Changelog

Fixed an issue where connected channels (LINE, Facebook, Instagram) showed as “Not connected — credentials needed” on the chat agent detail page, even when the integration was working correctly.

Root cause: The mapDocToEventTrigger() function was not including chatAgentId from Firestore documents, so the page couldn’t match triggers to their bound agent.

The setup wizard now also correctly writes the channels[] field when binding a trigger, ensuring future agents display channel status accurately.

JieGou now offers professional integration services to get customers from setup to value in days, not months.

Service tiers:

• Guided Setup ($1,000) — 1-hour call + async support + custom recipes
• Migration Package ($3,000–5,000) — n8n workflow migration with security audit
• Full Integration ($3,500–5,000) — end-to-end custom build with training
• Managed ($2,000–3,000/mo) — ongoing optimization with monthly reports

Console intake form at /help/integration-request — structured request form with industry, channels, automation goals, and n8n-specific fields for migration requests.

Integration Health Dashboard at /deploy/integration-health — per-platform health scores showing credential status, error rates, activity metrics, and actionable recommendations. Identifies expiring credentials and suggests expansion opportunities.

3 new industry packs:

• F&B Social Content — 8 recipes + 2 workflows for restaurants (daily specials, menu carousels, food captions)
• E-commerce Social Content — 7 recipes + 2 workflows (product launches, testimonials, seasonal campaigns)
• Professional Services Content — 7 recipes + 2 workflows (thought leadership, case studies, compliance review)

Marketing pages: /solutions/integration-services and /solutions/n8n-migration-services with pricing, case studies, and security comparison for n8n refugees.

Marketing agencies and integration partners can now manage multiple client accounts from a single dashboard.

Partner Dashboard at /partner — aggregate view of all client accounts with health scores, recipe/workflow counts, and total runs. Quick-switch to any client’s console.

3 partner tiers:

• Starter (Free) — 3 clients, 1 template, 80/20 revenue share
• Professional ($99/mo) — 15 clients, 10 templates, priority support, 85/15 revenue share
• Enterprise (Custom) — unlimited clients, dedicated support, 90/10 revenue share

Client provisioning — create new client accounts with automatic pack installation and owner invitation, directly from the partner dashboard.

Template marketplace — publish custom recipe/workflow packs as templates for other partners and customers to install.

Certification hub — 5 training modules (Platform Overview, Integration Setup, Recipe Customization, Workflow Design, Client Onboarding) with certification tracking.

Revenue tracking — dashboard showing total/monthly services revenue, platform fee breakdown, and per-client revenue attribution.

Marketing page at /partners with tier comparison, benefits, and registration flow.

The Cmd+K command palette has been completely overhauled from 7 actions to 55 quick actions — making it the fastest way to do anything in the console.

New actions include: create schedule, trigger, batch run, chat agent, document, knowledge base, bundle, bakeoff, glossary term, prompt fragment, brand voice, group, collection — plus team management (invite, API keys, billing), monitoring (approvals, audit log, analytics, ROI), integrations (connect LINE/FB/IG/WhatsApp, social inbox, install pack), and compliance (governance, SOC 2, EU AI Act).

Fuzzy alias matching: Type naturally — “invite”, “subscription”, “dark mode” — and the right action appears, even without exact pattern matching.

Entity secondary actions: Press → on any recipe or workflow result to expand inline actions (Run, Edit, Schedule, Batch Run) without leaving the palette.

Page-aware suggestions: Open Cmd+K on a recipe page and see “Run this recipe”, “Edit”, “Schedule” at the top. Context-sensitive actions for 10+ page types.

Recent actions memory: Your last 10 spotlight actions are remembered and shown when the palette opens.

Entity preview panel: Select a recipe or workflow result to see a live preview on the right — description, department, input/output fields, and quick action buttons.

Operator commands: clear cache, switch to dark/light, copy account id, keyboard shortcuts, what's new.

All 137 spotlight keys are translated across all 7 locales.

A guided setup wizard for Facebook and Instagram content publishing — go from zero to publishing in under 5 minutes.

The wizard at /deploy/integrations/social-content walks you through:

1. Connect Meta accounts — enter Page ID, Access Token, Instagram User ID, App Secret with a “Test Connection” button for instant credential validation
2. Brand Voice setup — create a knowledge base with your tone, style, and content themes
3. Content preferences — select platforms (FB/IG/both), posting frequency, default tone, and target audience
4. Auto-install — installs the Social Content Pack (14 recipes, 6 workflows), creates a Content Manager agent, and sets up the Brand Voice knowledge base
5. Verify — guided test to create your first AI-generated post

The wizard uses a new outbound mode for the IntegrationWizard — no webhook needed since content publishing is outbound-only. Skips event trigger creation and uses a custom agent system prompt optimized for content generation.

