Skip to content

CHANGELOG.md

Latest commit

 

History

History
337 lines (256 loc) · 25.3 KB

CHANGELOG.md

File metadata and controls

337 lines (256 loc) · 25.3 KB

Changelog

All notable changes to CrewForm will be documented in this file.

The format is based on Keep a Changelog.

[Unreleased]

[1.9.1] - 2026-04-23

Added

  • 7-Day Team Trial — Every new signup automatically receives 7 days of full Team-tier access (Orchestrator Mode, Collaboration Mode, Custom Tools, Messaging Channels, A2A, Team Memory, RBAC). Trial status tracked via trial_expires_at column (Migration 077) with automatic plan resolution
  • Trial Banner — Dashboard displays a contextual banner showing trial days remaining with upgrade CTA. Three visual states: active (green), expiring soon (amber), expired (red) with progress bar
  • Coolify Deployment Guide — Step-by-step documentation for deploying CrewForm on Coolify v4 via Docker Compose or Git-based methods, including Ollama integration with GPU passthrough
  • Cloudflare Turnstile — Bot protection added to login and signup forms

Fixed

  • Trial Feature Gating — useEELicense hook now falls back to workspace effectivePlan when no explicit ee_licenses row exists, correctly unlocking Orchestrator and Collaboration modes during active trial
  • Auth Callback PKCE Timeout — /auth/callback no longer shows a dead-end "Authentication timed out" error when email verification links open in a different browser context. Now checks for existing session (auto-redirects) or shows a friendly "Email verified!" success screen with Sign In button

Documentation

  • Coolify deployment guide added to docs (Getting Started section)
  • Changelog updated with 1.9.1 release

[1.9.0] - 2026-04-22

Added

  • Workflow Templates — Reusable workflow blueprints that bundle agents, teams, and triggers into single installable packages:
    • Template Definition Schema — workflow_templates table (Migration 076) with JSONB template definitions, variables, install counters, and full RLS policies
    • Template Marketplace — Browse and filter published templates by category, with install count and resource summaries
    • One-Click Install — Install Modal auto-creates all agents, team configuration, and pipeline steps from a template; users fill in {{variable}} placeholders before install
    • Create Template Wizard — 4-step wizard (Select Agents/Team → Auto-Scan Variables → Metadata & Trigger → Preview & Publish) accessible from Marketplace header and Agent Detail page
    • Variable Auto-Detection — Wizard scans agent prompts for {{variable}} patterns and auto-generates variable definitions with labels, placeholders, and defaults
    • 5 Built-in Starter Templates — Weekly Sports Coach, Content Research Pipeline, Daily News Digest, Code Review Assistant, Weekly Report Generator
    • Template Categories — Coaching, Research, Content, DevOps, Reporting, Sales & Marketing, Support, General
    • CRON & Webhook Triggers — Templates can include scheduled triggers that are created alongside agents and teams

Documentation

  • Workflow Templates Guide — New documentation page covering browsing, installing, creating, and the template definition schema
  • Changelog — Updated with 1.9.0 release
  • README — Updated Features table, Marketplace description, Documentation table, and comparison table to include Workflow Templates

[1.8.2] - 2026-04-11

Added

  • AG-UI Multi-Step Wizards — Agents can orchestrate complex, stateful, branching user interactions via requestWizard(). Supports 5 input types (approval, confirm_data, choice, text_input, form) with conditional step branching, cancellation, and a glassmorphism modal UI
  • Marketplace: Creator Analytics Dashboard — Comprehensive usage analytics for template authors replacing the basic stat cards:
    • 4 summary cards: Published Agents, Total Installs, Average Rating, Total Reviews
    • Install trend area chart (30-day daily + cumulative)
    • Rating distribution bar chart (1–5★ color-coded)
    • Per-agent performance table with installs, rating, reviews, and publish date
    • Recent reviews feed with star display
  • Navigation: Settings Sub-Navigation — Migrated 14 settings tabs from horizontal scroll bar to collapsible sidebar groups, organized into 3 categories (Configuration, Integrations, Workspace) with URL-based routing (/settings/:tab)
  • Navigation: Admin Panel Sub-Navigation — Migrated 8 admin tabs (Overview, Workspaces, Abuse, Activity, Beta Users, Licenses, Marketplace, Review Queue) to collapsible sidebar group with URL-based routing (/admin/:tab), auto-expand, and super-admin gating

