See the work before you choose an expert
- -Explore concrete examples: a GitHub lead shortlist, customer-feedback treemap, X conversation chart and competitor posting comparison. Names, messages and numbers are clearly labeled as illustrative.
- -Public Connect your agent now opens a compact dropdown for Codex, Claude Code, Cursor and OpenClaw. The expert workspace keeps its focused setup dialog.
- -Models, Tools and Skills each show four aligned entries. Expert and skill labels use clean text instead of generic icons, and the work example follows your chosen agent.
Your experts, with interfaces for their work
- -Experts uses a two-column gallery. Connect your agent opens one compact setup dialog for Codex, Claude Code, Cursor or OpenClaw, with terminal sign-in also available.
- -GitHub research has an inline audience explorer: search saved profiles, filter by source availability, inspect evidence and export a shortlist. Existing reports open in the same versioned result folders.
- -Use expert saves a workspace shortcut without starting a task. Company context and preferences are shared with your coding agent, with conflict checks when saving changes.
- -Your experts reflects saved workspace connections and accessible work. Existing X research keeps its folder workspace; no extra collection or outreach is started by opening a view.
One column for your results and expertise
- -Results, Experts and My expertise now show one item per row on every screen size.
- -Home stacks its account panels and expert list vertically. Search, filters, navigation and result folders keep their existing behavior.
Clearer billing, API keys and expert discovery
- -Billing separates your balance, task usage, credit transactions and payment receipts. Payment methods, monthly plans, grants and spending settings stay together under Balance.
- -API keys has a dedicated page with named keys, creation dates and status. Create a unified key in a focused dialog, save its secret once, and confirm a specific key before revoking it.
- -Find expertise by task or role. Outcome cards include examples and a tailored request for your connected agent; X conversation and product-feedback requests keep separate drafts.
- -The directory includes a clearly marked Answer with Books integration. Tinker audience is no longer listed. Existing workspaces, saved results and pilot access checks remain in place.
Clearer payment recovery and research delivery
- -Billing checks previous top-ups when you return. Pending payments keep their original checkout; confirmed payments show the credit grant and receipt instead of asking you to pay again.
- -Native research delivers a versioned result folder with saved findings, captured sources and limits. Runs with no accepted findings still deliver an explanation. Historical saved runs enter the same delivery queue without new analysis.
- -Request status and run history show the same saved outcome and recorded charge. Missing billing details remain explicitly unavailable; they are never displayed as a free run.
- -Sources now supports Add evidence for pasted text or a text/Markdown file, and Manage accounts for existing research contexts. Saving evidence returns source IDs and does not start analysis.
- -File-read failures offer reselect and paste recovery. Saved skills open their canonical URL; partially unavailable libraries retain the records that loaded.
- -Profile links are validated before save and at storage. Privacy and Quickstart now describe the current storage and folder experience; unknown routes show a useful not-found page.
A shared workspace for your X research
- -Open the X research expert workspace to browse Context, Requests, Results and Sources in one folder view.
- -Continue in your agent carries the selected request or versioned result into your coding harness. Existing briefs and source records remain available without another setup wizard.
- -Reports, HTML deliverables and tables open inside the workspace. Folder navigation and document reading now scroll independently, including in standalone result packages.
One place to create skills and read results
- -Create a skill from one card, then choose a conversation, GitHub repository or files.
- -Private result readers keep the collapsible workspace sidebar, with layouts that adapt to the available reading space. Shared customer links remain standalone.
- -Results and Work open the folder reader. Supported older deliveries can be organized from their saved output without another paid run.
- -Saved video deliveries include workspace-only playback, the script and original source links alongside versioned result files.
Your saved work, easier to find
- -Managed skill tasks open their versioned result reader from older Work links. Workspaces without research topics can still show saved skill tasks and results.
- -Archive a workspace for your account from Manage workspaces, then restore it anytime. This organizes your switcher; tasks, billing, shared access and saved results continue unchanged.
- -Saved skill cards are clickable throughout. Expert panels use clearer What it does, Your work and Agent setup tabs, without stacking another workspace dialog on top.
Save a skill first. Use it on your terms.
- -Saved skills now open a private workspace with Overview, Method, Tools, Checks and History. Saving does not start a task or publish your expertise.
- -Create skills from selected ChatGPT, Claude, Cursor or Codex conversation exports, a supported public ChatGPT share, public GitHub files, or a file/folder upload. Conversation extraction is a creation feature with a cost review before execution.
- -Instructions, supporting files and tool requirements stay together in immutable versions. Optional checks remain marked untested; credentials and raw conversation histories stay outside the skill package.
- -Use skill copies an authenticated reference for your own agent. Supported cloud tasks remain optional; generic skills cannot accidentally run as a personal-retro task.
- -Customer results retain their separate folder reader and sharing controls. Making a private skill available for others to hire remains a separate, forthcoming step.
One expertise library, with credit-backed managed runs
- -My expertise brings saved research setups and private methods together. Playbooks is no longer a separate destination; existing links and saved work remain available.
- -Turn a useful conversation into reusable know-how through a CrowdListen expert, or import selected repository instructions. General expertise belongs to a workspace and does not require a research topic.
- -Supported personal-retrospective and method-extraction tasks show the workspace balance and maximum charge before running. Actual usage is charged once; unused reserved credits return to the balance.
- -Managed tasks continue in the background and retain status, cancellation, source references and versioned results. Private methods stay separate from shareable deliverables.
- -Results follows the selected workspace and includes published results from its members, with current access checked before display.
- -Coding agents can list, quote, run, inspect and cancel saved expertise. Generated methods remain drafts until reviewed; arbitrary scripts, automatic training and automated behavioral grading are not enabled.
- -The 2.3.1 agent setup includes shared workspaces and opens versioned result files from Copy context for your agent through the same connection. Failed tasks show their settled charge and do not imply that a result is still on the way.
Versioned results and private expertise tests
- -Results supports search and expert filters, with a simpler card layout and one Copy context for your agent action.
- -Result packages bring Markdown, HTML, source records and interactive tables into one folder-based reader with version history and selected-file sharing.
- -Approved managed tasks publish their saved deliverables into result packages in the background. Private methods remain separate from customer results.
- -My expertise introduces a private pilot for importing conversations or selected repository methods, reviewing instructions, and testing personal-retro or chat-skill-miner capabilities in a contained Codex environment with Fireworks inference.
- -Generated methods remain drafts for review. Arbitrary tool installation, general paid execution and automated behavioral evaluation are not enabled by this pilot.
Clearer capabilities and a more consistent website
- -Billing shows the available balance in credits without a duplicate dollar-equivalent line or routine credits-ready banner.
- -Docs keeps the workspace sidebar when you are signed in, including direct links and refreshes. Documentation links preserve in-app navigation; signed-out readers retain the public site layout.
- -Public signup buttons now say “Get 500 free credits”, making the existing welcome offer explicit.
- -Capabilities are grouped into By CrowdListen and By other experts. First-party X research is available; third-party capabilities are labeled planned.
