Skip to main content

Release Notes — v1.3.1

July 28, 2026

This release builds Agents out into a usable, discoverable feature, adds skill slash commands and in-conversation skill execution to chat, and introduces shared platform knowledge — documents, data sources, collections, skills, and agents that live above any single organization.

New Features​

Agents​

Agents were introduced in v1.3.0. This release makes them practical to use day to day.

An Agent combines instructions and a specific set of skills into a reusable AI assistant. The agent's instructions apply for the entire conversation, and only the skills assigned to that agent are available during the session. Agents can carry recommended starter prompts so users know what the agent is good at.

  • Featured agents on the chat landing page — mark an agent as Featured and it appears as a quick-pick pill directly below the composer on a new conversation. Expanding a pill shows the agent's description and recommended prompts, and selecting a prompt submits it immediately.
  • Start a chat from the Agents page — launch a conversation with an agent directly from your organization's agent list.
  • Redesigned agent form — a pinned header and footer with independently scrolling regions, a Featured toggle, and an empty state with a create CTA.
  • Manage agents — admins create, edit, feature, and archive agents from the Agents page.

Agents replace the earlier Featured Skills functionality, which has been removed.

Agents remain gated per organization. Contact your Revelry Ai team if you'd like Agents enabled for your instance.

Skill Slash Commands​

Type / in the chat composer to search and run any skill in your library. A caret-anchored, keyboard-navigable typeahead filters as you type.

  • Skills with no required inputs run immediately.
  • Skills that need input prompt you for it automatically.
  • Unmatched commands pass straight through to the AI as normal text.

Every skill now has an auto-generated slug that you can edit per organization — that slug is what you type after the /.

Run Skills Inside Your Conversation​

Skills can now run directly in your current conversation instead of opening a separate session. Both options remain available:

  • In context — the skill sees everything you've already discussed, and its work happens inline in the chat.
  • Separately — the skill runs in isolation and returns only its final result.

Pick based on how much of the conversation the skill needs.

Shared Platform Knowledge​

Documents, data sources, and collections can now exist at the platform level, accessible to every organization without duplicating uploads or configuration. Skills and agents can also be published platform-wide.

  • Platform documents and data sources support multi-namespace vector retrieval, and results are permission-checked before they reach you.
  • Platform collections group platform documents and data sources.
  • Published resources are labeled so you always know where they come from, and global agents carry an indicator.
  • Global agents get read-only views inside each organization.

For enterprise partners, this means an admin can maintain one central knowledge base that every organization can search and reference.

Drag-and-Drop File Attachment​

Drop a file anywhere in the chat area to attach it. Dropped files run through the same upload and validation flow as the attach button.

Adjustable Side-by-Side View​

When an artifact opens alongside chat, drag the divider between the two panes to resize them. Behind the inline artifact popout flag.

Message Timestamps​

User messages and assistant answers now show the time they were sent, in your local timezone.

User-Aware Responses​

The AI now knows who you are — your name and email — and uses your local date and timezone rather than UTC. Agent context in the system prompt was also expanded, so an agent's configuration is better reflected in how it responds.

Knowledge Base document queries and segment searches now return an app URL for each result, and the AI is instructed to cite it. Links are organization-scoped and safe for platform-level documents.

Web Search and Knowledge Tools Default On​

Web search now defaults to on in chat settings. A companion setting defaults the Knowledge (Data Center) tools on as well, mirroring web search behavior.

Condensed Streaming View​

A feature-flagged, calmer streaming view replaces thinking blocks and tool-call cards with a single plain-language progress indicator — per-tool display labels, an elapsed-time clock, and past-tense completion or failure chips. Contact your Revelry Ai team if you'd like it enabled.

BuildingConnected (Preview)​

Foundation for an Autodesk BuildingConnected tool family in chat, including bearer auth with transparent token refresh, structured error mapping, and connection gating. Behind a feature flag and not yet generally available.

Super Admin — Credit Management​

Super admins can now manually deposit and charge organization credits with a required note and full audit logging. Credit activity gained date-range filters, and the credit balance in the top navigation now polls so it stays current.

Organization Data Export & Import​

A new export/import framework moves an organization's data between instances. Includes a provenance concept so repeated imports stay idempotent, and portable encoding of organization settings.

UI Improvements​

  • Agents index clarity — "Global Agents" naming, and subtitles that distinguish organization agents from platform agents.
  • Skill name shown during execution — chat displays which skill is running.
  • Category picker when saving an artifact — clicking a category name always selects it, drilling into subcategories moved to a dedicated chevron, plus direct save-to-current-category cards and uniform row heights.
  • Onboarding — refreshed chat colors and styling, plus a confirmation email notification. Onboarding also seeds organization-scoped default instructions and optional web-search enrichment.
  • Copy consistency — global agent and skill wording now matches documents, data sources, and collections.
  • Removed unwired upgrade CTAs; trial credit balance now shows as a percentage.
  • Landing page card cleanup and a typo fix.

Security​

  • Fixed a potential privilege escalation for organization admins.
  • Patched dependency vulnerabilities (hackney, tesla) and bumped Phoenix LiveView to address a CVE.
  • Added daily automated security scans with results reported to the team.

Bug Fixes​

  • Chat layout is more stable: the composer and header stay pinned while scrolling, the chat pane no longer collapses on initial render, and the artifact popout can't be collapsed by dragging.
  • Disabled tools can no longer be called.
  • MCP connection refreshes no longer produce disconnection errors (Amplitude and Metabase), with serialized credential refreshes, preserved refresh tokens, and retries on transient failures.
  • Web fetch and structured web search failures fixed.
  • Platform-provided documents can now be attached in chat.
  • Deleting a document from Knowledge now cascades to dependent chat media instead of erroring on a referenced image.
  • Fixed a document attachment leak.
  • Other projects' collections no longer leak into restored chat defaults.
  • Prevented recursive skill execution in subagent chains and a subagent progress-tracker crash on non-string tool arguments.
  • MCP-created skills are attributed to the right user; API-created skills are labeled correctly.
  • Agents load correctly on a disconnected mount and no longer flash under the loading indicator when opening an existing chat.
  • Fixed mobile navigation overlay bleeding page content at the bottom.
  • "Quick Start Chat" now shows correctly when onboarding is disabled.
  • Pipedrive organization create/update accepts full field passthrough, including address and custom hash keys.
  • Textract results are fetched from the S3 output location, and documents fail cleanly when their job is discarded.
  • Handled a crash on an unexpected Accept header, and stringified complex arguments in tool calls.
  • Tool modules are loaded before their display labels are read.

Under the Hood​

  • Renamed globally_available to published on agents and skills.
  • Dropped the legacy model_requests table; removed legacy Extended Thinking fields as part of a thinking-config rework across providers.
  • Upgraded Phoenix LiveView (1.1.x → 1.2.x), Ecto 3.14.1 with writable-violation detection, req_llm, Tailwind 4.2.3, and added ColocatedJS/Hooks/CSS support with prettier-based inline formatting.
  • Seeded current default model configurations with pricing; dropped outdated seed models from fresh installs.
  • Raised Oban queue concurrency limits and increased the Firecrawl page-limit default from 50 to 100.
  • Adjusted BEAM memory settings and installed recon for troubleshooting.
  • Feature-flag gating for the chat model selector and the skill & workflow management UI.
  • Amplitude session replay behind an organization feature flag, plus expanded chat and onboarding analytics.