Changelog

Added

• Close and reopen controls for sessions, in the app and through the MCP. Closing stops new participant activity and keeps all threads, messages and summaries. Reopening lets participants continue.

Fixed

• A confirmed short session can now be closed.
• Session collaborators can now see participant-only results.

Changed

• Improved Delphi’s second and third rounds, with a tighter opening and stricter limits on what each round produces.
• A project’s sources list is now paginated.

Added

• Chain roster: shows each participant’s status for every step (not ready, notified, claimed, started, completed, no-show) and a completion count per step. Hosts can copy or resend a participant’s step link from the roster, and resends are recorded.

Changed

• Harmonica now picks context sources based on the participant’s latest message, not only the host brief. The first turn still uses the brief, and Harmonica falls back to the brief if the turn-based lookup fails.
• Improved Cynefin chains. Chain templates now declare which project type they create.
• When a session uses several context sources, each source’s share of the retrieved context is capped so one large source can’t crowd out the rest.

Fixed

• Confirming a recap directly now counts as confirmation.

Fixed

• Seven fixes to how Harmonica runs conversations. It offers an inference for the participant to confirm rather than asserting it, reframes by the participant’s role, keeps replies aligned to what was asked, and no longer accepts a close without real confirmation. Delphi’s first round is now host-driven, so it covers the deliverable, stays concrete, and asks neutrally.
• Question marks are no longer replaced with periods, and questions stay in the participant-facing question field instead of appearing in the body of a turn.
• A chain’s step-one prompt is now kept when a session is created through the API or MCP.
• Role-ready emails now record when they were sent, so the roster shows the notification.

Changed

• Improved chain configuration, handoffs between steps, and chain start-up.
• A session’s distribution channel is validated against the reserved list of channels. A channel with no dispatcher is rejected with a 422 instead of being stored and ignored.

Added

• Privacy inventory for a project record, showing for each item its private wording, the shared wording that would be published, and which surfaces it appears on.
• POST /sessions can start a chain, and the response says whether the chain started. (API)
• widget_spec is now passed through participant chat and returned on ChatResponse, so a client can render Harmonica’s polls and ratings. (API)

Added

• Link previews for public project records: a shared record shows the project’s title and intro instead of a generic card. Private records are not indexed.

Fixed

• Endorsed preview responses are now included in the API response.
• When Harmonica asks a question, the turn now stays open for the answer instead of closing the round.

Added

• Hosted MCP server: connect Claude, Cursor or any MCP client to Harmonica with a bearer token, with nothing to install. Read tools list and search sessions, participants, responses, questions and summaries. Write tools create and update sessions, close and reopen them, generate summaries, post chat messages and submit questions. The npm package is still available and supports the same tools.

Changed

• The chat’s session list moved into a drawer, giving the embedded chat more room.
• A project record’s Memory tab is now for curating what the record says, rather than adding to the conversation.

Fixed

• Fixed the Emerging group labelling its own members as uncorroborated.
• MCP calls that are refused or reference a missing id now return an error message.

Added

• GitHub repositories as a context source for Team OS projects. Connect a repository from the project’s Memory record and its markdown is used as context in every session in the project. It reads all *.md files by default, can be narrowed to selected paths, and shows which files were read and which were skipped against the size limit.

Fixed

• Fixed MCP session search ignoring the search term and returning the first 20 sessions.

Added

• People roster in Settings, Context: a list of the people who take part in your sessions, with role, expertise and contact details. Harmonica receives only an aggregate, for example “12 people, 3 Backend Engineers, expertise in onboarding and payments”. Names, emails and notes are not sent to the model.
• Email domain access: open a project’s record to anyone with an email address at a domain you choose. Free email domains are not accepted.
• Methodology labels on projects: launching a Wardley chain labels the project “Wardley Mapping project”. The label is set by the first chain and does not change.