- -The landing page explains the difference between skills and hired expertise, with interactive expert, toolbox and outcome examples. Notification delivery and brief reuse remain labeled previews.
- -The centered Research, Marketing and Sales headline rotates continuously every six seconds, with a gentler transition, pause control and reduced-motion support.
- -The outcomes section now matches the surrounding white sections. Slack was removed from the list of agent interfaces.
- -About uses the same public navigation as the rest of the website. Sign-in and signup backgrounds fill the available space, and the duplicate Back to CrowdListen link is removed.
Workspace billing: balance, payments, history and grants
- -Billing now has Balance, Payment methods, History and Credit grants tabs for the selected workspace, including workspaces without a research topic.
- -Add to balance accepts a whole-dollar amount from 1,000, shows the credit allowance and continues to Stripe Checkout. Standard top-ups provide 100 credits per $1.
- -Monthly plans fund the same balance: 90 for 10,000, or $200 for 25,000. Unused credits carry over; renewal can be canceled in Billing.
- -Each workspace owned by a verified account can receive its one-time 500-credit welcome grant. Refreshing or returning from checkout does not duplicate the grant.
- -Payment methods can be managed through Stripe. History distinguishes usage, grants and payments; Credit grants shows remaining promotional credit and supports code redemption.
- -Choosing a different monthly plan replaces an unpaid checkout. Balance is added only after successful payment verification; canceling checkout does not add credit or charge a card.
Managed X playbook tasks: limited pilot
- -Added a managed X playbook task experience for an explicitly enrolled operator pilot, using saved X evidence in an isolated hosted environment.
- -The task viewer supports background progress, clarification questions, answers, cancellation and revision requests, with previous reports retained when a revision fails.
- -Citations open the exact saved source version behind a claim. Tasks remain bound to one authorized workspace and source collection; agent access follows account permissions.
- -Incomplete provider usage is surfaced for reconciliation instead of being treated as a free or successfully completed run.
- -This remains a sponsored pilot, not a general paid launch. Real-source output quality, successful revisions and paid settlement still require validation; the landing page does not promise that every expert uses this runtime.
Expert capabilities and simpler agent setup
- -CrowdListen is now presented as “Expertise your agents can hire”, with concrete capabilities and usage-based pricing across the public website.
- -Available X tasks cover conversation research, account activity and new questions over saved evidence. The competitor posting playbook is distinguished from generally available research.
- -Copy setup for agent provides instructions for Codex, Claude Code, Cursor, OpenClaw and the terminal. The extra terminal command view toggle is removed.
- -Signed-in navigation brings Home, Experts, Playbooks and Results together. Public pages stay accessible from sign-in and signup.
- -New API keys are created in a modal with combined collection and export permissions. Existing keys retain their current access.
Working signup emails and readable documentation
- -Public pages, docs, sign-in and signup render useful HTML before JavaScript loads, improving first paint and access for crawlers and agents.
- -Confirmation emails use the CrowdListen sender with a working verification button, fallback link and plain-text body. Disposable-account signup, activation and password sign-in were verified.
- -Signup includes resend, delivery feedback and an option to correct the email address. Existing-account and unconfirmed-email errors offer recovery actions.
- -Google consent uses CrowdListen branding. Unverified personal social links were removed from the public footer.
- -Authentication docs explain coding-agent login, token refresh and key permissions. A searchable API catalogue and downloadable schema accompany the task-oriented guides.
Prepaid research with estimates before you run
- -Self-serve research can use workspace credits without the legacy subscription gate. Verified owners can enable usage and open a one-time credit checkout.
- -Run review shows an estimated range, a maximum charge and available balance. Insufficient-credit messages lead back to workspace Billing.
- -Actual recorded usage is settled once per run, and unused reserved credit returns to the balance. Work already performed can consume credits even when a run is partial or fails.
- -Fixed the browser preflight that blocked approved runs using an idempotency key. Empty-result runs now explain that no matching sources were captured.
- -Reading saved reports, inspecting evidence and checking progress do not consume additional task credits.
Versioned reports with controlled sharing
- -Research briefs open as versioned web pages inside the CrowdListen viewer, with saved history and links to supporting evidence.
- -Reports are private by default. Owners can choose Only me or Anyone with the link, then revoke access or replace the link.
- -A shared page keeps its published version while private revisions continue. Viewing history does not overwrite the original report or evidence.
- -The viewer was simplified by removing Ask for changes and Download HTML while keeping history and sharing controls.
Reusable workflows and evidence you can inspect
- -Create, import and revise reusable workflows through the workspace or a connected agent. Saved versions share one history, and conflicting edits are rejected instead of overwriting newer work.
- -Result themes expand their supporting posts inline, with official X embeds where available, saved source inspection and links to the originals.
- -Evidence grids adapt from four columns on desktop to one on narrow screens. Opening or closing evidence keeps the selected heading in place.
- -Copying a result to an agent includes revision-specific findings and access to cited and uncited saved sources, subject to the account’s permissions.
- -Fixed a navigation race that could reopen a workflow immediately after it was closed.
Insights and Experts: a simpler workspace
- -Home, Insights and Experts are the primary destinations. Docs and one-time Agent access remain secondary.
- -Choose a source collection on Home. Insights brings source configuration, saved originals and evidence-backed findings together for that workspace.
- -Insights includes saved analysis workflows and run history. Experts separates specialist requests, delivered reports and supporting tools; expert work starts through your connected agent.
- -Previous Skills, Directory, Loops, Sources, Workflows, Artifacts and Results links remain usable, including saved edits, OAuth returns and report deep links.
- -New personal API keys start with ingest permission instead of full access. Browser-login agent setup requires no additional key.
- -The public demo uses fictional source evidence and readable reports. Original product screenshots are restored and supported tools appear as an authentic logo grid.
UI redesign: new sidebar, home page, tools, agent page, demo entity
- -Sidebar redesign: brand header (CrowdListen / Signal Engine), workspace selector on top, SCOPE nav (Home, Insights, Agent, Tools, Docs), entities list with status dots (green/amber/gray) and actions dropdown (configure/delete), user profile footer.
- -Home page (/listening): Harvest Engine for quick entity creation → redirects to full /ingestion config. Entity overview card with inline signal counts (Signals, Testimonials, Claims, Feedbacks). Theme treemap and velocity chart with tab toggle. “Settings” button replaces old “Configure New Pool”.
- -Demo entity: new users auto-get a real OpusClip demo entity (55 insights, 6 categories) pre-loaded. Data comes from live Supabase, not mocks. “Demo” badge in sidebar, prominent “Create Your Own Entity” banner on home page. Dismissible via entity dropdown menu.
- -Agent page (/agent): dedicated AI analysis page replacing floating Insights FAB. Painting background hero (nature oil paper) with centered input bar and quick prompt chips. Toggleable sessions sidebar matching ProjectChatView card container design. AI Specialists section (6 skill cards) with “Install Skill” CTA.
- -Tools page (/tools): three-tab layout (Platform Analyzers, Continuous Sync, AI Utilities). Inline URL analyzer with flower garden painting background, platform circle selector, real API integration, sentiment gauge results. Agent Mode section with MCP install. AI Utilities tab now has CTA buttons matching Sync tab style.
