customer Context Graph, was rolled out to all customers. Read the announcement and set it up.

• Citations on every answer: Every aggregate number returned by Enterpret MCP now carries a deep link back to the source records in the Enterpret dashboard, just like the in-product chat experience. No other MCP server on the market does this.
• Works on any MCP host: The server is now tools-only, so it runs on any MCP-compliant host, including ChatGPT, Codex, n8n, and LibreChat, alongside Claude, Cursor, Glean, and Notion.
• Simpler setup: The initialize_wisdom prompt and usage-guidelines resources are gone. The server self-initializes from cleaner tool descriptions, nothing to run before you start asking questions.
• More context-efficient: Schema and instructions now load on demand instead of being dumped upfront. Sessions last longer before compaction, and customers with large schemas get materially faster, cheaper answers.
• Verbatim customer quotes: A new quote tool returns validated, word-for-word customer quotes with speaker attribution: actual words, not paraphrases.
• Renamed: Enterpret Wisdom MCP is now Enterpret MCP.

Existing connections keep working. To pick up the new experience, reconnect and re-authenticate the server in a fresh session.

Claude Opus 4.8 is now available in Wisdom

• Anthropic's latest Opus model is live. Opus 4.8 replaces 4.6, Opus 4.7 remains available, and Sonnet 4.6 is still the default.

Integration Transparency

Enterpret customers now have in-product visibility into the health of every integration.

• Nine user-facing statuses on every integration row: Active, Ingesting, Setup incomplete, Connection expired, Slow ingestion (Recovering), Ingestion blocked, Issue detected, Paused, and Deleted — each with the right color, label, and tooltip.
• Proactive email alerts to org admins: A first alert on Day 0, reminders at Days 3, 6, 9, and 12, a final notice at Day 15, and a recovery email once the issue clears, so a broken integration no longer goes unnoticed for days.
• Upgraded detail pages: A new ingestion activity chart with 30-day, 60-day, 90-day, and all-time ranges, UTC timestamps, and all-time totals.

Adaptive Taxonomy: AI-Assisted Editing, Validation, and Chat

Refining your taxonomy is now faster, safer, and easier with four new capabilities — available today and included by default with every Enterpret subscription, from Edit Mode on the Taxonomy page in Settings. Read the announcement.

• Chat with your taxonomy: Describe what you want in plain natural language — "Merge all login-related sub-themes under Authentication," "Move Notifications from Desktop to Mobile App," or "Read this feature doc and tell me what edits to make." Enterpret reads the taxonomy, interprets your intent, proposes the operation, and lets you review the change set before anything is applied.
• Quality checks before changes go live: Every proposed edit is checked for near-duplicates, naming conflicts, semantic overlap, and structural issues. When a change can't be applied directly, Enterpret suggests an alternative path — for example, reparent first, then merge — so you can review the full sequence before accepting.
• Edit any node, with guardrails: Rename, merge, split, move, create, delete, and reparent nodes — down to the themes and sub-themes closest to customer language — directly in the UI. Changes accumulate in a draft, nothing touches production until you save, any change can be rolled back, and Enterpret automatically reclassifies affected records on commit.
• Faster Taxonomy Edits & Edit During Backfill: Backfills now run faster, and you can keep editing the taxonomy while they run in the background.

Quality & Reliability Improvements

A broad set of accuracy and reliability upgrades across our agents and platform — sharper detection, cleaner alerts, and fixes throughout.

• Escalation Agent: Quality Upgrade: Two improvements rolled out to all customers — more meaningful alerts, with less noise out of the box.
  • Smarter detection: We upgraded the AI models behind the Escalation Agent, so it surfaces the issues that matter with greater accuracy.
  • Fewer low-value alerts by default: Billing disputes, refunds and chargebacks, and account bans are now automatically excluded for every customer — addressing the common "too many billing alerts" problem from day one. You can turn these back on or fine-tune what's excluded for your team at any time.
• Quality Monitor Agent: We upgraded the AI models behind the Quality Monitor Agent and streamlined how feedback is analyzed, so issues are detected more sharply and alerts are cleaner.
• Accurate feedback counts in Wisdom: Fixed a deduplication bug where source-citation drill downs counted duplicate feedback records, so source counts and opened records now match.
• More reliable user quotes: The quote flow now looks for speaker names across many more metadata fields and linked user records, instead of a small fixed set — so real customer names resolve correctly rather than falling back to "Customer."
• API error-code propagation: Improved handler-level error wrapping across core services so APIs return the intended error codes while preserving existing behavior.
• Ingestion & pipeline hardening: Fixed content-field override when feedback is re-ingested after translation, added SQL validation in the backfill service (rejecting conflicting ORDER BY / LIMIT / OFFSET), and resolved a query performance anomaly.
• App Store / Play Store enrichment: Hardened the app-store enrichment pipeline and backfilled historical records, plus added partial country-level metadata across more connectors — unblocking country-level filtering on app-store reviews.