Changed

• Creating a session inside a project now goes through the template picker instead of skipping method selection. “Start from scratch instead” is available for a blank session.
• Pre-survey answers now reach Harmonica as a participant profile rather than as an opening message in the participant’s name.
• When a participant says they have nothing to add, Harmonica ends the conversation instead of asking another question about the remaining agenda.
• Session briefs no longer repeat the host’s own words back, and open with exactly one question.

Fixed

• The rating request now shows on mobile. Its buttons also have screen reader labels and full-size touch targets.
• A private record now shows as private instead of looking like a broken link.

Added

• Organization access: bind a project to an organization so its members can read the results without having taken part. The organization is picked from a list.
• Contesting record entries: anyone who can read a published record can start a conversation about an entry, and the entry’s state updates as a result. Contesting a settled entry reopens it.
• Lately strip on the record page, showing what changed recently.

Changed

• The contest action is prominent on entries still being decided, less prominent on long-settled ones, and hidden on retired ones.
• A reader without access to an organization-gated record now sees an explanation instead of a page that looks broken.
• A conversation about a record entry no longer opens by restating what the reader just said.

Added

• Project types: Standard (sessions and reports, private by default), Broad Listening (a public conversation with an opinion map), and Team OS (a record of what a team has settled and what is still open). The type is chosen once, at creation, shows in the project header, and determines which publishing options are available.
• Memory record for Team OS projects. It persists across sessions instead of being regenerated each time. Entries are grouped into decisions, tensions, questions and themes, each with a state: emerging, contested, settled, reopened or retired. When synthesis and the host disagree, the host’s choice is kept.
• Record publishing: a Team OS project can publish its record at its own URL, with an intro written by the host. All projects now have a visibility setting: public, members, or dashboard only.

Fixed

• Fixed the publish link returning a 404. The publish check now counts only what will appear on the page.

Added

• Invite progress on a host’s first session: the Invite Participants card shows progress toward three people who have responded, counting responses rather than link clicks.

Fixed

• Fixed chat messages occasionally failing to save.
• The Refresh button on a session report now stays usable while the report is stale.
• The participant count on a session’s scratchpad now matches the server.

Added

• Statement management on the Opinion Map tab: a drawer lists every statement by source, with its vote tallies. You can exclude a statement from the public page without losing its votes and restore it later. An excluded statement is not regenerated.
• Meeting transcripts as a source type. Harmonica detects and labels uploaded transcripts and reads long ones into the project map and seed statements. You can also mark or unmark a source as a transcript yourself.
• What’s New popup on the notification bell, announcing each release with links to the relevant help guides.

Changed

• Statement generators now produce specific, decision-relevant statements people can disagree with, instead of generalities.
• Anchored deliberation on a cluster or statement now opens on the current page instead of navigating away from the map.

Fixed

• Quotes from hard-wrapped source text, such as forum threads and pasted articles, are now grounded correctly. Most of them were being dropped.

Added

• Per-question depth: for each core question, a session shows whether it has been worked through, touched but not explored in depth, or not yet explored. Shown alongside coverage and saturation, for information only.
• Consensus provenance: each emerging-consensus item shows how it was reached (Instant, Worked through, or Validated in reflection), with a one-line timeline.
• Deliberations section on the project’s Opinion Map tab. It groups anchored conversations by what people reacted to, with a short summary and the full transcript for each thread.

Changed

• The coverage signal no longer flags a session that is still in progress.

Fixed

• Removed a stray session card that appeared on the dashboard for anchored deliberations and broke when opened.
• A conversation link that is no longer available now says so, instead of showing a generic error.

Fixed

• Session coverage now counts only the questions a conversation actually covered.
• Fixed overlapping opinion map labels. Very thin regions of the map now render instead of disappearing.
• Follow-up questions in anchored sessions are more reliable.