Three comparison pages updated with the latest competitive intelligence:

New: JieGou vs LangSmith Fleet — LangChain’s new governance product for managing LangGraph agents at scale. Fleet governs what your engineers build. JieGou governs what your departments run. 24 key differences, 9 security comparison rows, and 6 FAQ covering the buyer persona difference (developer vs. department lead).

Updated: JieGou vs n8n — n8n’s security crisis has deepened. Now includes 6+ critical CVEs, total credential compromise (encryption key decryption), v1.x end-of-life with forced migration to v2.0, CISA KEV active exploitation listing, and the FCEB March 25 patch deadline. If you’re evaluating n8n, check the comparison.

Updated: JieGou vs Make — Make launched visual AI Agents and the Maia natural-language interface. We’ve updated the comparison with Make’s multi-model support (GPT-4, Claude, Gemini, Grok), 3,000+ app ecosystem, and credit-based pricing — plus why JieGou’s 10-layer governance still sets it apart.

JieGou now integrates with OpenClaw autonomous agents — bridging ungoverned agent execution with enterprise governance.

JieGou MCP Server for OpenClaw — 8 tools that let OpenClaw agents invoke JieGou’s governed workflows:

• list_recipes, run_recipe, run_workflow, get_run_status, search_knowledge, get_analytics, create_schedule, publish_social_post

Every invocation runs through JieGou’s full governance stack — RBAC, approval gates, audit trails, and token budgets.

OpenClaw Channel Adapter — OpenClaw agents can now route conversations through JieGou chat agents via a webhook endpoint with HMAC-SHA256 verification. This makes OpenClaw the 13th messaging channel alongside LINE, Instagram, WhatsApp, Slack, and others.

OpenClaw-Ready Badges — Recipes compatible with autonomous agent invocation are marked with an orange badge. A new filter on the recipe list page shows only OpenClaw-ready recipes.

Positioning: OpenClaw is the agent. JieGou is the governance layer. Together, they give autonomous agents structured workflows with enterprise controls.

Explore the integration at /solutions/openclaw-integration.

Two improvements to JieGou’s 12-channel chat agent system:

Embedded FAQ mode — Chat agents can now use full FAQ documents as LLM context instead of relying solely on embedding search. Three modes available:

• Rules (default) — keyword/pattern matching
• Embedded — inject the entire FAQ document for comprehensive answers
• Hybrid — try rules first, fall back to full-document context

This is ideal for teams with structured knowledge bases where every answer matters — support teams, HR onboarding bots, and internal help desks.

DM vs. group chat thread isolation — Messages in group chats and direct messages are now tracked in separate conversation threads. Previously, group chat context could bleed into DM responses. Each thread is now scoped by conversation type, so your bot’s DM replies stay personal and group replies stay contextual.

JieGou now includes a full social media content automation suite for Facebook and Instagram — from drafting to publishing to analytics.

17 new MCP tools:

• Facebook — publish text, photo, video, and link posts; schedule posts; get insights; list posts; reply to comments
• Instagram — publish photos, videos, reels, and carousels; schedule posts; get insights; list media; reply to comments; search hashtags

14 recipes in the new Social Content pack:

• Post Writer, Caption Generator, Content Repurposer, Carousel Planner, Hashtag Researcher, Performance Analyzer, Weekly Planner, Comment Responder, A/B Content Evaluator, Blog-to-Social Converter, Comment Moderator, Reel & Story Planner, Cross-Platform Analytics Reporter, Competitor Content Analyzer

6 workflows:

• Content Publish Pipeline (with approval gates), Weekly Performance Review, A/B Test & Publish, Auto-Publish from Blog, Comment Moderation Pipeline, Cross-Platform Analytics Report

Content Calendar — a new week/month view at /deploy/content-calendar with platform color coding, status badges, post preview, and drag-to-reschedule for drafts and scheduled posts.

All recipes and pack descriptions are translated across all 7 locales.

Google sign-in now works reliably on iOS Safari and Android Chrome. Previously, mobile users could encounter blank screens or failed redirects when logging in with Google.

The fix migrates to Google Identity Services (GSI) for mobile browsers, which handles the popup/redirect flow more reliably than the previous Firebase auth approach. Microsoft and GitHub login buttons are hidden on mobile where popup-based auth isn’t supported — Google remains the primary mobile sign-in method.

If you previously had trouble logging in on your phone, try again — it should work now.

With Microsoft’s Agent 365 (E7 license) approaching GA on April 1, we’ve published a detailed comparison page at jiegou.ai/vs/agent-365 — available in all 7 locales.