- -Project chat: toggleable sessions sidebar on left edge of chat container. Shows recent workspace analyses with current session highlighted. PanelLeftOpen icon when collapsed.
- -Entity creation flow: Harvest Engine creates entity → redirects to /ingestion → enrichment progress banner shows 5-step setup (entity created, enriching, select sources, configure focus, run sweep). Entity auto-selected after creation.
- -Signed-out nav: Home and Tools now visible for unauthenticated users. Home links to landing page, Tools to /tools.
- -Dark mode removed — light-only design system. AppHeader bar removed — workspace info consolidated in sidebar.
Routing redesign, $149 Tester tier, research queue, backend cleanup
- -Routing redesign: removed Analyze as standalone page. Analysis now accessible via floating “Analyze” panel on Insights page. Sidebar nav: Dashboard, Ingestion, Insights, Tools, Docs.
- -Restored Listening (renamed Dashboard) and Ingestion pages — accidentally deleted in earlier cleanup. All 6 listening chart components restored.
- -New $149/mo Tester tier: 3 entities, manual upload via Sources, insights + analysis sessions, free analyzer tools. No automated research. Self-serve signup.
- -Pricing page updated to 4-tier grid: Tester (400), Growth ($800), Enterprise. Starter gets weekly research, Growth gets daily.
- -Human-in-the-loop research queue: users request research → operator processes locally → results appear. Entity config tracks research_status (requested/processing/complete).
- -Backend cleanup: deleted 27 legacy platform search tools (~13K LOC). Created _platform_extractors.py shared module. All imports reconciled. Platform knowledge preserved as 6 SKILL.md packs in MCP harness.
- -Free analyzer rate limit changed from 3/hour to 3/day per IP. Auth-aware ToolsNav shows Dashboard button when signed in instead of login/signup.
- -Tools page: added AI Specialists section (6 role cards) and Agent Mode section (MCP install + 3 core tools).
- -Ingestion page: tester tier sees “Upload Content” + upgrade CTA. Starter+ sees full research config with “Request Research” flow.
- -Local operator scripts: push.py gains mark_research_started/complete functions. research.sh auto-sets status at start/end. New check_requests.py polls for pending requests.
Frontend refactoring, Research hub, ScrapeCreators removal
- -Service consolidation: merged contextService + wikiService into knowledgeService (single service per pages table). 12 files updated, 200+ LOC saved.
- -ProjectChatView split: extracted ChatPanel (481 LOC) and StudioPanel (751 LOC) from 5,411-line monolith. Main file reduced to 4,318 LOC.
- -Blog externalization: created BlogPostLayout shared component for SEO/header/CTA/FAQ chrome. Converted 8 blog posts to use shared layout. Fixed prerender-meta.js to detect SEO in JS object syntax.
- -Research hub: new /research “Coming Soon” page with three content types (Ranks, Reviews, Marketed vs Real). Email notify CTA. Added Research link to ToolsNav.
- -Navigation fixes: pricing and research pages now show ToolsNav for unauth users instead of sidebar + duplicate auth buttons.
- -Dead code cleanup: deleted 54 dead blog stubs, 3 dead pages (ForCreators, NarrativeExtraction, RedditAnalysis), unused ApiKeyCard component, unused exports from insightService and autoTaggingService.
- -ScrapeCreators disabled: API key removed from production, all deep search flags set to false. Was burning 100+ credits per analysis (all 6 platforms maxed out).
- -TikTok transcripts replaced: new yt-dlp + Gemini pipeline (auto-subs tier 1, audio download + Gemini 2.5 Flash tier 2). Zero ScrapeCreators dependency.
- -TikTok direct API hardened: User-Agent rotation (5 modern Chrome UAs), exponential backoff on 429/403, consecutive error limits, msToken expiry detection.
Product page redesign: Tools, Ingestion, Listening dashboard
- -Tools page redesign: category tab switcher (All Tools, Video, Forum, Community, Utility), bento grid with platform SVG icons and brand-colored hover borders, pricing badges, “Activate” CTAs. Removed Agent Roles and Developer Docs sections (moved to Analyze page).
- -Ingestion page: 4-stat monitoring header with monospace labels and colored progress bars (Monitoring Channels, Ingested Documents, Normalized Signals, Synthesized Insights). Quick Ingest Toolset with 4 buttons (Upload File, Upload Survey, Upload Link, Paste Text).
- -Listening dashboard: Executive Summary moved inline into header card (was separate dark card). Sentiment meter repositioned to top-right of header. Treemap reverted to original Recharts visualization. Removed signal cards below treemap.
- -Landing hero dashboard: 6 clickable mock pages (Listening, Ingestion, Insights, Analyze, Tools, Docs) with concrete AI Coding Agents data. Platform brand icons on Tools demo.
- -GEO audit and fixes: FAQ schema on landing page, BreadcrumbList on tool pages, sameAs links (LinkedIn, Discord, GitHub), OG tag alignment, removed fabricated AggregateRating, 5 new AI crawlers in robots.txt, context-transfer added to sitemap.
- -Duplicate login/signup buttons fixed on landing page. Dead nav block removed. Dark mode locked to light-only. Changelog page restored at /docs/changelog.
Landing page redesign: interactive dashboard hero, updated feature sections, mobile responsive
- -Interactive hero dashboard: Linear-inspired real DOM dashboard replaces static chat mockup. Shows sentiment meter, executive summary, 5-category treemap with hover effects, velocity chart, and tier-2 drill-down on click.
- -Feature sections renamed and updated: Knowledge→Ingest (topic detection via browser agents), Workspace→Insights (synthesize feedback into scored insight cards), Workers→Analyze (deep dives with agent skills and outcomes).
- -New landing page screenshots for all three feature sections (ingestion, insights, analysis) with clean bordered frames replacing painting backgrounds.
- -Mobile responsive: dashboard collapses to header bar on mobile (<640px), tap to expand. Treemap switches from 3-column grid to 2-column layout on mobile. Hero heading scales down to text-3xl on small screens.
- -Fixed persistent blank space below dashboard caused by min-h-screen propagating through flexbox into dashboard main element (min-h-0 fix).
- -Sidebar nav in dashboard demo: CrowdListen logo, tracked entity indicator, 6 nav items (Listening, Ingestion, Insights, Analyze, Tools, Docs).
Notion-style workspace switcher, local ingestion operator, platform diversity
- -Notion-style workspace switcher: replaced Slack-style rail with dropdown (user info, workspace list, settings, sign out). Entity selector labeled “Tracked Entity” with type subtitle.
- -Sidebar “Sessions” section shows recent projects (was “Recent analyses”).
- -Analyze page: Web UI / Agent Mode toggle. Agent Mode shows MCP install + 3 core tools. AI Specialists merged from Workers page.
- -Ingestion page: split into dashboard view (configured entities) vs stepper (new entities). “Run Sweep” button at top for configured entities. Entity creation step for new entities.
- -Tools page: cleaned up layout (matches app design), removed separate nav, added G2 Review Analyzer.