Added

• Completeness and saturation: a session shows when it has covered its questions and when new responses stop adding new information.
• Reflection rounds: once a session settles, you can invite participants back to react to the group’s map. The group then either confirms the consensus or raises new points.
• Inherited project memory: each new session in a project starts from what earlier sessions established. Hosts can turn this off.
• Anchored sessions: start a session based on where the whole project has landed, so people respond to it directly.
• Public anchored deliberation: on a public topic page, visitors can start a conversation with Harmonica about a specific cluster or statement.
• Participant pre-surveys: templates can ask a few structured questions before the conversation starts, on single sessions and on chains.
• Seed statements from sources: turn an uploaded document into statements people can vote on, review them, and show the source of each statement on the public page.
• Project management API: create and manage projects, and set a session’s project, through the REST API. There is no dashboard UI for this yet.

Changed

• Onboarding: a guided welcome flow for first-time hosts, and a reminder on the dashboard if they leave before creating their first session.

Fixed

• Editing a pre-survey question now keeps its answer options.
• Pro plan overage limits are enforced correctly in the credit path.
• Linking and unlinking sessions on a project now checks your permissions.
• Public opinion maps: endorsements from preview responses now count, and duplicate statements are detected more accurately.

Added

• Participant pre-surveys defined by templates: templates can define a short pre-survey participants complete before the conversation (name, email, role and method-specific fields), including on the first step of a chain.
• Redesigned onboarding: a four-step welcome flow for first-time hosts, and a reminder on the dashboard if they leave before creating their first session.
• Project archiving: a redesigned projects page and an Archive action.

Changed

• The project Knowledge tab now uses a fixed layout with findings first (Decisions, Tensions, Actions, Open questions, Themes) that stays the same when knowledge is regenerated, with a colour legend.
• Included preview runs now appear on the Responses tab and count toward the Report tab, with a redesigned in-chat preview banner.
• Session Design now opens with the Facilitation Approach and shows the Session Report editor.
• Editing a session’s brief regenerates its recap and prompt to match.

Fixed

• Empty or whitespace-only chat messages are blocked.
• The text-mode fallback no longer shows raw turn JSON in the conversation.
• Deleted-project metadata no longer leaks on the preview page.
• Auto-created chain steps inherit the session’s pre-survey.
• Three small session page fixes.

Fixed

• Participant transcript emails: when a participant opts in, their transcript is now sent by email. The email’s question-and-answer layout was also redesigned.
• The “view results” links in participant and milestone emails now open on a different device from the one used to chat, and the buttons have clearer labels.
• Timer-based chain advancement and milestone emails work again. A scheduled-job authentication change had stopped both.
• Fixed a chain step being able to pull in another step’s context.
• Lifetime-deal accounts on hosted models now see the full platform model list.

Added

• Visual artifacts for more methods: Cynefin and SWOT render as an interactive quadrant canvas, and Six Thinking Hats as a mind map per hat.
• Drag components on the canvas to fix a placement, and click any item to see the participant quotes behind it.
• Quadrant artifacts export to PNG.
• Chaordic Design Process chain template.
• Project summaries for chains use a summary prompt based on the chain’s method.
• Create and publish projects through the API: v1 endpoints create a project and publish it as a public Broad Listening page. (API only)

Fixed

• Chains warn and offer to regenerate before advancing on a stale summary.
• Three Horizons opens its strategy step by presenting the H2 portfolio.
• Summaries no longer show a stray code-fence wrapper, and long code blocks stay contained.
• Harmonica no longer asks participants to act on context they can’t see.
• Private session topics are redacted on the public preview.

