Changelog
Recent updates and improvements to Agent.ceo
v2.8.0 — Cyborgenic Installer & On-Prem Support
Cyborgenic installer for on-prem Kubernetes deployments, real VM validation on k3s, multi-cloud support (GKE/EKS/AKS), and 265 tests passing.
- Multi-tenant isolation — namespace-per-org with automatic provisioning
- Tier auto-detection — Free, Pro, Enterprise tiers based on org configuration
- Dynamic agent lifecycle — full CRUD via REST API (POST/GET/PATCH/DELETE /api/v2/agents)
- Agent management UI — Add Agent button with form dialog, Remove button with confirmation on each agent card
- Org tree visualization endpoint — GET /org-tree returns nested JSON hierarchy
- 300+ backend tests (conductor platform), 163 DevOps lifecycle tests
- 1,858 website tests passing across 181 test suites
- Responsive dashboard — all pages optimized for mobile (375px), tablet (768px), desktop
v2.6.0 — Dashboard Command Center Expansion
Major dashboard expansion: Paperclip board embedded with SSO, plus five new management pages — cost breakdown, approval queue, org chart, task board, and audit log.
- Paperclip board embedded at /paperclip via reverse proxy — no separate login needed
- Cost breakdown page (/costs) — per-agent token usage, budget bars, spend summary
- Approval queue (/approvals) — review and approve/reject agent spend/action/deploy requests
- Org chart (/org-chart) — interactive hierarchy with live agent status, budget utilization, click-to-drill-down
- Task board (/tasks) — filterable task list by status and agent with cost per task
- Audit log (/audit) — filterable event feed with CSV export
- Beta signup now stores to Firestore when Supabase is not configured
v2.5.1 — Multi-Org Health, Token Usage, Dashboard Cleanup, Chat Demo
Health page monitors all organizations simultaneously. Usage page shows real token counts. Chat page accessible without login for demo. Removed Learning and GEA pages. Billing adds credit card entry.
- Health page monitors all orgs at once — no more single-org limitation
- Usage page fetches live token counts from API even before usage doc is written
- Billing page adds 'Add Credit Card' button for free-tier users
- Navigation cleaned up: removed Learning and GEA Evolution pages
- Chat demo mode now accessible without signing in — try AI agents before registering
- GEA and Learnings routes fully removed from navigation and routing
v2.4.0 — PinchTab Browser Automation, Voice Languages & GitHub Config
PinchTab browser automation extension, multi-language voice chat, organization GitHub settings, and improved agent credential management.
- PinchTab extension — token-efficient browser automation (800 tokens/page vs 10K with screenshots), headless & headed modes, multi-instance support
- Browser Automation skill added to Extensions skills catalog with Fullstack and Frontend agent archetypes
- Voice language selector — multi-language chat support with automatic language detection
- Organization settings page with GitHub repository configuration
- GitHub config API — token storage, create-repo endpoint, Firestore security rules
- Invite users with owner roles and agent assignment support
- Agent credentials shown for all provisioned agents (not just running)
- Usage page displays real organization data from Firestore
- Health page agent capabilities read from Firestore instead of hardcoded values
- Command Center chat clearly labeled as org-admin agent interface
v2.3.0 — User Invitations, Real Activity Data & Extensions Skills
Organization invite flow, Activity Dashboard powered by real Firestore data, Extensions skills catalog, and Command Center UX improvements.
- User invitation flow — /join page to accept org invites, with auto-redirect for unauthenticated users
- Dashboard invite banner — blue notification card when pending invites exist, with Accept Invite link
- Invite APIs — GET invite details, POST accept invite (adds to org members), GET check pending invites
- Activity Dashboard real data — removed all hardcoded mock metrics, seeded randoms, and fake charts
- Real metrics cards — active/total agents, last active time, and command count from Firestore
- useOrgActivity hook — fan-out query across agent commands subcollections for real command history
- Charts replaced with 'Activity charts coming soon' placeholder — no fake numbers
- Removed hardcoded AGENT_ROLES — agent cards now use agent.role with deterministic color hashing
- Skills catalog on Extensions page — 15 skills across 5 categories with apply dialog
- Command Center chat header shows org agent name and purpose
- Single-org auto-redirect — users with one org skip the org list page
- Organization cards expanded by default for faster access
v2.2.0 — One-Time Terminal Credentials
Secure credential management for customer organization agents. Terminal passwords are revealed once, then masked. Users can reset passwords to generate new ones.