Thank you for pushing us forward. Even more coming in June — stay tuned!

We shipped Enterpret AI assistant on Slack, launched Sales Intelligence, upgraded Wisdom to run on Claude Opus 4.7, new YouTube Comments and Trustpilot and a revamped Slack integrations, and made deep quality investments across our agents, audio pipeline, and taxonomy. Here's what's new.

Enterpret Everywhere: Bringing Customer Intelligence to Slack, Claude, Notion, Glean, and Beyond

• Enterpret on Slack: @mention @Enterpret(Wisdom) in any Slack channel or DM to ask anything about your customers and get cited, conversational answers without leaving Slack. Learn more.
• Available on the Slack Marketplace: Enterpret Slack app is now listed in the Slack Marketplace for one-click install in any workspace.
• New native Enterpret MCP integrations: Enterpret MCP is now natively live in:
  • Claude (via Anthropic's MCP Gallery) — available for all Claude users
  • Notion Custom Agent — Enable Enterpret MCP as custom agent. Learn more.
  • Glean — available through the Glean admin console. Learn more.
  • Cursor — available self-serve through setup. Learn more.
  • Coming soon in their marketplaces: Cursor, ChatGPT, LibreChat, n8n

‍
‍

‍

Sales Intelligence

• Sales Intelligence is here — understand every deal to win more. Sales Intelligence automatically analyzes every closed deal across your CRM (Salesforce, HubSpot) and call transcripts (Gong), and surfaces the real reasons you win and lose, with revenue impact. Powered by Enterpret's Adaptive Taxonomy, it replaces six-week interview projects and unreliable CRM picklists with customer truth — ranked by ARR.Built for the whole GTM team:
  • Product (which gaps are blocking revenue)
  • RevOps (attribution beyond picklists)
  • Sales leadership (the real story behind losses)
  • and Product Marketing (battlecards from what reps actually hear).
  Available as a sales-led add-on for B2B customers. Learn more
• Four ways to explore deal insights:
  • Exploration — Browse every opportunity by team, quarter, use case, competitor, segment, or ARR, with win/loss reasons, product gaps, value props, and competitor presence at a glance — breakdowns Salesforce reports were never built for.
  • Deep dives — Full AI-generated reports for any opportunity covering win/loss reasons, competitors, use cases, product gaps, and prospect language, with every claim linked back to its source.
  • Aggregate analysis — QBR- and board-ready views of revenue impact across win/loss reasons, product gaps blocking deals, and value props driving wins.
  • Ask anything across your deals, in Slack and Claude — Ask anything across your deals in real time ("Why are we losing to Competitor X this quarter?", "Which use cases convert in mid-market but stall in enterprise?", "What feature gaps are linked to the most losses?") with cited evidence, anywhere via MCP.

‍

Wisdom can now run on Claude Opus 4.7

Anthropic's latest Opus model is live in Wisdom. Expect sharper reasoning on complex questions and stronger handling of multi-step analysis.