Key comparisons:

• Pricing — Agent 365 costs $99/user/month. For a 50-person team, that’s $4,950/month. JieGou Enterprise is $499/month total.
• Integrations — Agent 365 works inside Microsoft 365 only. JieGou connects to 250+ tools including HubSpot, Shopify, Slack, LINE, Salesforce, and Zendesk.
• Governance — JieGou’s 10-layer governance stack covers EU AI Act, NIST AI RMF, and ISO 42001. Agent 365’s governance is limited to M365 tenant boundaries.

If your team runs on HubSpot, Shopify, Slack, or any non-Microsoft stack, Agent 365 doesn’t reach your tools — but it still costs $99/user.

Our bug bounty program now has a dedicated page at jiegou.ai/security/bug-bounty — available in all 7 locales.

The page outlines our responsible disclosure process, scope, and how security researchers can report vulnerabilities. This joins our existing security trust signals: SOC 2 Type II audit in progress, Gartner Peer Insights listing, and continuous compliance monitoring via Vanta.

The Enterprise plan is now fully self-serve — sign up, enter your payment info, and start using Enterprise features immediately. No “Contact Sales” required.

What’s included at $499/month (flat, not per-user):

• Unlimited users
• 250+ tool integrations
• 10-layer governance stack
• Priority support
• All department packs

Promotional codes (like early-adopter discounts) work seamlessly with the new self-serve flow — credit card is captured upfront, and billing starts automatically when the promotional period ends.

JieGou now ships with complete internationalization across 7 locales — English, Traditional Chinese (zh-TW), Simplified Chinese (zh-CN), Japanese, German, French, and Spanish.

What changed:

• 425+ hardcoded strings replaced with Paraglide i18n keys across the entire console
• Locale-aware packs, recipes, and workflows — zh-TW accounts see native Chinese template names, descriptions, and instructions
• Team activity, notifications, and chat agent pages fully localized
• Inbox items resolve titles and descriptions via i18n keys with nested parameter support

This isn’t just UI translation — the AI templates themselves adapt to your account’s locale. When a zh-TW user installs the Marketing Starter Pack, every recipe name, description, and field label appears in Traditional Chinese.

JieGou’s LINE integration now supports group chats with intelligent mention-only behavior.

New capabilities:

• Mention-only default — in group chats, the bot responds only when @mentioned, avoiding noise in busy channels
• Typing indicator — a loading animation shows while the AI processes your message, both in 1:1 and group chats
• Styled Flex Messages — responses are wrapped in branded Flex Message bubbles with colored headers and action buttons
• Per-message run history — every AI response is tracked as a first-class run with full audit trail

These improvements apply to both the chat agent and workflow trigger paths, enabling enterprise LINE use cases like team support channels and client group chats.

We completed a comprehensive dependency vulnerability sweep across all 7 sub-projects in the JieGou monorepo, resolving every Dependabot alert:

• 2 critical vulnerabilities resolved
• 65 high severity issues patched
• 66 medium severity issues addressed
• 29 low severity items cleaned up

All fixes have been deployed to production. Combined with our existing security measures — Semgrep SAST on every PR, Trivy container scanning, and continuous monitoring via Vanta — this sweep ensures JieGou meets the dependency management standards required for SOC 2 Type II.

All 10 findings from our SOC 2 penetration test have been resolved. This milestone clears the path to our Type I report. Combined with Vanta continuous monitoring, 412 policies tracked across 17 compliance domains, and 24,000+ automated tests, JieGou’s security posture meets enterprise requirements. SOC 2 Type I audit is now in progress.

Verification emails now come from noreply@jiegou.ai with full JieGou branding — logo, styled layout, and clear call-to-action — instead of the generic Firebase default. Emails are sent via Resend with proper SPF, DKIM, and DMARC authentication, so they land in your inbox instead of spam. After clicking the verification link, you’re automatically signed in and redirected to the console — no more re-login required.

The marketing site now features dedicated pages for each of our 5 industry packs: Healthcare, Financial Services, Legal, Retail, and Manufacturing. Each page details the industry-specific recipes, compliance considerations, and workflow templates included in the pack — making it easier for teams in regulated or specialized industries to see exactly how JieGou fits their needs before signing up.

New users now get an onboarding experience tailored to their RBAC role. Owners see account setup and team management. Admins get governance configuration and integration setup. Managers are guided through department packs and approval workflows. Editors jump straight into recipe and workflow creation. Viewers get a quick tour of dashboards and run history. Each flow highlights the features most relevant to the user’s permissions, reducing time-to-value and eliminating confusion from seeing controls they can’t access.

The site-wide search modal (⌘K / Ctrl+K) has been completely rewritten. The previous implementation had a bug where the search input would disappear after the first use. The new version uses the Pagefind JS API directly with a custom UI, ensuring the search input is always visible and results appear instantly. Keyboard shortcuts, backdrop dismiss, and Escape to close all work reliably.