Added

• Project-level knowledge: knowledge gathered across every session in a project, synthesised by the model and shown in collapsible sections, including what is extracted from imported sources. Material the host curates also feeds cross-pollination.
• Project-level context sources: add file, session, MCP and URL sources once and share them across every session in the project.
• PDF support for host file uploads.
• URLs as a fourth context source type: paste a link and the page is scraped into context with Firecrawl.
• Broad Listening Hub (formerly /explore): a gallery with admin curation, a starter framing picker, live polling and pseudonymized participants on the public page, prompts from Harmonica to publish, and a project tab and public header that show the project’s artifacts. See the Broad Listening page.
• Multi-lens opinion maps, starting with a Reasoning lens, with statements built from project knowledge and theme reach shown on the map.
• New layouts for project knowledge: tension comparisons and theme clusters.
• Template picker with two ways to browse, showing which methodology each template comes from.
• Sociocracy 3.0 starter chains.
• HARMONICA.md refreshes automatically from recent sessions, with a different structure for organizations and for individual practitioners.
• Chain roles can be claimed, first come first served.
• API: v1 REST API endpoints to create a project and publish a sensemaking topic. (API only)

Changed

• /explore is now the Broad Listening Hub, and the term “public sensemaking” is no longer used in the product.
• The publish dialog is split up, and projects with a public page show a Live label.
• Project members get access to the project’s sessions, up to editor level.
• Project knowledge moved to its own tab, with collapsible sections.
• Files, sessions and MCP sources now share one search index per session.
• Material the host curates now feeds cross-pollination.
• New colour palette across the opinion map.

Fixed

• Chain steps that advance automatically, and first-step prompts, now include platform guidelines correctly. Chain conversations now close by reflecting back and checking with the participant, instead of a stock closing line.
• Models other than Claude no longer drop the list of options they offer participants.
• Summaries regenerate less often when several changes arrive close together.
• Collaborators can take part in sessions instead of only previewing them. Orphan permission grants surfaced.
• Broad Listening publishing: a clearer live URL and publishing status, the URL no longer overflows the Public dialog, and the public snapshot builds when a project is published.
• Page titles and link-preview metadata are set per page, with the brand fallback fixed, “Dashboard” removed from link previews, and the mobile wordmark restored.
• OpenAI-compatible providers now use the chat completions API.
• Project page fixes: the Opinion Map gate, duplicate sources, a single summary prompt, and the edit button.
• Results page: a quieter empty state, a pre-filled email for opt-in, and card spacing.

Added

• Project summaries: generate and regenerate one summary across every session in a project, and edit it on the project page.
• Template management through the v1 REST API: create, read, update and delete templates, including chain configuration. (API only)

Changed

• Bring Your Own Model now sets default models per provider, filled in when you update your API key.
• Harmonica limits how often it shows polls and selects in chat, with stricter prompts, so they interrupt conversations less.
• The create flow saves your work as you go, so a refresh doesn’t lose it.

Fixed

• The lifetime-deal checkout page (/checkout/ltd) no longer returns a 500 error.
• Sign-up buttons now open the sign-up screen instead of the login screen.
• The end-of-chat opt-in form now shows for participants when a conversation ends.
• A duplicate email identity during onboarding now shows an error you can recover from, instead of failing.

Added

• Example Mapping chain template: an eighth multi-step chain, for finding the rules and concrete examples behind a feature or decision.
• Host preview mode: try your own session as a participant before sharing it, without affecting real results.
• Structured session knowledge (API): after its summary, each session generates typed knowledge (decisions, action items, tensions, and claims with message-level sources and quotes), stored and available through the v1 API. There is no results page UI for it yet.
• When the platform model is unavailable, the chat shows a degraded state instead of failing silently.

Changed

• Harmonica’s model calls fail over to another provider on error, so one provider’s problem doesn’t drop a conversation.
• The session creator shows the template’s own summary prompt instead of a generic default.

Fixed

• Participant join links in chains now open the correct session.
• Structured summaries no longer mention people who didn’t take part.
• Wardley maps: fixed a blank fullscreen view, and overlapping component labels are spread out.
• On Firefox mobile, the send button is no longer hidden behind the on-screen keyboard.