- -Treemap tier 2 click navigates directly to insight detail page (removed redundant inline list).
- -Skill-informed suggested actions: insight categories map to CrowdListen skills (pain_point→heuristic-evaluation, competitive→competitive-analysis, etc.).
- -Local ingestion operator (localhost:5555): auto-processing dashboard with gstack browser research. Entities auto-queued, searched across TikTok/Reddit/YouTube via browser, pushed to CrowdListen.
- -Agentic research SKILL.md: 4 signal types (user desire, distress, gaps, consensus), iterative search→evaluate→refine loop.
- -Vercel proxy now default Reddit search path. ScrapeCreators Reddit usage reduced to fallback only.
- -Platform intent weights raised: TikTok 0.4→0.7, YouTube 0.3→0.8, Instagram 0.4→0.75 across all intents.
- -Deep-extract pipeline: browser-collected URLs now processed through /ingest/deep-extract for transcript + comment extraction.
- -Fixed import os scoping bug in analysis.py that blocked the analysis pipeline.
- -All automated entity research scheduling disabled (manual-only via dashboard or UI).
Unified pipeline, skill agents, MCP harness v3
- -Unified enrichment pipeline: posts and comments processed in one code path. Comments auto-contextualized with parent post + transcript. All opinions produced in English for cross-language clustering.
- -Opinion splitting fixed: splitter now uses raw review content (Pros/Cons structure) instead of condensed summary. Reviews correctly split into 2-5 atomic opinions.
- -LLM cluster verification: naming step sees ALL evidence per cluster, flags misfits for reassignment. Cross-cluster merge pass consolidates duplicate themes.
- -Tutorial content correctly labeled as marketing_narrative instead of user pain points.
- -Post-naming quality gate: removes zero-impact and off-topic insights, fixes placeholder titles.
- -Skill agent routing: skills in project chat now route through hosted agent with real tools (search_content, cluster, compare) instead of question rewriting.
- -MCP harness v3: simplified to 3 tools (recall, analyze, ingest) + 8 skills. recall supports 5 modes: search, entities, workspaces, insights (with evidence), content.
- -Embedding deadlock fixed: replaced sync self-HTTP-calls with direct OpenAI client + batch API. Pipeline normalization went from 8 minutes to seconds.
- -Source cap at 100 before LLM relevance filter for faster analysis.
- -Workspace tooltips on sidebar rail. Share workspace button in header.
- -Sign Up link added to landing page nav.
- -Pin/acknowledge insights: three-section layout (pinned, active, acknowledged).
- -Theme-to-insight navigation: click “View full insight” in Listening page to jump to Insights page with detail panel open.
- -Velocity chart: “All” time window default, smart monthly/weekly bucketing, evidence dates from parent posts.
- -Evidence panel in ThemeTreemap tier 2: expandable source quotes per theme.
- -Progress bar stays visible throughout analysis with step labels.
- -Related questions prompt: platform-agnostic (no more “on Reddit”, “on G2”).
- -Demo accounts created for AvePoint, Vizard, OpusClip, EvenUp, Harvey, Otter.ai, UpGuard workspaces.
- -New entities: Hootsuite, Brandwatch, Sprinklr, Nectar Social tracked in CrowdListen workspace.
- -Dead code removed: compiled_truths and insight_contradictions tables dropped.
Insight quality overhaul, 4-tier pricing, entity tracking paywall
- -LLM source relevance filter: GPT-5.4-mini checks each content source for direct relevance to the tracked entity before clustering. Filters out tool integration bugs from other products that merely mention the entity.
- -Entity-aware insight naming: clustering LLM receives entity name and description. Off-topic clusters about different products are automatically dropped.
- -Unicode normalization: entity keywords with non-breaking hyphens now match content with regular hyphens. Fixed 89% of opinion_units being incorrectly filtered.
- -Adaptive HDBSCAN: clustering parameters scale with dataset size to prevent all-noise results on small datasets.
- -Pipeline status in header: visual stages (Find Sources → Understand Sources → Produce Insights) moved from bottom bar to compact inline display in the library toolbar.
- -Default insight tab: Evidence tab is now the default view when opening an insight detail (was Analytics).
- -Actions tab removed from insight detail (redundant with Analytics).
- -4-tier pricing: plans renamed to tester/starter/growth/enterprise with updated Stripe checkout.
- -Entity tracking paywall: EntityActivationGate modal shown when users hit plan entity limits.
- -Sentiment normalization hardened: handles all input scales (0-1, 0-100, unexpected) robustly.
- -Entity-scoped connectors: observation connectors inherit entity context automatically.
- -Subscription check on entity research trigger with per-plan entity limits.
Workspace viewer mode: share your workspace with design partners
- -Workspace viewer mode: invited users see a clean read-only experience. All write actions (create, edit, delete, run analysis) are hidden for viewers across ProjectChatView, AppSidebar, AnalyzePage, and Library.
- -Viewer-aware footer: workspace viewers see “You’re viewing this workspace” with a “Request analysis” button instead of the sign-in prompt shown to anonymous users.
- -Sidebar protection: rename and delete actions hidden for viewers. Share link remains accessible.
- -AnalyzePage protection: “Create new context” tile and edit/delete/toggle actions hidden for viewers.
- -RLS migration: workspace-sharing SELECT policies added to pages, content_store, entity_insights, insight_evidence, and knowledge tables. Viewers can now read all data linked to workspace entities and projects.
- -Existing infrastructure activated: useWorkspacePermission hook, ShareWorkspaceDialog, AcceptInvitePage invite flow, and 12+ isReadOnly guards in ProjectChatView were already built but not wired to workspace roles. Now connected.
Insight cards, hybrid search, exploratory analysis UX
- -Card-based analysis: LLM-orchestrated insight cards replace the fixed analysis template. 14 card types: takeaway, sentiment, themes, dimension breakdown, engagement highlights, competitor mentions, platform breakdown, actions, sources, and more.
- -Hybrid search (gbrain-style): content_store embeddings + keyword relevance + RRF fusion. Questions now find semantically relevant sources, not just the most recent.
- -Exploratory UX: clicking a card triggers a contextual follow-up analysis. Themes expand in-place. Sentiment segments generate natural language questions.
- -Generated Artifacts panel: “Artifacts” button in the analysis view opens a workspace-wide list of all generated documents (PRDs, battlecards, support docs). Actions card now generates artifacts inline.
- -Topic Glossary: left sidebar shows entity KB topics. Clicking a topic runs a focused analysis.
- -Source citation modal: clicking a source opens a popup with full content, platform, engagement, and external link.
- -Performance: simplified analysis pipeline — removed redundant intent classification, entity extraction, and KB context fetch. Knowledge mode goes straight to synthesis (~10s vs ~30s).
- -Sources card: 4 per page pagination, collapsed by default in chat, expanded in full analysis.
- -Panel layout: flex-basis 0px fix prevents overflow. Right panel toggles via “Generate Full Analysis Details” card.
- -Comment enhancement: short comments get parent post context prepended for better embedding/search.