Changed

  • Settings and Admin pages no longer render their own tab bars — navigation is driven entirely by sidebar sub-items
  • TopBar dynamically shows per-section titles for all Settings and Admin sub-pages

[1.8.1] - 2026-04-10

Added

  • Canvas: Copy/Paste Nodes — Ctrl+C / Ctrl+V to duplicate agent nodes on the workflow canvas. Auto-connects pasted nodes based on team mode (brain link in orchestrator, mesh in collaboration)
  • Canvas: Sticky Notes — Right-click canvas → "Add Note" for free-text annotations. 5 color presets (yellow, blue, green, pink, purple) with inline editing. Notes persist across saves and page navigation
  • Canvas: Node I/O Inspector — Click any agent node during/after a run to see collapsible Input/Output sections showing exact data the agent received and produced
  • Canvas: Autosave Infrastructure — draft_config column on teams (Migration 075), useSaveDraft / usePublishDraft hooks for future Save vs. Publish workflow
  • Canvas: Re-run Step Hook — useRerunStep mutation hook for re-executing individual pipeline steps (backend endpoint coming in a future release)
  • Keyboard Shortcuts — Added ⌘C, ⌘V to the shortcuts overlay

Fixed

  • Canvas: Node Blur — Reduced backdrop-filter blur from 12px to 4px and raised background opacity. Added will-change: transform for crisp text rendering on all GPUs
  • Canvas: Copy/Paste Not Working — Rewrote paste logic from broken nested setNodes/setEdges pattern to direct state updates
  • Canvas: Notes Disappearing — Notes now persist in config._canvas_notes and are restored when rebuilding the graph from config
  • Orchestrator: Lazy Final Answer — Brain's final_answer is now augmented with aggregated worker outputs when insufficient, ensuring full content reaches webhook dispatchers
  • Orchestrator: Webhook Output — All three executors (pipeline, orchestrator, collaboration) now consistently pass output to the webhook dispatcher
  • Webhook: DB Fallback — Dispatcher fetches output from task_runs.output as a safety net if the in-memory value is empty
  • Trello: Robust Parser — Output parser now uses 4 fallback strategies (heading-based, separator-based, paragraph-based, single-card) for reliable multi-card delivery
  • Trello: Multi-Card Delivery — Fixed test handler and card creation for structured multi-card outputs

Changed

  • Roadmap — Updated Visual Workflow Builder status to Phase 3 (copy/paste, sticky notes, I/O inspector, draft autosave)
  • README — Major overhaul for v1.8.0; comparison table reformatted