Added

• Chain templates with role-based participation: multi-step pipelines such as a Delphi panel, Six Thinking Hats and Wardley mapping, where each step can require specific roles. The output of one session becomes context for the next. A project is created automatically to hold the chain.
• Public chain navigation: visitors to a session’s results page can go through every step of the chain and see how the output changed from stage to stage.
• Wardley mapping output: the Wardley Mapping chain ends with a rendered map, with anchors and components placed along visibility and evolution axes, evolution arrows, and strategy markers. Hosts can correct positions.
• Review my session: for Pro, an AI review of Harmonica’s behavior on every finished session. It points out patterns such as terse turns, missed follow-ups and compound questions, with evidence excerpts, and proposes prompt rules you can add to Session Design in one click.
• Grounded summaries with verbatim quotes: summary claims show the participant quotes that support them, taken from transcripts. Hosts can add quotes to existing summaries from the results page.
• Host notification center: a bell icon with an unread count for chain progress, required actions and completions.
• Chain step emails for participants: when a chain reaches a step that needs their role, assigned participants get an email with a signed link, the session topic, the project, and a join button.
• Edit history for session prompts: every change to the facilitation prompt is recorded, with who made it, when, and from which part of the UI, including which AI suggestion it came from.
• Methodology-aware session creation: the create flow asks questions for the template you pick instead of one generic form, each with an optional AI Suggest button.
• Suggested next session: after a session ends, the results page offers a follow-up session with the same project and brief structure filled in.
• MCP and chat skill updates: harmonica-mcp passes the full set of editable fields through update_session (including widgets, results visibility, summary prompt, host context and template metadata) and accepts widgets_enabled and results_visibility on create_session. harmonica-chat was rewritten as a short entry file with reference docs per subcommand, loaded on demand, plus design steps for widgets and results visibility.

Changed

• A new results_visibility field on session creation lets hosts make the results page public, link-only or restricted.
• Role-weighted synthesis on chains: roles in a chain step can have a weight (normal, elevated or lead), and synthesis gives more weight to those roles. A k-anonymity guard keeps any individual from being identified as the source.
• When participants give short answers early, Harmonica is prompted to follow up.
• Faster session launch: welcome message and metadata generation no longer hold up the launch, and the loading spinner clears as soon as the session is reachable.

Fixed

• On iOS Safari, a message could go missing if a participant switched away from the chat tab while Harmonica was responding. The chat now detects this and recovers.
• On mobile, Enter adds a line break, and messages are sent with the send button.
• Switching to a different account no longer fails with an Auth0 invalid_request error.
• Adding a context source no longer blocks the Next button in the creation wizard.
• AI Suggestion requests that hang now time out and can be retried.
• Pages you can’t access now say you don’t have permission, instead of saying you can’t send messages.
• The export dialog no longer gets stuck.

Added

• Custom avatars: upload a profile picture instead of using initials.
• Session Design tab: HARMONICA.md, platform guidelines, facilitation approach and summary generation, in one panel.
• Skippable onboarding: new hosts can skip the /welcome form and return to it later from Settings.

Changed

• Settings are grouped into four tabs: Profile, Billing, AI models, and Harmonica API keys.
• The dashboard opens on Active sessions, with Draft and Finished one click away.
• Templates and chains appear in one grouped picker when creating a session, with a short preview of each.

Fixed

• Password reset emails are now sent through Auth0’s SMTP relay, using the same Loops templates as other transactional email.
• Email transcripts include a direct link to the session results page in every milestone template.

Added

• Polls and ratings in chat (first version): Harmonica can show polls and ratings during a conversation. Later template updates will turn widgets on per methodology.
• Brief regeneration: when you edit a session’s brief, Harmonica offers to regenerate its facilitator prompt and recap, using your project memory and active template.
• Video intro on the landing card: hosts can paste a video URL into the Session Design Landing tab, and it shows as a playable thumbnail above the welcome text.
• Project-wide invites: inviting a collaborator to a session also adds them to the project’s collaborators, and they are invited to future sessions in that project automatically.

Changed

• When a session ends, participants see a Results link when results are available, and new closing text.

Fixed

• Markdown links in the welcome card now render correctly, and bare URLs become links.