- One-time terminal credential reveal — password shown once on first access, masked on subsequent requests
- Credential reset — generate a new password and restart the agent with one click
- Download credentials as text file for safekeeping
- K8s secret-backed credentials — passwords stored securely in Kubernetes secrets, mounted into pods
- Firestore tracking of reveal state — prevents re-exposure of revealed passwords
- Wildcard SSL for customer subdomains — all orgs served under *.agent.ceo
v2.0.0 — Public Launch: Smart Agent Scaling & Defense-in-Depth Security
Major release for public launch. Smart Agent Scaling automatically pauses idle agents and wakes them on demand, reducing costs while maintaining availability. Full defense-in-depth security with 24/24 penetration test pass.
- Smart Agent Scaling — auto-pause idle agents, wake on terminal access, inbox messages, or cron schedule
- Agent Lifecycle pause/resume — scale agents to 0 or 1 replicas from the dashboard or via MCP tools
- Ralph Loop anti-drift pattern — agents checkpoint progress and resume after restarts without losing context
- 33 MCP tools available via gateway API for agent orchestration, provisioning, and monitoring
- Defense-in-depth security — 24/24 penetration test scenarios pass, org-scoped data isolation enforced
- Common auth path redirects — /login, /signup, /register redirect to proper sign-in/sign-up pages
- 1,627 tests across 168 test suites — full lifecycle, provisioning, and admin control coverage
v1.9.0 — Launch Prep: Agent Lifecycle Polish & Status Indicators
Pre-launch polish with updated agent status colors, retry controls for failed agents, uptime tracking, and comprehensive lifecycle test coverage.
- Agent status colors match launch spec — green (running), yellow (provisioning/deploying), red (error/terminated), gray (paused/stopped)
- Retry button for failed or terminated agents — admin-only, triggers re-provisioning via PATCH API
- Agent uptime display showing time since last start (e.g. '2h 15m', '3d 5h')
- Usage hours display — mock-ready for metering backend deployment
- 39 component tests covering all agent lifecycle states, admin controls, retry, uptime, and usage hours
- Invite auto-join moved to login flow with 7-day invite expiry
- Extension catalog fallback when Firestore is unavailable
v1.8.0 — Launch Ready: Multi-User Orgs, Terminal Proxy & Metering
Major release with multi-user organization support, secure terminal WebSocket proxy, agent-hours metering, email verification, and agent run/pause controls.
- Multi-user organization support with members subcollection — add team members by email with admin roles
- Terminal WebSocket proxied through Next.js — browser never connects to internal gateway directly
- xterm.js terminal for customer org agents via gateway WebSocket proxy
- Agent-hours metering with free hours counter and usage dashboard widget
- Email verification required on signup — unverified users redirected to verification page
- Agent run/pause controls on dashboard — scale agents to 1 or 0 replicas from the UI
- Provisioning UX — terminal button disabled until agent is active, deploying status indicators
- Voice button restored as standalone floating widget on org and terminal pages
- Voice TTS model selection fixed — CJK languages use sonic-multilingual, Hebrew returns graceful error
- Auth redirect loop fixed for returning users with restored Firebase sessions
- re4ai.com onboarding — CRE role, org template, and showcase entry as first customer
- SELL-001 payment enforcement — free tier capped at 3 agents, billing mode toggle
v1.7.0 — CLI Type Badges & Extensions Catalog
Dashboard shows agent CLI types and available extensions. New /extensions catalog page for browsing MCP servers, skills, hooks, and plugins.
- Agent cards display CLI type badge — Claude Code, Codex, Gemini CLI, etc.
- Agent detail page shows Extensions tab with MCP servers, skills, hooks, and plugins (read-only)
- New /extensions catalog page with type and CLI compatibility filters
- Verified badge for platform-approved extensions
- Tests for CLI badge, extensions page, and Firestore type interfaces
v1.6.0 — Gateway Provisioning & Org Isolation
Organization provisioning wired to Gateway API with terminal namespace routing and strengthened data isolation.
- Org provisioning creates namespaces and deploys agents via Gateway API
- Terminal routing uses org namespaces for customer agent access
- Org isolation hardened — static imports, fail-safe ownership checks, delete protection
- Chat panel embedded in organization detail page with per-agent deep linking
v1.5.0 — Launch Polish: Sorting, Hierarchy & Agent Creation Fixes
Fixed 4 founder-reported bugs: agent sorting, organization tree diagram, version display, and agent creation reliability.