[1.8.0] - 2026-04-08

Added

  • Embeddable Chat Widget — Drop-in <script> tag to embed any CrewForm agent as a floating chat widget on any website:
    • Standalone Vite build outputs widget.js bundle served at /chat/widget.js
    • Customizable position, theme colors, and welcome message via data-* attributes
    • Domain whitelisting via cf_chat_ API keys for production security
    • Settings UI with embed snippet generator, API key management, and live preview
    • Chat history stored in chat_sessions / chat_messages tables (Migration 071)
    • Built and bundled inside the task runner Docker image for zero-config deployment
  • Knowledge Base Enhancements — Upgraded KB from basic vector search to production-grade retrieval:
    • Retrieval Testing UI — Interactive playground to query uploaded documents, see matched chunks with color-coded similarity scores, toggle between search modes, and filter by document or tag
    • Metadata Tag Filtering — Tag documents with labels (e.g., "FAQ", "Technical") and filter search results by tags. Inline tag editor on each document row with GIN-indexed tag queries
    • Hybrid Search + Reranking — Combines vector similarity (cosine) with PostgreSQL full-text search (ts_rank_cd), weighted scoring (default 70% vector / 30% text), and over-fetch + rerank strategy for better recall
    • New POST /kb/search endpoint for direct retrieval without task execution
    • Migration 072: tags column, tsvector generated column, GIN indexes, hybrid_search_knowledge RPC
  • Agent/Team Export & Import — Portable JSON data format for sharing agent and team configurations:
    • Export Agent — One-click download of agent config (model, prompt, tools, voice profile) as crewform-agent-{name}.json
    • Export Team — Self-contained JSON with all member agents embedded inline, preserving pipeline/orchestrator/collaboration config
    • Import — Upload any export file to create agents/teams with (imported) suffix; team imports rewrite all agent ID references to maintain referential integrity
    • Export buttons on Agent Detail and Team Detail pages; Import button on Agents list
    • Versioned crewform-export format (v1) for forward compatibility
  • AG-UI Rich Interactions — Agents can now pause execution and request user input via three interaction types:
    • Approval — Agent asks for permission before proceeding (Approve / Reject buttons)
    • Data Confirmation — Agent presents data for user to verify or edit before continuing
    • Choice Selection — Agent presents options for the user to pick from
    • New INTERACTION_REQUEST, INTERACTION_RESPONSE, and INTERACTION_TIMEOUT AG-UI event types
    • New POST /ag-ui/:agentId/respond endpoint for submitting interaction responses
    • Executor helper functions: requestApproval(), requestDataConfirmation(), requestChoice()
    • InteractionModal component with glassmorphism styling, countdown timer, and slide-up animation
    • useAgentStream hook now exposes pendingInteraction state and respond() callback
    • Tasks transition to waiting_for_input status while awaiting user response (5-minute default timeout)
    • Migration 069: waiting_for_input task status + interaction_context JSONB column
  • Marketplace Agent README — Agents can now include rich Markdown documentation visible to potential users:
    • New marketplace_readme field on the Agent model (Migration 070)
    • Markdown textarea with live preview in the Publish Agent modal
    • README section with formatted rendering in the Agent Detail modal
  • License Key Validation — EE license keys are now cryptographically verified:
    • HMAC-SHA256 signature verification on task runner startup (when CREWFORM_LICENSE_SECRET is set)
    • New validate-license Edge Function for on-demand key verification
    • 24-hour periodic re-validation with 7-day offline grace period
    • Validation status display (last validated, stale indicator) in the License Admin panel
    • Invalid/forged keys are automatically marked as invalid in the database
  • Observability & Tracing — Opt-in OpenTelemetry + Langfuse integration for production debugging of multi-agent workflows:
    • New tracing.ts module with zero-overhead lazy initialization (only loads SDKs when env vars are set)
    • Langfuse support — LLM calls appear as Generations with model, token counts, cost, and prompt/output previews
    • Generic OTLP support — works with Datadog, Jaeger, Grafana Tempo, and any OTLP-compatible backend
    • Task execution traces with spans for MCP discovery, LLM calls, and tool invocations
    • Team run traces with child spans for pipeline/orchestrator/collaboration execution
    • Env vars: LANGFUSE_PUBLIC_KEY, LANGFUSE_SECRET_KEY, OTEL_EXPORTER_OTLP_ENDPOINT

Changed

  • AG-UI Protocol Version — Health endpoint now reports version 1.1 (up from 1.0)

Documentation

  • MCP Protocol — Expanded docs to clearly document MCP Client Execution — agents autonomously invoke external MCP server tools during task runs via callMcpTool(). Added architecture diagram, runtime execution lifecycle (8-step flow), and example usage
  • Observability — New doc page covering Langfuse and OTLP setup, trace structure for tasks and team runs, environment variables, and Docker/self-hosted configuration
  • API Reference — Added AG-UI Protocol endpoints section (POST /ag-ui/:agentId/sse, POST /ag-ui/:agentId/respond, GET /ag-ui/health), added waiting_for_input to task statuses
  • Agents Guide — Added MCP Server Publishing section (publish toggle, tool name mapping, config snippet, client setup)
  • Visual Workflow Builder — Added Fan-Out Visualization section (fan-out/branch/merge node types, branching pattern, per-branch execution states)
  • AG-UI Protocol — Added Rich Interactions section (interaction types, /respond endpoint, React hook usage, timeout behavior)