- -Unified knowledge table: pages + content_store merged into single `knowledge` table (26K+ rows). One search function for everything.
- -MCP harness v2: 21 tools consolidated to 5 (save, recall, analyze, entities, observe). All SKILL.md workflows updated.
- -Database cleanup: 39 dead tables dropped. Observations and artifacts merged into unified storage.
- -Promotional content separation: company marketing filtered before synthesis. “Official” badge on sources.
- -Evidence cards: platform thumbnails/icons on left side of each evidence item.
Insights-first library, artifact generation, and performance boost
- -Insights tab: default view in the library. Pre-computed insights with category badges (pain point, feature request, praise, churn risk, competitive gap, trend), urgency and impact scores.
- -InsightCard component: shows cluster size, top evidence snippets, and engagement-weighted ranking.
- -InsightDetailView: full-width detail with Evidence, Actions, and Artifacts tabs. Expandable source context with platform badges.
- -Suggested actions: auto-generated per insight based on category (e.g., feature requests show “Add to roadmap”, pain points show “Create support article”).
- -Artifact generation: one-click document creation from insights — PRD, battlecard, support doc, outreach template, content guide. GPT-4o-mini powered.
- -Performance: analysis pipeline now ~15s (down from 40s) via parallelized execution. Console spam fixed (profile upsert only on sign-in).
- -Data ingestion: Discord export, Reddit URL + comments, YouTube video + comments. Auto-triggers insight pipeline after import.
- -MCP harness: 2 new tools (get_insights, generate_artifact). Tool count 13 → 15.
Feature simplification: tasks, scheduling, and agent partners removed
- -Removed: Kanban/task management UI and all backend task endpoints (/agent/v1/tasks/*).
- -Removed: Goal decomposition pipeline and endpoints (/agent/v1/hosted/tasks/decompose).
- -Removed: Background task queue and scheduler endpoints (/agent/v1/hosted/schedules).
- -Removed: Agent Partners enrollment network and endpoints (/api/agents/*).
- -Removed: Skills generation and refresh endpoints (/projects/{id}/skills/refresh).
- -Simplified: MCP tool surface reduced from 22 to 13 canonical tools. Execution pack removed.
- -Kept: Analysis pipeline, entity management, observation pipeline, hosted agent chat, context/KB, audio transcription.
- -/agents route now redirects to home.
Context Library, KB-grounded analysis, Intercom analyzer, batch URL import
- -Context Library page (/library): unified three-tab page with Content (masonry gallery), Themes (engagement-weighted synthesis), and Docs (TipTap editor). Entities in app sidebar as foldable section.
- -KB-grounded Analyze page: default search mode changed to knowledge base. Searches pre-indexed content (~2s) instead of live social media (~60-90s). Web search available as explicit fallback.
- -Batch URL import: paste URLs in the Import zone, system extracts content + comments, promotes high-engagement opinions, clusters into themes, writes to living KB doc. Full pipeline in one click.
- -Intercom Conversation Analyzer (/tools/intercom-analyzer): new tool page for customer support conversation analysis. Cross-channel correlation with social content.
- -Channel Source Adapter: uniform interface for Discord, Slack, Intercom. Each adapter implements 3 methods, shared pipeline handles thread promotion and response formatting.
- -Painting hero header: search bar with oil painting background matching analyzer pages. Content/Themes/Docs tab toggle with platform filter pills.
- -Auth redirect fixes: existing users no longer stuck in sign-up loop. All signin/signup links preserve the page users came from.
Embeddable UGC widgets, AI Context Transfer tool, Apollo-style navigation
- -Embeddable UGC widgets: Brands can embed crowd-verified TikTok content on their website with one script tag. Notion case study on landing page with People/Agents toggle view.
- -Self-serve embed demo: Enter your website URL and email, CrowdListen searches TikTok for positive UGC about your brand, shows a live preview with real creator videos.
- -AI Context Transfer tool (/tools/context-transfer): Move AI memory between ChatGPT, Claude, Codex, Gemini, and any agent. Copy export prompt, paste results, upload chat history, or connect via MCP.
- -Progressive disclosure UX: Context transfer shows copy prompt first, then reveals paste/upload/MCP sections after clicking “Unlock full coverage with CrowdListen Harness”.
- -GEO-optimized: Context transfer page has SoftwareApplication + HowTo + FAQPage JSON-LD schemas, 19 target keywords, rich noscript content for AI crawlers.
- -Apollo-style navigation: Shared ToolsNav component with multi-column dropdowns (Analyzers with descriptions, Community, Utility, Featured card). Consistent across landing page and all tool pages.
- -Resources dropdown updated: About, Blog, Changelog under Learn. Discord, X/Twitter, Contact under Connect.
- -Tools section on landing page: Workers renamed to Tools. 8 analyzers + Context Transfer shown in categorized grid (Video, Discussion, Community, Utility).
- -Embed showcase on landing page: Notion case study with 3 real TikTok creator videos. Inline brand search with lead capture. What People See / What Agents See toggle.
- -Sign-in/sign-up redesign: Apollo modal-over-dashboard pattern with blurred workspace background, CrowdListen brand colors (burnt amber #C2662D), email-first layout, Google + Microsoft side by side.
- -Analysis progress events: Backend now emits real SSE progress events during analysis (relevance filtering, clustering, synthesis, KB write). Frontend progress bar shows actual pipeline stage instead of fake timer.
Full pipeline: source detail modal, comments, workspace scoping, onboarding redesign
- -Source detail modal: Click any raw source row to open a modal with full post content, platform badge, engagement metrics, link to original, and threaded comments with author + likes.
- -Raw sources table: Now filters out comment rows (shows posts only). Click to open modal with comment thread.
- -Knowledge page: Dynamic folders from database. Users can create, rename, delete custom folders.
- -Onboarding: 5-step wizard (Setup → Verify → Sources → Frequency → Focus) with oil painting header and company verification.
- -Workspace scoping: Entities and folders workspace-scoped. Switching workspaces shows only that workspace’s entities.
- -Competitor discovery: “Find competitors automatically” button uses Exa + Gemini to suggest competitors.
- -Review sites: Trustpilot, Capterra, G2, App Store added as search sources.
- -File upload: DOCX, XLSX, CSV, MD, TXT, PDF files as entity sources.
- -Workers page: Renamed from Capabilities. Skills renamed to Roles.
- -React.Fragment fix: Resolved production crash in RawSourcesTable.
Source picker labels, platform name rendering, UI cleanup
- -Source picker: Hacker News and Product Hunt now show full names instead of abbreviated labels.
- -Knowledge page: Fixed “HackerNews” display to “Hacker News” (proper spacing).
- -Platform names and colors: All 12 analysis platforms now render with proper display names and brand colors in source post lists.
- -Source Posts section hidden from expanded analysis and analysis results views for a cleaner layout. Data still available in XLSX export.
Typography refresh, landing page redesign, platform cleanup, paywall banner
- -Typography: Switched from Instrument Serif/Sans to Plus Jakarta Sans across the app.
- -Landing page: Redesigned problem section, updated hero copy with rotating channel names, renamed Harness to Workspace throughout.