Added

• Three new models in the picker: Gemini 2.5 Flash-Lite (Free), Claude Haiku 4.5 (Pro), and Gemini 2.5 Pro (Pro). Existing defaults are unchanged.

Changed

• Voice input shows live transcripts and a waveform as you speak.
• Bring Your Own Model (BYOM): clearer error handling, better fallback, and a simpler model selection flow for hosts using their own LLM keys.
• Pro summaries now run on the large model tier (Claude Sonnet 4.6 by default) instead of the main tier. They take a few seconds longer and handle long, multi-thread synthesis better.
• Welcome card: shorter copy and less clutter.

Added

• Organizational context (HARMONICA.md): a markdown file for you as a host that is added to every new session you create, with your team’s vocabulary, decision-making norms and standing context.
• Context sources: attach reference material to a session, including files (split into chunks and embedded for search), earlier Harmonica sessions, or MCP servers. Harmonica draws on them during the conversation.
• AI-drafted welcome messages: new sessions get a welcome message generated from your brief, plus link-preview metadata for shared URLs.

Changed

• If the cross-pollination check takes too long, Harmonica replies without waiting for it.

Fixed

• A billing lockout when subscription events arrived out of order
• Paid-tier settings occasionally resetting on sign-in

Added

• Public results pages that can be shared with people who don’t have a Harmonica account, with collapsible summary sections and a stats bar above the thread.
• Invitation landing page: when someone invites you to a session as a co-editor or viewer, you see a preview of the session and can accept or decline in one click.
• Summary tools in the MCP: harmonica-mcp added generate_summary and summary_prompt, so agents can run a synthesis on demand and change how the recap is written.
• Participant tools in the MCP: a new list_participants tool and per-participant filtering on get_responses. Hosts can also set cross_pollination and distribution through update_session.

Changed

• Welcome, invitation and milestone emails are now sent through Loops, with consistent templates.
• Manage access shows each collaborator’s email under their display name, and always shows their role.

Added

• Cross-pollination now clusters responses before sharing them across threads, so shared points are more relevant and repeat less.
• Session scratchpad: Harmonica builds and refines its understanding of the conversation as it goes, and adapts later questions to what it has already heard.
• Pre-session questions in the MCP and chat skill: harmonica-mcp create_session accepts a questions array, and harmonica-chat has a configurable pre-session questions step.
• New list_telegram_groups MCP tool, so agents can list connected Telegram groups when creating a session.
• New chat skill commands: /harmonica-chat edit updates session metadata, and /harmonica-chat review analyses transcripts for facilitation issues and proposes prompt fixes.

Changed

• harmonica-mcp retries with backoff when the API rate-limits a request, instead of failing the tool call.

Fixed

• Session prompt edits now carry through the session builder
• Form field regressions during session creation

Added

• Public REST API v1: bearer-token authentication with hm_live_… keys for creating sessions, collecting responses, getting summaries and chatting. Keys are managed from your profile.
• harmonica-mcp: an MCP server on npm, so any MCP-compatible AI agent can create and query Harmonica sessions.
• harmonica-chat v2.0.0: a Claude Code companion covering session design, creation, status checks, follow-ups and invitation drafts.
• help.harmonica.chat: a docs site with the API reference, MCP server docs and host guides.

Changed

• Updated dependencies to patch known vulnerabilities. No changes for hosts or participants.

Changed

• PDF uploads now accept files up to 10MB, so multi-page reports and transcripts fit.

Added

• Edit a session’s facilitator prompt from session settings, to change how Harmonica opens, probes and closes a conversation without leaving the session view.
• Edit the summary prompt from session results: change the synthesis question, regenerate, and compare in place.

Changed

• Redesigned session results, with clearer separation between the synthesis, the responses and each participant’s thread.
• Templates page update: back navigation and clearer template cards.