New & Improved Integrations

• YouTube Comments: Bring comments and replies from your owned YouTube channel into Enterpret with one-click OAuth — 90-day backfill on day one, continuous sync after, video metadata for context, Shorts and multi-channel support, and built-in spam filtering. Now unified with your tickets, calls, and surveys in Wisdom. Set up YouTube Comments →
• Slack, rebuilt and on the Slack App Directory: Event-driven ingestion delivers messages in seconds, with complete threads, preserved forwards, and live edits/deletions. Noise (join/leave, bots, prompts) is filtered out, real names and @mentions resolve everywhere, and new metadata (Slack Connect, email domains, workspace names) powers customer/account identification. New channels appear in 1–2 minutes, and Enterpret is now one-click installable from the Slack Marketplace.
• Trustpilot: Native Trustpilot connector for pulling public reviews directly into Enterpret.
• HubSpot Users & Accounts: Bring HubSpot contacts and companies into Enterpret to enrich feedback records with CRM identity and account context.
• Zendesk backfill improvements: Smarter historical backfill for faster, more complete catch-ups when connecting Zendesk Support.
• Snowflake setup transparency: New "Validate Connection" step on the integration page during onboarding — preview sample columns and rows, get clear error messages on bad configs, and edit account, database, schema, warehouse, or table inline without re-creating the integration.

Quality Improvements

• Quality Monitor Agent: Detection accuracy improvements thanks to a new model family, rewritten prompts, sharper titles and root causes, and smarter grouping.
• Audio & Call Intelligence: New insight extraction architecture for audio chapters improved call summary quality. Added end-to-end translation support for non-English Gong and Zoom calls.
• Inferencing & Taxonomy: Taxonomy and prediction improvements improved precision and moved a meaningful share of records out of MISC into specific, actionable themes.
• More of your metadata, automatically available in Wisdom: Wisdom now picks up metadata fields that were previously skipped because only a small fraction of your records had them populated — ticket categories, assignee names, SLA fields, plan tiers, and custom Gong call attributes. These are now queryable, filterable, and citable in answers, without any setup on your side.
• Bug fixes: Resolved an error users hit when opening a Wisdom-cited record from the dashboard, with clearer error messaging in the meantime.
• Performance & reliability: More resilient ingestion for Twitter and Play Store integrations with retry/backoff and new metrics.

Thank you for pushing us forward. Lots more coming in May — stay tuned!

‍

In March 2026, we focused on expanding our admin tooling, improving reliability across integrations, and tightening up platform permissions and Slack experiences.

Customer Insights in Claude

• Enterpret Customer Insights Claude plugin (Cowork + Claude Code): Install the plugin to bring structured customer insights workflows into Claude, including feedback analysis, root cause investigation, and branded report generation — powered by the Wisdom MCP Server. Read more

Admin Controls

• Per-domain auto-approve controls: Admins can now configure auto-approve independently for each approved domain (Workspace Settings → Approved Domains).
  • Previously, auto-approve was all-or-nothing across domains.
  • Now, each domain can be configured separately. For example:
    • example.com → auto-approve ON (users can sign up and get access immediately)
    • acme.com → auto-approve OFF (users must be invited to get access)

MCP Server Compatibility

• Improved MCP authorization compliance: The Wisdom MCP server's OAuth and Dynamic Client Registration (DCR) now conform to the MCP specification, allowing any MCP host that supports tools, resources, and prompts to self-register and authenticate without server-side changes. Customers can now connect Wisdom from hosts like Codex, Windsurf, Notion, and Glean alongside Claude and Cursor.

Reliability & Quality Improvements

• Quality Monitor Agent improvements: Alerts are now cleaner and more actionable. Fewer unrelated issues get merged, and Root Cause Analysis (RCA) are grounded in what users actually reported.
  • Smarter alert grouping: Clique-based grouping reduces over-clustering and prevents borderline judgments from snowballing into oversized alert groups.
  • More reliable RCAs: Root causes are derived from underlying user reports (not anchored to potentially misleading alert titles).
  • Clearer problem statements: The agent describes the problem instead of restating the solution, improving specificity.
  • Better titles: Both RCA specificity and alert titles improved materially across both orgs.
• Members can edit their own agents: Resolved an issue where users with the Member role were getting logged out when attempting to edit agents that they created

Thank you for your feedback and support as we continue to make Enterpret more powerful and delightful. Stay tuned for more updates in April!

In February 2026, we focused on improving product reliability and operational quality.