- -Platforms: Removed TruthSocial and Bluesky, Xiaohongshu set to Coming Soon, source limits increased from 15 to 20 per platform.
- -Paywall: Persistent analysis limit banner with email capture.
- -Kanban: “Start Task” button replaces “Assign to Agent”, task completion polling.
- -Fixed: Project and context deletion now works (RLS policy fix), decomposition status constraint fixed.
- -Reddit: Exa fallback when Reddit API returns 0 results, global search fallback for niche queries.
16-platform intelligent search, context UX improvements
- -15-platform search: added Hacker News, YouTube, TikTok, Instagram, Product Hunt, Threads, Polymarket, and Truth Social to the source picker with SVG icons. Platform selector expanded from 4 to 13 platforms.
- -Intelligent search pipeline: new query planner with intent classification, weighted RRF fusion (replaces hardcoded 60/40 blend), semantic clustering via embeddings, and progressive retrieval for thin-result platforms.
- -Polymarket adapter: search prediction markets for real-money-backed forecasts and odds on any topic.
- -Truth Social adapter: search posts via Mastodon-compatible public API.
- -Context detail expand/collapse: maximize button on context detail dialog toggles between standard and full-width view.
- -Hidden scrollbars: observation stream, themes dashboard, knowledge page, and context detail modal now use invisible scrollbars for a cleaner look.
- -Fixed UUID validation error when auto-saving virtual pages (observations and themes) that have non-UUID IDs.
- -Plugin manifests: .claude-plugin/plugin.json for Claude Code marketplace and gemini-extension.json for Gemini CLI.
Task orchestration, entity source capture, KB browser cleanup
- -Homepage goal → task board: type a goal on the landing page and it routes to the task board where an agent breaks it into subtasks in the Proposed column.
- -Approve All now launches a sequential hosted batch instead of only moving cards, with progress streamed back into the task UI as cards advance Proposed → In Progress → In Review → Done.
- -Graceful fallback: if the agent backend is unavailable, the goal is saved as a single task card so nothing is lost.
- -Task board workflow restored to explicit Proposed, In Progress, In Review, and Done stages instead of collapsing execution into two ambiguous board states.
- -Entity creation now captures preferred source platforms plus official Twitter/X, YouTube, and blog/RSS channels to make entity crawls more grounded.
- -Context tab replaced with workspace-scoped Knowledge Base browser: browse, search, create, and edit pages inline.
- -Add Task button on KB pages: manually create a task card from any knowledge page.
- -Generate Tasks from KB pages: AI-powered task generation from page content.
- -Removed DecompositionPanel floating review UI — decomposed tasks now appear directly on the board.
- -Context panel button renamed from “Open Doc” to “Open Panel” for clarity.
Context Library source, entity integration, docs overhaul
- -Context Library: new source option in workspace chat, alongside Neural Web Search, Reddit, Twitter, etc. Queries your saved pages and knowledge base.
- -Entities on Context page: always-visible section showing tracked companies, competitors, and products with an “Add Entity” CTA when empty.
- -Entity-observation connections: observations in Feedback Insights now show entity name badges when linked. Entity filter added to observation stream.
- -Theme-entity linking: theme cards in Feedback Insights surface which entities are referenced by observations in each theme cluster.
- -Task navigation: workspace task cards now link directly to the global task board. Task count in studio panel is clickable. Deep links via ?card= URL param.
- -Removed Web/Knowledge/Both toggle from workspace chat — Context Library is now a proper source in the source picker instead.
- -FAQ section restyled with bordered cards instead of flat divider layout.
- -Collapsed sidebar workspace icon alignment fixed to match nav item padding.
- -Sources MCP docs removed: all social listening tools are now part of the Harness. /docs/mcp/sources redirects to /docs/mcp.
- -Documentation overhaul: fixed stale tool counts (21 canonical tools across 7 packs), removed Gemini CLI from auto-configured agents, updated product surface names (Kanban → Tasks).
Tasks rename + workspace-scoped task boards
- -Route rename: /kanban → /tasks. Old URLs redirect automatically.
- -All user-facing “Kanban” text renamed to “Tasks” (sidebar, footer, landing page, onboarding).
- -Workspace-scoped task boards: each workspace gets its own board. Switching workspaces shows a different board.
- -Existing boards auto-migrated to your default workspace.
Workspaces — group projects by client or team
- -Workspace switcher: Slack-style selector in the sidebar to create, switch, edit, and delete workspaces.
- -All project lists (Analyze, Sidebar, Context) are now scoped by active workspace.
- -Context page (Business Context, Observations, Themes) filters by workspace projects.
- -Default workspace auto-created on first login; existing projects backfilled.
- -New projects automatically assigned to the active workspace.
Search intelligence, sign-in improvements, related questions
- -Search Intelligence: adaptive query expansion, source quality scoring, and cross-platform deduplication for improved results.
- -Microsoft sign-in button added to /signin page, matching the OAuth flow already available on the landing page.
- -Related questions now cache to localStorage with 24h TTL and show shimmer skeletons while loading.
Landing page conversion overhaul, privacy policy, FAQ section
- -Hero section split: unauthenticated visitors now see inline email signup + Google/Microsoft OAuth buttons (Apollo-inspired); authenticated users still see “Connect your agents” tool selector.
- -Three-button header: “Log in”, “Get a demo”, “Sign up free” replaces the previous two-button layout for unauthenticated visitors.
- -Lead capture modal: two-step progressive form (email → name + use case) with auto-triggers on scroll depth, time-on-page, and exit intent.
- -FAQ section: 8 CrowdListen-specific questions in a two-column accordion layout, placed after Changelog on the landing page.
- -Privacy policy page (/privacy) with 11 sections covering data collection, PII redaction, CCPA/GDPR rights, and third-party services.
- -Microsoft OAuth support via Supabase Azure provider alongside existing Google OAuth.
- -Unified contact email: all references updated to terry@crowdlisten.com across the entire frontend.
- -SignUp page now pre-fills email from query params when redirected from the hero signup form.
- -SQL migration for leads table with RLS policies for anonymous insert and admin-only read.
Blog typography overhaul, reports moved to S3
- -Fixed blog typography: replaced broken text-xs CSS with comprehensive text-lg typography system for headings, lists, tables, blockquotes, and links across all blog posts.
- -Upgraded OpenClawMarkdownPost renderer with Card wrapper, shadow, wider container, and larger hero title to match polished inline JSX posts.
- -Improved 15 blog posts across Guides, AI Knowledge Base, and Frameworks clusters with expanded content, consistent formatting, and concrete examples.
- -Moved 42 report HTML files (~339MB) from git to AWS S3 bucket with Vercel rewrite proxy — repo shrinks from ~350MB to ~11MB.
- -Added scripts/upload-report.sh for future report uploads to S3.
Context page redesign, card unification, feedback insights, memory import fix
- -Context page redesigned as 2-tab hub: Business Context (all documents with path-based filter chips) + Feedback Insights (observations + themes across all projects).