Added

• Session templates for common types of deliberation: retrospectives, brainstorms, stakeholder consultations and hiring debriefs. Each template includes a facilitator prompt and a synthesis prompt.
• A shorter path from a question you want to ask a group to a session link you can share.

Added

• Voice typing in the chat input, so participants can speak their responses. If the browser doesn’t allow microphone access, typing still works.

Changed

• The model is configurable per tier: hosts can route the small, main and large tiers to different providers, with Gemini as a fallback when the main provider is overloaded.
• Cross-pollination is off by default, and hosts turn it on per session. Participants had told us they expected stricter isolation by default.

Added

• Multi-step session creation: the first step asks what you want to learn, before any questions about format. New users start with a guided introduction instead of a blank form.

Added

• Redesigned participant chat, built for mobile first, with cleaner threading and better keyboard handling.
• Chat state now persists through page reloads and reconnections.
• Share settings: set who can view, edit and take part in each session.
• Live summary updates: hosts see an indicator while the summary regenerates, and frequent edits are batched.
• Exclude a participant from the summary without losing their thread.
• Quick create from the dashboard, skipping the templates page.
• Email invitations through Gmail SMTP, replacing the previous third-party sender.

Changed

• Projects are sorted by last modified.

Fixed

• Mobile chat no longer autofocuses and zooms the page on iOS
• Session invite encryption
• Chat scrolling on long threads
• Response export for sessions with many participants
• Several form state and styling regressions

Added

• Session rating: participants can rate a session at the end.
• Editors can add sessions to projects. Before, only project owners could.
• Owner permissions for viewing and editing sessions.

Changed

• Summary prompt editing moved to its own dialog.

Fixed

• 200-session limit on the dashboard
• Renaming projects from the profile page
• AI Suggestion regression

Added

• Cross-pollination: when relevant, Harmonica shares what one participant said with another, in that person’s own thread, without naming the source. Harmonica decides in the moment whether sharing helps.
• Session knowledge (first version): Harmonica keeps context and learnings from earlier sessions in the same project.
• Better continuity when running a series of workshops with the same group.

Added

• Cross-pollination between threads (first version): Harmonica finds points worth sharing across discussions and generates follow-ups based on them. The production version launched in v0.0.9.
• AI Suggestion button next to the input field, which drafts a response. Powered by GPT-4o-mini.

Changed

• Gemini 2.0 is now the main model, for its larger context windows and faster handling of long threads.
• Cross-pollination happens at most once every three minutes per participant.

Added

• Projects: a workspace view for organising several sessions in one place. Existing sessions were migrated into it.

Changed

• Better detection of off-topic or non-substantive replies, so Harmonica can move on instead of chasing them.
• Smoother keyboard and scroll behaviour in long sessions.
• Chat results now show on the session page, so you don’t need to switch tabs to see what each participant said.
• Cleaner results page layout.

Added

• Data collection forms for gathering input before a session, such as emails or context.
• User analytics for workshop hosts, covering engagement and completion.
• Session timestamps, and sorting by creation date.
• Links on results pages to start a new session.

Changed

• Better chat error handling, and higher timeout limits for slower models.
• Editing a session’s creation fields now regenerates its template.
• Page titles and social metadata, for better link previews.
• Session visibility limits.

Fixed

• Session loading and data fetching across pages
• Message ordering in chat
• Anonymous user handling
• Chat export for sessions with several participants
• Session count on the dashboard

Added

• Personal dashboard, with cookie-based memory so you can return to your own work without logging in again.
• Role-aware session controls: hosts see management options and participants see the chat, on the same URL.
• Result links for individual participants, for sharing specific responses.
• Live summaries that build as participants respond, instead of waiting for the session to end.
• Session completion prompts, with a “Create New Session” button when a conversation ends.
• Logged-out test mode: “Create Test Session” lets people try Harmonica without signing up.
• Direct links to specific session results.

Changed

• Summaries show in a single card.
• Empty transcripts are hidden from participants.
• Participant names show in lists where appropriate.
• Transcripts are formatted as a chat.