[1.7.1] - 2026-04-01

Added

  • MCP Server Publishing — Expose your CrewForm agents as MCP tools so external clients (Claude Desktop, Cursor, other agent frameworks) can call them
    • New mcpServer.ts handler implementing MCP Streamable HTTP transport (POST /mcp)
    • Supports initialize, tools/list, tools/call, and ping JSON-RPC methods
    • Agents opt-in via is_mcp_published flag — each published agent becomes an MCP tool with auto-generated name, description, and input schema
    • Auth via dedicated MCP API keys or existing A2A keys (Bearer token)
    • Tasks created with full audit trail and AG-UI event streaming
  • MCP API Key Generation — Generate, regenerate, and revoke MCP API keys from the Settings UI
    • One-click key generation with cf_mcp_ prefix
    • Key shown once with copy button; masked preview afterward
    • Regenerate/Revoke actions with confirmation for key rotation
  • MCP Connection Config — Auto-generated Claude Desktop / Cursor config snippet with copy button in Settings → MCP Servers
  • Ollama Auto-Discovery — Dynamically detect installed Ollama models via GET /api/tags and merge them into the model selector
  • Custom Base URL — Per-provider base URL field for Ollama (and future providers), enabling remote Ollama instances on different hosts
  • MCP Publish Toggle — "MCP Publish" / "MCP Published" button on each agent's detail page to opt-in/opt-out of MCP exposure

Changed

  • LLM providers updated from 15 to 16 (Ollama with auto-discovery)
  • Task runner now dynamically resolves Ollama base URLs from the API key record instead of hardcoded localhost

[1.6.0] - 2026-04-01

Added

  • Fan-Out Pipelines — Pipeline steps can now branch into multiple agents running in parallel, with configurable merge agent and failure modes (fail_fast / continue_on_partial)
  • Fan-Out Config UI — New "Add Fan-Out" button in pipeline config panel with multi-agent checklist, merge agent selector, merge instructions, and branch failure mode
  • Fan-Out Canvas Visualization — Workflow canvas renders fan-out steps as branching nodes with amber-colored edges and merge points
  • Fan-Out Progress Rail — Real-time execution rail shows parallel branch status indicators per agent
  • Visual Builder Phase 2 — Glassmorphism node styling, live transcript panel, keyboard shortcuts, and tool usage heatmap
  • Multi-Directional Handles — Canvas supports both top-to-bottom and left-to-right layout directions with correct edge handle switching
  • Pipeline Step Insertion — Right-click any edge on the canvas to insert a new agent step between existing steps
  • README Comparison Table — "How CrewForm Compares" section with 11-row feature comparison (CrewForm vs alternatives)
  • README Hero GIF — Animated pipeline run GIF as README hero image

Fixed

  • Fan-Out Canvas Roundtrip — Canvas changes no longer destroy fan-out configuration; graph-to-config reverse mapping correctly groups fan-out branch/merge nodes
  • Edge Handle Persistence — Edge handles now persist correctly when switching between TB/LR layout directions
  • Edge Handle Direction — Edges connect to correct side handles based on layout direction

Changed

  • Updated LLM provider count from 15 to 16 (Moonshot added)
  • Pipeline validation now skips agent_id check for fan-out steps

[1.5.0] - 2026-03-27

Added

  • Model Performance Comparison — New analytics chart comparing models by speed, cost/run, and tokens/run with inline mini-bars
  • Marketplace Category Filters — Preset category groups (Customer Support, Content Creation, Data Analysis, Code Assistant, Research, Sales & Marketing, DevOps, Education) for browsing marketplace agents
  • Marketplace Category Query — Category filter plumbed through to Supabase query layer for server-side filtering