- -Sources, agents, and connectors moved to gear-icon config panel instead of toggle buttons.
- -Unified card hover pattern across all pages: flat base, subtle shadow + accent border on hover. Applied to 7 card components.
- -Observation stream filter bar reorganized: type pills now include category icons, “Hide Resolved” moved to header as a small text toggle.
- -Observations and themes now support global view (across all projects) with project name badges.
- -Fixed critical bug where AI memory export parsed entries but never saved them to the database.
- -Added user-scoped queries: getAllObservations() and getAllThemes() in observationService.
- -Removed project filter from kanban board (kanban is global).
- -Updated CLAUDE.md with card component pattern documentation and context import pipeline reference.
Context editing, related questions, kanban improvements, docs nav redesign
- -Rich markdown editing for context detail pages using TipTap editor with live markdown conversion.
- -LLM-generated related questions (OpenAI gpt-4o-mini with Gemini fallback) focused on what people think, believe, and debate across communities.
- -Docs navigation moved to sidebar on /docs/* routes, replacing inline details/summary menu.
- -Inline agents panel on Context page — toggles like Sources instead of navigating away.
- -Kanban board is now global: removed project filter.
- -Task agent selector shows only connected agents from Supabase agent_connections instead of hardcoded executor types.
- -Workspace pagination reduced from 7 to 3 projects per page (1 row + create tile).
- -Fixed React strict mode bug where related questions got stuck on loading.
- -Fixed expandedThemes not resetting when switching analyses.
- -Fixed treemap theme accordion animating on initial mount.
- -Added stripContextPrefix utility for cleaning [PRIOR CONTEXT] prefixes.
Unified pages table, knowledge browser, watch/sync CLI
- -All knowledge (wiki pages, memories, ingested files) now stored in a unified `pages` table with path-based identity (UNIQUE per user+path) — replacing separate memories and wiki_pages tables.
- -New /knowledge page: two-panel file tree + markdown viewer for browsing all pages, with inline editing and search.
- -Harness CLI: `npx crowdlisten-harness watch ~/folder` auto-syncs local markdown files to the pages table with content hashing for idempotent updates.
- -Harness CLI: `npx crowdlisten-harness sync ~/folder` for one-shot folder sync.
- -Recall tool now uses semantic search (pgvector cosine similarity) with keyword fallback, searching the unified pages table.
- -All wiki tools (wiki_list, wiki_read, wiki_write, wiki_search, wiki_ingest, wiki_log) retargeted to the pages table. project_id is now optional.
- -Agent backend: all Python services updated to query pages instead of memories/wiki_pages. search_pages RPC replaces match_memories.
- -Frontend services (wikiService.js, contextService.js) updated to read from pages table.
ChatGPT and Claude support, unified agent setup, agent status detail
- -Added ChatGPT and Claude as remote MCP agents on the landing page with step-by-step configuration instructions for mcp.crowdlisten.com/mcp.
- -All agent setup dialogs now show the learn.md prompt as the primary setup method — agents read the guide, install MCP, register via API, and return a verification link.
- -Clicking any agent card now opens a detail dialog showing tier, online/offline status, analyses count, active tasks, completed tasks, and last activity.
- -Reordered landing page agents: OpenClaw, Claude Code, Codex, ChatGPT, Claude. Removed Gemini CLI.
- -Renamed hero section from “Connect your AI tool” to “Connect your agents”.
- -Expanded ExecutorSelector with OpenClaw, ChatGPT, Claude, Cursor. Removed Gemini CLI.
- -Standardized tool counts across all docs from outdated “10”/“20”/“49” to ~28 tools across 6 packs.
- -Removed Gemini CLI, Amp, and Qwen Code from all agent lists, docs, badges, and executor configs.
Context-first workspace, project wiki, and shareable knowledge base
- -Context page redesigned around connected sources, agents, and a searchable Context Library with cleaner gallery-style browsing and file navigation.
- -Project workspace now uses a clearer context-first studio flow, with wiki pages opening directly inside the editor workspace.
- -Added a unified markdown wiki per project: page CRUD, templates, maintained index/log pages, and URL-addressable wiki navigation.
- -Analyses and saved highlights can now auto-ingest into the wiki, updating relevant pages and keeping project knowledge structured over time.
- -Context Library supports Mine/Shared tabs, publish-to-team sharing, Shared badges, and keyword-first recall without embedding generation.
- -Added wiki migration tooling plus fixes for NULL project rows, local env loading, and new-page auto-open behavior.
OpenClaw playbooks, insight pipeline, and frontend cleanup
- -Published OpenClaw playbooks and blog content in the docs hub, including user interviews, sentiment analysis, PRD-to-agent requirements, and context engineering.
- -Added insight pipeline settings and service wiring so extracted insights can move more cleanly into the project board workflow.
- -Introduced the extraction service path and related service refactors to support the new insight pipeline and docs content.
- -Expanded the .md knowledge base frontend with sharing UI, Mine/Shared tabs, Shared badges, and keyword-based recall.
- -Removed dead frontend pages and utilities, and simplified Integrations after the earlier expert-system removal.
MCP tools connected to project board, Ollama removed
- -MCP add_context and record_learning tools now mirror knowledge (decisions, patterns, learnings) to the project board Insights tab.
- -Project board insights from MCP tools show a small “AI” badge to distinguish from manually created insights.
- -remember/recall MCP tools now write and read from Supabase context_blocks instead of local JSON, with local fallback preserved.
- -Context blocks support optional project scoping — blocks on the /context page show project and AI source badges.
- -Removed Ollama LLM provider from CLI setup, API, and configuration UI.
Inline digest settings, navigation upgrades, and agent list cleanup
- -Email Digests settings tab now shows full configuration inline (toggle, frequency, sources, focus, save) instead of opening a modal.
- -Settings page supports ?tab= deep linking (e.g. /settings?tab=digests) for direct navigation from other pages.
- -Context page “Raw Data Feed” and “Email Digests” links upgraded from tiny text to prominent outlined buttons.
- -Removed Amp and Qwen Code from supported agents list in kanban setup dialog, agent connections, and executor selector.
Landing page agent-first redesign
- -Hero H1 redesigned: “Turn {rotating topic} into context every agent acts on.” with subtitle “What one agent learns, the whole swarm inherits.”
- -Input placeholder updated to “What context do your agents need?”
- -New problem/solution section with Smithsonian painting background — “Agents fly blind without compoundable context” flowing into “Knowledge from one becomes knowledge for all.”
- -“From research tool to context layer” section redesigned with painting background.
- -Added “Trusted by AI-native teams” logo bar: TikTok, Moveworks, Scale AI, Portia Labs, Vizard.
Feed redesign, source management, and ingestion backfill
- -Reverted Raw Data view to table layout with Platform, Source, Content, and Date columns.
- -Replaced horizontal source strip with collapsible “Connected Sources” panel — responsive grid with health indicators, sync tooltips, and pause/delete controls.
- -Added content deduplication — strips duplicate title prefix from raw content display.
- -Added Sync button with spinner animation to Connected Sources header.
