Release Notes

Follow the product changes that matter to product leaders, PMs, customer teams, and technical evaluators.

Knowledge intake takes a whole folder, the ideas atlas reads truthfully, and chat reasons from more of your evidence.

Product-knowledge intake can now process an entire folder of documents in one run — using each file's own outline for structure — and it stays honest and recovers when the AI service is slow or unavailable. The Discovery ideas atlas treats Size as a visual control, shows every idea (unscored ones included, in edge lanes), and large opportunity trees open again at scale. Workspace chat and connected agents can query your signals and idea votes, and the assistant answers faster and holds its state.

v0.52.0 · Released on 2026-06-15

Product Highlights

The changes most likely to affect daily product work.

  • Bring in a whole folder of product documents in one run, with sections drawn from each file's own outline
  • Knowledge intake stays honest and resilient when the AI is slow or unavailable: it says so, recovers stuck runs, and offers a retry instead of reporting your files as unreadable
  • In the ideas atlas, Size resizes nodes in place, count-based axes (votes, linked insights, ARR) show every idea, and ideas still missing impact or effort sit in labeled edge lanes instead of disappearing
  • Workspace chat and connected agents can query your signals — by topic, source, account, product, and status — and answer idea-demand questions from vote analytics
  • Opportunity trees open on large opportunities again, loading the graph from bounded counts instead of a deep relation fan-out
  • The workspace assistant is faster and steadier: curated context is pulled directly into each turn, and the header holds its state as you move

Why It Matters

Bring in product knowledge a folder at a time

Knowledge intake used to process one file per run. It now takes an entire folder, builds sections from each document's own outline, and keeps an import digest of what came in — with the integrity and performance work a real client workflow needs. It also handles AI trouble honestly: when the model service is slow or unavailable, the run says so, bounds the slow call, recovers instead of sitting in 'processing', and lets you retry rather than marking your documents unreadable. Only genuinely transient errors suggest retrying, and a stale workspace vector store re-provisions itself instead of failing every call.

  • Import a folder of documents in one run, with sections that follow each file's outline
  • An AI outage is reported as something to retry, not as bad files, and stuck runs recover on their own
  • Intake keeps working even when a workspace's vector store is stale or missing

The ideas atlas tells the truth

Size used to re-query the backend and reshuffle the map, clicking an idea refetched it twice, and count-based axes hid every idea whose count was zero. Now Size only resizes the nodes already loaded, selecting an idea doesn't refetch, and a zero count reads as zero instead of missing data. Ideas still missing impact or effort are no longer dropped behind a 'hidden from this map' note — they sit in labeled edge lanes (needs-effort, needs-impact, not-scored) where you can act on them. The atlas also loads from a single ordered query, so large workspaces map faster.

  • Size, selection, and axis changes feel instant instead of reloading the map
  • Count-based views — votes, linked insights, linked ARR — actually populate
  • Unscored ideas appear in edge lanes instead of vanishing from the map

Chat reasons from more of your evidence

Workspace chat and connected agents can now pull signals filtered by topic, source, account, product, or status, with related-account summaries for evidence questions, and answer idea-demand questions from vote analytics. Curated workspace context is injected directly into each turn rather than routed through a separate agent, so answers come back faster without losing recall.

  • Ask about raw customer signals and get answers tied to their sources and accounts
  • Ask which ideas have the most votes, or which accounts are asking
  • Faster responses, with the same curated context behind them

Full Changelog

Smaller fixes, polish, docs, and platform updates.

Discovery and product knowledge

Folder-scale knowledge intake, a truthful ideas atlas, and Discovery surfaces that hold up at scale.

  • Knowledge intake processes a whole folder in one run, with sections from each file's outline and an import digest
  • Knowledge intake resilience: separates an AI outage from unreadable files, bounds slow calls, recovers stuck runs, and offers a retry; only transient errors suggest retrying
  • Intake prompts are self-contained, and a stale workspace vector store re-provisions itself instead of failing every call
  • Ideas atlas: Size resizes loaded nodes in place; count-based axes (votes, linked insights, linked ARR) show every idea; selecting one no longer refetches; unscored ideas sit in labeled edge lanes; the atlas loads from a single ordered query
  • Opportunity trees open on large opportunities again — the visual graph loads from bounded counts instead of a deep relation fan-out
  • Idea trace flow layering stabilized, rapid idea-metric saves coalesced, and the opportunity narrative uses the standard record shell
  • Discovery section view dropdowns (Database, Trends, Themes, Canvas) refined, and a hover-focus underline regression fixed

Chat, agents, and the app

More of your evidence is reachable from chat and connected agents.

  • Workspace chat and MCP can query signals by topic, source type, account, product, status, and external id, with related-account summaries
  • Chat answers idea-vote demand questions — top-voted ideas and voting accounts — from analytics
  • Workspace context is retrieved deterministically per turn, replacing a nested context agent and cutting latency
  • Workspace header continuity stabilized, and opening an initiative routes to the workspace overview
  • Jira import flags only unsupported custom fields you might map, not system fields, with a cleaner configuration layout
  • Settings Team navigation aligned with the rest of settings, with adjacent layout and error edges tidied
  • Harvest reliability: steadier voice input, accessible icon buttons, and an error-noise filter