[1.4.1] - 2026-03-27

Added

  • Canvas Undo/Redo — Full undo/redo support with Ctrl+Z / Ctrl+Shift+Z keyboard shortcuts and toolbar buttons (30-entry history stack)
  • Canvas Auto-Layout — One-click dagre-based graph layout via toolbar button (TB for pipeline/orchestrator, LR for collaboration)
  • Canvas Position Persistence — Node positions saved to teams.config JSONB and restored on reload — no DB migration needed
  • Per-Task Token Breakdown — Prompt vs completion token split persisted per agent; stacked bar chart on Analytics page
  • Cost Forecasting — 30-day cost projection using linear regression with trend indicator, daily average, and mini forecast chart

[1.4.0] - 2026-03-27

Added

  • Visual Workflow Builder — Interactive Editing — Full Phase 2 canvas editing capabilities:
    • Drag agents from the sidebar onto the canvas to add them to a team
    • Delete agent nodes from the canvas (with brain agent protection in orchestrator mode)
    • Connect nodes by dragging edges to define pipeline execution order
    • Edit step name, instructions, expected output, and on-failure handling inline on the canvas sidebar
    • Auto-save with validation — invalid configs are rejected with rollback and toast notification
  • Canvas Error Boundary — Any canvas crash automatically switches to Form view with error toast, keeping data safe
  • Draggable Agent Palette — Sidebar agents show grip handles and are draggable onto the canvas

Changed

  • Workflow Sidebar — Pipeline step properties are now editable inputs (text, textarea, select) instead of read-only labels

[1.3.0] - 2026-03-27

Added

  • Tier Limits for Knowledge Base — Free: 3 docs, Pro: 25 docs, Team+: Unlimited. Quota enforced on upload with upgrade prompt
  • Tier Limits for A2A Publishing — A2A agent publishing gated to Pro+ plans; consuming remains free on all tiers
  • Embedding Provider Fallback — kb-process Edge Function now tries all available providers (OpenAI → OpenRouter) instead of failing on the first quota error
  • README Screenshot Gallery — Added 6 product screenshots in a collapsible gallery (Dashboard, Agent Creation, Pipeline Setup/Run, Marketplace, A2A Settings)
  • Landing Page Screenshots — Added "See It In Action" section with 5 product screenshots and hover effects
  • Pricing Table Updates — Added MCP Protocol, AG-UI Protocol, Knowledge Base, A2A Consume, and A2A Publish rows to both README and in-app pricing table

Fixed

  • Knowledge Base Upload — Fixed silent upload failures caused by missing storage RLS policies for the knowledge bucket
  • KB Processing Errors — Replaced fire-and-forget processing with proper error handling; documents no longer get stuck on "pending" — errors surface via toast notifications
  • Edge Function Auth — Deployed kb-process with --no-verify-jwt to prevent 401 errors (function handles auth internally)
  • Upload Button Styling — Fixed upload button text color for brand consistency

Changed

  • Provider Count — Updated from 14 to 15 LLM providers (added Ollama)
  • Landing Page Pricing — Updated tier features with Knowledge Base limits, MCP Protocol, and A2A Publish

[1.2.0] - 2026-03-27

Added

  • A2A Protocol Support — Agent-to-Agent interoperability: publish agent cards (/.well-known/agent.json), delegate tasks to external A2A agents via the a2a_delegate tool, and manage remote agents in Settings → A2A Protocol
  • AG-UI Protocol Support — Real-time SSE streaming for frontend integration via POST /ag-ui/:agentId/sse, in-process event bus, and React hook (useAgentStream.ts) for live agent-to-UI communication
  • RAG / Knowledge Base — Upload documents (TXT, MD, CSV, JSON), auto-chunk and embed with pgvector, and search via the knowledge_search agent tool
  • MCP Tool Discovery — Browse and discover available tools from connected MCP servers directly in Settings UI
  • Ollama / Local Model Support — 11 popular local models (Llama 3.3, Qwen 2.5, DeepSeek R1, Mixtral, Phi-4, Gemma 2, etc.) via Ollama — zero API keys, fully local inference
  • Abuse Dashboard — Spike detection, key rotation alerts, and workspace suspension enforcement in the Super Admin panel
  • Activity Workspace Filter — Filter the Activity tab by workspace in the Super Admin panel
  • MCP & Knowledge Base Docs — Added MCP Protocol and Knowledge Base (RAG) documentation pages to Mintlify