- -Moved Insights/Raw Data toggle inline with Feed title to avoid notification overlap.
- -Added useSourcesData hook for centralized source and feed state management.
- -Discord and Slack first-crawl backfill — new sources now fetch the past 7 days of messages instead of only the latest 100.
- -Added immediate ingestion trigger when adding new channel sources.
- -Added API Key Management portal at /api — create, list, and revoke API keys from the dashboard.
- -Expanded API Reference docs from 7 to 17+ endpoints: cluster, enrich, LLM proxy, premium analyze/insights/research, and key CRUD.
- -Created migration 010_api_key_crud.sql with api_keys table, usage_records table, RLS policies, and increment_usage RPC.
Visual extraction, paid API, and search parallelization
- -Added visual extraction system: BrowserPool, RequestInterceptor, VisionVerifier, and 4 platform adapters (Twitter, TikTok, Instagram, Xiaohongshu).
- -Added paid API endpoints: POST /api/v1/cluster, POST /api/v1/enrich for scope-gated analysis.
- -Added LLM proxy: POST /api/v1/llm/complete, GET /api/v1/llm/models for keyless multi-provider LLM access.
- -Added premium endpoints: POST /api/v1/analyze, /insights, /research with tier-based rate limiting.
- -Added unified API key validation (api_key_auth.py) with SHA-256 hashed lookup and legacy fallback.
- -Added shared rate limiter (rate_limiter.py) with Redis monthly counters and tier-based caps.
- -Parallelized hybrid search with ThreadPoolExecutor (~50s to ~20s).
- -Deleted 922 lines of dead code (InstagramBrowserSearch, TikTokBrowserSearch).
- -Bumped Exa results from 20 to 30, total cap from 30 to 40.
Agent engagement guidance in learn.md and SKILL.md
- -Updated learn.md (v3.1.0) — renamed “Share Your Research” to “Share and Engage”, added engagement ladder table and community engagement subsection.
- -Expanded SKILL.md (v3.2.0) Knowledge Exchange section with full behavioral guidance: when to browse, vote, feedback, and comment.
- -Added comment type taxonomy, good/bad examples, and knowledge deposit quality traits to SKILL.md.
- -Documented stigmergy feedback loop with CrowdListen’s agent and four engagement progression patterns.
- -Added “Engagement Ladder in Practice” recipe with full browse→vote→feedback→comment curl flow.
Agent knowledge exchange and Xiaohongshu support
- -Added agent knowledge exchange — read-only comment section in Feed for agent-authored comments.
- -Published SKILL.md (v3.1.0) at crowdlisten.com/SKILL.md — comprehensive Intelligence API onboarding for agents.
- -Added comment endpoints to Intelligence API (GET/POST/DELETE) for agent knowledge deposits.
- -Added Xiaohongshu (小红书) platform support with source picker integration.
- -Added DELETE /api/agents/owned/{id} endpoint for agent disconnection.
- -Reduced deploy size by ~97 MB via painting image compression.
- -Fixed agent registration idempotency to prevent duplicate cards.
Mobile responsiveness and background image fixes
- -Fixed hero overlap — Humans/Agents toggle no longer collides with heading on mobile.
- -Background images now fully cover hero and Cal.com sections on all viewports using inset layer pattern.
- -Tightened agent tab content for mobile with smaller text and reduced spacing.
- -Increased hero card min-height on mobile for better breathing room.
- -Redesigned docs hub layout with improved navigation.
- -App-wide palette aligned to earthy tones across cards, articles, and feature showcase.
Landing page redesign with earthy palette
- -Replaced abstract blur hero with painting-style background (Alaskan Coast Range).
- -Switched to earthy/powder-petal color palette across hero, tabs, and sections.
- -Added painting-backed demo frames in FeatureShowcase (Feed, Workspace, Kanban).
- -Added “Knowledge from one becomes knowledge for all” core narrative section.
- -Added “Stay on the frontier” section with source composer and feature cards.
- -Added Cursor-inspired changelog card grid on landing page.
- -Removed ProductsShowcase and CuratedSection from landing layout.
Agent management improvements
- -Added disconnect button to agent cards for revoking agent connections.
Kanban Tasks tab, expert system removal, and E2E tests
- -Added Tasks tab to studio with project-scoped kanban boards and context-aware task generation.
- -Added Project Context tab to kanban with full analysis text display.
- -Removed expert system entirely from frontend (34 files).
- -Added kanban onboarding flow for new users.
- -Fixed Run button flex alignment in KanbanCard.
- -Added Playwright E2E tests and updated CI workflow.
Unified composer, skill packs, docs redesign, and agent discoverability
- -Added MCP Servers section: /docs/mcp (overview), /docs/mcp/kanban (10 tools), /docs/mcp/sources (13 tools across 6 platforms).
- -Added Authentication page with claim flow, scopes, and security rules.
- -Added Errors page with error envelope format, status codes, rate limits, and retry logic examples.
- -Redesigned API Reference with per-endpoint blocks, method badges, parameter tables, and tabbed code examples (cURL/JS/Python).
- -Expanded Integrations from bare list to cards per agent stack with MCP details.
- -Grouped sidebar navigation: Getting Started, Agent API, MCP Servers, Guides, Resources.
- -Redesigned docs hub with 3-column card grid.
- -Created SKILL.md agent instruction set with routing table, decision tree, auth cheat sheet, and 5 recipes.
- -Created openapi.yaml (OpenAPI 3.0.3) with all 7 endpoints and two security schemes.
- -Rewrote llms.txt as structured index with MCP server listings.
- -Expanded llms-full.txt from 49 lines to ~900 lines with full API spec, MCP tools, and workflows.
- -Updated learn.md to v2.0.0 with MCP servers section and cross-links.
- -Added .yaml to Vercel static file route for openapi.yaml serving.
- -Created 6 reusable docs components: ApiEndpoint, ParamTable, CodeBlock, Callout, StepList, DocsCard.
- -Unified chat composer with source picker, skill picker, and URL paste support.
- -Added Claude Code skill packs and plugin manifest.
- -Added Human/Agent tab toggle to landing hero section.
- -Replaced old analysis page UI with workspace treemap visualization.
- -Renamed kanban workspaces to agent runs in UI and CLI commands.
- -Removed legacy mcp-server directory, ExpertCrowdToggle, and updated sidebar labels.
- -Hardened Slack/Discord feed integration.
Source routes, ingestion proxy, and digest fixes
- -Added source API routes and fixed analyses origin constraint.
- -Enabled Discord and Slack tabs in channel sources modal.
- -Added Reddit proxy on EC2 with multi-layer fallback for resilient ingestion.
- -Added digest pipeline diagnostic endpoint.
- -Fixed digest user preferences upsert and added debug logging.
Legacy cleanup and docs foundation
- -Removed BettaFish legacy code, dead backend files, and unused dependencies.
- -Reorganized footer into Product, Docs, Company, Contact, Channels columns.
- -Added /docs, /docs/quickstart, /docs/api-reference.
- -Added llms.txt and llms-full discoverability surfaces.
- -Improved /api/agents contract typing and OpenAPI metadata.