Wisdom

• Opus 4.6 Support: Wisdom now supports Claude's most advanced reasoning model, Opus 4.6, for deeper and more nuanced analysis on complex queries.
• Sonnet 4.6 + Model Switcher: Claude Sonnet 4.6 is now the default model on Wisdom, delivering faster responses and higher tool success rates on single-hop queries. A new model switcher lets users toggle between Sonnet and Opus directly from the Wisdom interface.
• Unified Fields in Wisdom: Wisdom can now query and use Unified Fields, enabling richer analysis for customers with complex metadata configurations.
• Context Graph Search Improvements: Overhauled how Wisdom searches the Knowledge Graph — each concept is now searched separately in parallel with fallback workflows, eliminating 70% of search failures and doubling eval pass rates from 50% to 100%.
• User Quote Attribution Improvements: Improved how Wisdom extracts and attributes quotes from Gong calls, reducing cases where internal employee statements were incorrectly attributed as customer feedback.
• Wisdom Rules on Wisdom Page: Users can now discover and configure Wisdom Rules directly from the Wisdom interface, with an embedded link to curated example rules. Learn more.
• Export Wisdom Citations: Users can now export Wisdom citations — both feedback and non-feedback (accounts, users, work items) — directly to email, bringing the same export workflow from the classic product into Wisdom.

New Integrations

• Linear Integration: Bring customer insights into the Product/Eng workflow by linking Enterpret feedback (records, themes, feeds, collections) to Linear issues/projects, and auto-enriching tickets with the relevant context so teams can build with the why at their fingertips. Learn more.

Bug Fixes & Quality Improvements

• Slack notifications no longer show empty “No feedback available” summaries: Fixed the Slack notification flow where long summaries could result in empty Slack content; improved gating so empty messages aren’t sent.
• Exports no longer fail due to empty records (Monday tenant): Fixed an export failure mode caused by effectively-empty feedback records.
• Gong calls open reliably from Feed again: Resolved an issue where Gong calls failed to open due to data inconsistencies.

Performance Improvements

• Improved performance on complex dashboards: Major improvements to unified fields, custom filter application, and query patterns reduced timeouts and improved load times.
  ‍

Thank you for your feedback and support as we continue to make Enterpret more powerful and delightful. Stay tuned for more updates in March!

In January 2026, we focused on platform reliability, performance improvements, and expanding our integration ecosystem.

Taxonomy Editing & Management

• Edit Predictions on Individual Records: Users can now edit predictions at the individual record level, enabling more granular and accurate taxonomy predictions.
• Edit Mode for Taxonomy Page: We've introduced a dedicated edit mode for the taxonomy page. Miscellaneous and General keywords are now only visible in edit mode, streamlining the standard view for most users.
• Instant Feedback for Taxonomy Edits: Keywords now appear immediately after edits, rather than disappearing during processing. This eliminates confusion and improves trust in the editing process.

Wisdom AI

• Reliable User Quotes: A new post-processing step improves quote accuracy and attribution. Quotes now show exact wording with clear speaker attribution.
• Citations GA: Wisdom citations feature is now generally available, providing transparent source attribution for AI-generated insights.
• MCP Server Authentication: Easier setup across Claude, Cursor, and other tools withOAuth support and unified authentication flow. Learn more.
• Wisdom Rules: Users can now create rules that control how Wisdom behaves across all queries. Workspace Rules (only admins can create) apply organization-wide, while User Rules allow personal customization. Rules can specify behavior preferences, require clarifications, enforce formatting standards, and define company-specific terminology. Learn more.

Performance & Reliability

• Dashboard Performance: Implemented month-based table partitioning, reducing query times by 50% and memory usage by 66%. Dashboards that previously timed out now load faster and more reliable.
• Unified Fields Materialization: Optimized how unified fields are queried, improving dashboard load times on complex dashboards.
• Quantify Query Optimization: Improved Quantify load times by removing redundant count queries.

Platform Stability

• Redis Infrastructure: Isolated high-load clients to dedicated servers and added fail-open guardrails to prevent cache-related outages.
• Enrichment Optimization: Significant performance improvements to the enrichment pipeline, reducing processing lag.

Bug Fixes

• Fixed: "No Value" display bug in Quantify where empty fields showed incorrect labels
• Fixed: Double count query in Quantify causing redundant database calls
• Fixed: Feed page intermittent load failures
• Fixed: Wisdom functionality within Feeds view
• Fixed: Broken links in Wisdom responses
• Fixed: Co-occurring themes from Gong not reflecting in Quantify
• Fixed: Link generation issues during high sync lag periods

Thank you for your feedback and support as we continue to make Enterpret more powerful and delightful. Stay tuned for more updates in February!

Questions or feedback? Reach out to our team — we'd love to hear from you!