- Active agents now sort first — expanded status detection to include RUNNING, deploying, restarting, and saving states
- Organization tree diagram wraps multiple root agents under a single Organization node instead of showing them as separate hierarchies
- Agent creation via chat now returns explicit error messages when database writes fail, preventing false success reports
- Version display updated to v1.5.0 with proper environment variable configuration
v1.4.1 — Org Chat Agent CRUD via Natural Language
Organization agents can now create, list, update, and delete agents via natural language chat using Anthropic tool_use.
- Chat with your org agent to manage agents: 'add a real estate agent', 'list my agents', 'delete the marketing agent'
- 4 Anthropic tools: create_agent, list_agents, update_agent, delete_agent — all backed by Firestore
- Auth-scoped: tools only available when user has an organization
- Safety limit: max 5 tool rounds per request to prevent infinite loops
- RoleId-first lookup for update/delete operations (handles both role names and doc IDs)
- 16 new tests covering all tool operations, auth, validation, and loop safety (1,574 total)
v1.4.0 — 100% Test Coverage & Org Creation Fix
Achieved 100% page and component test coverage. Fixed org creation to atomically provision template agents. Cleaned up production console logging.
- 100% page test coverage: all 37 pages tested (1,567 tests / 167 suites)
- 100% component test coverage: all 27 components tested
- Client-side org creation now atomically creates org + org-agent + template agents via Firestore batch writes
- Fixed 'Agents (0)' display bug on newly created organizations
- Removed stray console.log statements from production code
- E2E tests passing against production (19 public page tests)
- Zero TypeScript errors, zero ESLint warnings, zero console errors on all pages
v1.3.0 — Org Agent Chat, Full CRUD & API-First Hooks
Chat buttons now connect to your org agent. Full organization CRUD with template provisioning, agent management, and server-side validation.
- Chat with your org agent directly from the organization detail page
- Per-agent chat buttons with ?agentId= deep linking
- POST /api/organizations atomically creates org + org-agent + template agents
- PATCH /api/organizations/[id] for updating org name and status with validation
- POST /api/organizations/[id]/agents for spawning sub-agents
- DELETE /api/organizations/[id]/agents/[agentId] with org agent protection
- All mutation hooks use API-first pattern with Firestore fallback
- Paused and archived org status badges in list and detail views
- Removed fake metrics data — only real Firestore data is displayed
- 1,499 tests passing across 153 test suites
CSP Security Consolidation & Favicon Fix
Consolidated duplicate Content-Security-Policy headers into a single comprehensive policy. Added favicon.ico rewrite for browser compatibility.
- Merged two conflicting CSP headers — browsers enforce the intersection, which silently blocked WebSocket and Firebase resources
- Single CSP now covers all Firebase Auth, Firestore, and Cloud Functions domains
- Added /favicon.ico rewrite to serve from existing /icon.svg (eliminates 404)
- Full QA sweep: zero console errors across all production pages (desktop + mobile)
Server-Side Org API & Middleware Hardening
Server-side organization CRUD bypasses Firestore rules for reliable operation. Middleware fix prevents public pages from being auth-gated.
- Server-side GET/POST /api/organizations using Firebase Admin SDK — reliable even without Firestore rules deployed
- API-first pattern: hooks try server API first, fall back to direct Firestore for real-time updates
- Middleware fix: /license-agreement, /security, /changelog no longer incorrectly require authentication
- Comprehensive org-data hook coverage with server-side API integration
Go-Live: Full Test Coverage & Security Hardening
Achieved 100% test coverage across all pages, API routes, hooks, and components. Added input sanitization for voice session routes.
- Security hardening: input sanitization and auth enforcement across all API routes
- Voice hook tests: useVoice (19 tests) and useLiveKitVoice (9 tests) with full browser API mocking
- Input sanitization: agent_system_prompt, agent_id, and agent_name validated in voice session route
- Organization creation flow wired to gateway API with template picker and provisioning status
- Rich chat with markdown rendering, code blocks, and file attachment support
Launch Polish & SSE Integration
Final launch polish: mobile-responsive layouts, updated CTAs, graceful provisioning UX, chat fallback improvements, and live SSE gateway for real-time agent events.
- Mobile-responsive grids on landing page (stats, tech partners) at 375px
- Hero CTA updated: 'Get Started' → /sign-up with launch pricing banner
- Provisioning page: smooth progress indicators with status normalization
- Chat page: graceful fallback UX when gateway is unavailable
- SSE gateway proxy for real-time agent activity streaming
- Multi-tenant data scoping: all API routes enforce org-level isolation
- Multi-tenant data scoping verified across all API routes
Security: Org-Scoped Data Isolation
All API routes now enforce organization-level data isolation using authenticated session context, preventing cross-tenant data access.