Fixed

  • Task Runner OOM — Resolved out-of-memory crashes on Railway deployment
  • Realtime Reconnect Loop — Prevented infinite WebSocket reconnect loop; gracefully disables Realtime after 5 consecutive failures
  • Channel Task Tracking — Tasks originating from messaging channels (Discord, Slack, Telegram) now correctly update agent activity and analytics
  • MCP Discovery CORS — Proxied MCP tool discovery through Edge Function to bypass browser CORS restrictions
  • Webhook Server Binding — Bound webhook server to 0.0.0.0 for Railway/Docker compatibility
  • Mintlify Branding — Fixed CrewForm logo and favicon on docs site

Changed

  • README — Updated with MCP, RAG, A2A, AG-UI, and Ollama features; added self-hosting section with local model instructions
  • Zapier Integration — Added per-agent and per-team filtering for Zapier triggers and actions
  • Landing Page — Removed explicit pricing row; updated self-hosting section with Ollama

[1.1.0] - 2026-03-20

Added

  • Agent Tool Support for Teams — Agents in Pipeline, Orchestrator, and Collaboration team modes can now use their configured tools (e.g. web search) during execution
  • Tool Call Tracking — Tool call logs (arguments, results, duration) are captured and displayed in both Task Detail and Team Run Detail views
  • Change Password — Users can now change their password in Settings → Profile
  • Execution Mode Tutorials — Added comprehensive tutorials for all 4 execution modes

Fixed

  • Realtime Auto-Reconnect — Task runner now detects dropped WebSocket connections and auto-reconnects with exponential backoff, preventing 4-5 minute delays in team run pickup
  • Docker Build — Fixed Dockerfile COPY paths for repo-root build context
  • Onboarding Flow — Fixed invite redirect through email confirmation and onboarding bugs for new users

Changed

  • Zapier Integration — Hardcoded production API URL (api.crewform.tech), removed user-provided Supabase URL field for security compliance
  • Landing Page — Updated CTAs from "Join the Beta" to "Get Started Free"

[0.1.0] - 2026-03-09

Added

  • Agent Management — Create, configure, and monitor AI agents from a visual UI with system prompts, model selection, and tools
  • Pipeline Mode — Chain agents together in sequential workflows with automatic handoffs
  • Orchestrator Mode — Brain agent coordinates sub-agents via delegation trees (Pro)
  • Collaboration Mode — Agents discuss and debate tasks in real-time conversation threads (Team)
  • Single Tasks — Send a prompt to any agent and get results in real-time
  • Agent Marketplace — Browse and install community-built agent templates
  • BYOK (Bring Your Own Key) — Connect your own API keys from 14 providers: OpenAI, Anthropic, Google Gemini, Groq, Mistral, Cohere, NVIDIA NIM, Perplexity, Together, OpenRouter, HuggingFace, MiniMax, Moonshot, Venice
  • Team Memory — Shared pgvector semantic search across agents (Team)
  • RBAC — Role-based access control and workspace member invitations (Team)
  • Messaging Channels — Trigger agents from Discord (slash commands), Slack, Telegram, Email, and Trello
  • Output Routes — Deliver results to Discord, Slack, webhooks, MS Teams, Asana, Trello, and Email
  • Zapier Integration — Connect CrewForm to 7,000+ apps with triggers and actions
  • Real-Time Execution — Live task execution updates via Supabase Realtime
  • Usage Tracking — Monitor token usage, costs, and performance per agent and task
  • Self-Hosting — Docker Compose deployment for production
  • AES-256-GCM Encryption — Secure API key storage
  • Row-Level Security — Workspace-scoped data isolation via Supabase RLS
  • REST API — Full CRUD via Supabase Edge Functions with API key authentication
  • Mintlify Docs — Documentation site at docs.crewform.tech
  • Landing Page — Marketing site with provider/integration marquees, feature grid, and pricing