- Board, innovations, meetings, and activity APIs scoped to auth org
- Query parameter org_id ignored in favor of server-side auth context
- Fallback responses include org_id from authenticated session
Pre-Launch Quality Hardening
Comprehensive quality sweep: WCAG 2.1 AA compliance, verified performance budgets, zero console errors.
- 100% API route coverage for auth, chat, organizations, and billing flows
- WCAG main landmarks on all pages including legal, auth, and error pages
- E2E smoke test: signup → setup → dashboard flow verified on production
- Performance budget verified: Homepage FCP 336ms, Pricing FCP 244ms
- Zero TypeScript errors, zero console errors across all 20+ pages
SEO & Bundle Optimization
Improved search engine visibility with structured data and OpenGraph tags, plus significant bundle size reductions through lazy loading.
- JSON-LD structured data on changelog, innovations, and organization pages
- OpenGraph tags added to all marketing and auth pages
- Lazy-loaded HeroTerminalChat: homepage bundle reduced 238kB → 114kB
- Deferred Firebase Auth on marketing pages: -121kB on /pricing and /innovations
- Tree-shaken D3 imports on agents-canvas: -13kB
Agent Activity Dashboard
Comprehensive activity dashboard with agent grid, organization metrics, token usage charts, and session timeline.
- Agent grid with live status indicators and role badges
- Organization metrics: tokens, tasks, escalation rate, self-fix rate
- Token usage trend chart (7-day) and estimated daily cost
- Monthly budget usage progress bar with spend tracking
Page Load Performance
Reduced gateway timeout from 3s to 1s across all API routes, making pages like /innovations load 2-3x faster.
- Innovations, boards, and activity stream APIs optimized
- Faster fallback to built-in data when gateway is unavailable
Production Stability Fixes
Eliminated WebSocket 404 errors on the agents canvas page and updated stale innovation pipeline data.
- WebSocket connection now only attempts in development mode
- Innovation pipeline data reflects current progress (100% task completion)
- Updated SEO metadata to match current marketing positioning
Innovation Pipeline Dashboard
New /pipeline page showing active innovations, validation progress, and task completion across the platform.
- Interactive pipeline stage flow with click-to-filter
- Expandable innovation cards with metrics and progress bars
- Summary cards for active, in-validation, and shipped innovations
Activity Feed Enhancements
The live activity feed now supports event type filtering and shows metadata badges for richer context.
- Filter by event type: commits, tasks, messages, deploys
- Metadata badges showing branch names and environments
- Compact mode for dashboard widget (5 events max)
Clipboard Error Handling
Fixed unhandled promise rejections when clipboard access is denied across all copy-to-clipboard buttons.
- Credentials dialog, settings, MCP config, and image builder
- Graceful fallback when clipboard API is unavailable
Accessibility Improvements
Added missing ARIA labels to icon-only buttons for screen reader compatibility.
- Error dismiss buttons on organizations and agents pages
- WCAG 2.1 AA compliance for interactive elements
Live Agent Activity Stream
New real-time activity feed showing agent actions across your organization with auto-polling and connection status.
- Dedicated /activity page with full event history
- Dashboard widget with compact 5-event view
- Auto-polling every 30 seconds with connection indicator
Token Usage Dashboard
Monitor AI token consumption with burn rate tracking, budget status, and usage breakdown by agent.
- Daily burn rate with trend indicators
- Budget status with progress bar and days-until-limit
- Usage charts by day and by agent
- Recent usage events with model and token details
Voice Chat Widget
Voice-enabled conversations with your AI agents via Cartesia TTS and LiveKit WebRTC for real-time voice calls.
- Browser voice recording with Deepgram STT and RTL language detection
- LiveKit WebRTC voice calls with mute/unmute and agent speaking indicator
- Cartesia Sonic-2 TTS with 18+ language support
- Call/end call interface with connection status
Dashboard Onboarding
New users see a guided empty state with quick actions to create their first organization and explore the platform.
- Welcome banner with setup completion detection
- Quick action cards for common first steps
- Documentation section with getting started guide
Chat & Learnings Bug Fixes
Fixed authenticated user errors on /chat and /learnings pages caused by Firestore session management and error handling.
- Chat: session document created before message subscription
- Learnings: API fallback now triggers when Firestore errors
Innovations Showcase Page
Public /innovations page showing the platform's innovation pipeline with stage progression and metrics.
- Pipeline stages: Concept, Problem-Solution Fit, Validation, Growth
- Innovation cards with progress bars and task counts
- How-it-works section explaining the BML methodology
Want to see what we're building next? View our innovation pipeline