The new Agent Activity Feed on the chat agents list page shows per-agent operational metrics pulled from live conversation data: conversations today, escalations this week, and last message timestamp. Combined with the Governance Summary Bar (weekly actions, escalations, violations, cost), teams get a single-screen view of both operational health and compliance posture across their entire agent fleet — no separate analytics tool required.

Every chat agent now displays a real-time Governance Score (0-100) computed from 8 weighted factors: audit logging enabled (15%), escalation policy configured (20%), knowledge base attached (15%), rule table present (15%), RBAC roles assigned (10%), cost controls set (10%), memory/context limits configured (10%), and BYOK encryption active (5%). The score updates live as you configure each layer, giving teams immediate visibility into governance posture without waiting for quarterly audits. A companion Governance Summary Bar on the dashboard aggregates weekly actions, escalations, violations, and cost across all agents.

Create a production-ready chat agent in under 2 minutes. The new quick-create flow asks for just three fields — name, description, and channel — then auto-configures the provider, model, and governance settings using your account’s BYOK keys. Choose from 10 industry templates (E-Commerce Support, Healthcare FAQ, SaaS Onboarding, Restaurant Booking, Real Estate, Travel, Education, Finance, Legal, and HR) to start with pre-built system prompts, rule tables, and escalation policies tuned for each vertical. Templates apply smart defaults while remaining fully customizable.

Four new threat detectors protect production AI agents: prompt injection detection identifies adversarial input manipulation, data exfiltration detection catches attempts to extract sensitive data, privilege escalation detection monitors permission boundary violations, and resource abuse detection flags anomalous consumption patterns. All detectors run inline via execution hooks — threats are caught during execution, not after. Backed by 56 test cases covering adversarial inputs.

New AI Marketing Automation solution page targeting marketing teams. Features 6 pre-built marketing recipe templates covering content generation, campaign optimization, audience segmentation, social media automation, email personalization, and competitive analysis. Includes a dedicated Marketing department pack configuration.

Recipes and workflows can now specify a default LLM model that pre-populates the run page model dropdown. Creators set the preferred provider and model in the editor, and runners see it pre-selected while retaining the ability to override at runtime. Workflow-level defaults take precedence over recipe-level defaults for workflow runs.

JieGou is now available on AWS Marketplace, Azure Marketplace, and GCP Marketplace. Deploy via Helm chart with marketplace-specific values files for consolidated billing, procurement compliance, and simplified infrastructure provisioning. A new Cloud Marketplace landing page guides buyers through the deployment process.

New Chat Agents feature provides a 4-tier hybrid resolution cascade for messaging automation. Incoming messages are resolved through: (1) rule table matching using embedding-based cosine similarity against CSV-imported rules, (2) knowledge base retrieval via RAG, (3) LLM fallback with system prompt and conversation context, (4) human escalation when confidence is low. Supports all messaging channels (LINE, Instagram, WhatsApp, Facebook, YouTube) with conversation threading, thread compaction, and analytics. Manage agents, import rules via CSV, and test resolution interactively from the console.

Added a Market Validation Banner component displaying third-party survey results across the homepage and comparison pages. Key stats from enterprise AI adoption surveys reinforce JieGou’s governance-first positioning with real market data. The banner appears contextually on pages where social proof is most impactful.

New PHI detection module identifies protected health information in AI agent inputs and outputs. Detects Medical Record Numbers (MRN), National Provider Identifiers (NPI), ICD-10 diagnosis codes, health plan identifiers, and medical context phrases. Supports full redaction and partial masking modes. Integrates with JieGou’s sensitivity labels and data classification system. Validated with 32 test cases across all PHI categories.

Two usability improvements to the workflow run page: inline output display shows the workflow result directly after completion without requiring navigation to the run detail page, and a knowledge base dropdown automatically renders for workflow input fields with format: 'knowledge-base', replacing free-text input with a picker populated from the user’s available knowledge bases.

New Scale AI Adoption solution page with adoption gap analysis: 57% of enterprises have AI agents in production, but only 6% have scaled. Landing page includes market data from Gartner, G2, CrewAI, and IDC, along with a before/after comparison table, 4-step scaling guide, governance metrics, and ROI analysis. Two new blog posts: “The AI Agent Adoption Gap: 57% Have Agents, 6% Have Scaled” and “What’s the Cost of 15 Ungoverned AI Agents?” — analyzing the market data and making the case for governed scaling. External survey validation from CrewAI’s 2026 report (34% cite governance as #1 priority) prominently featured across marketing. New reusable MarketValidationBanner component for citing analyst data. Updated homepage with adoption gap framing. Added Scale AI Adoption and AI Marketing Automation to Solutions navigation.

New APAC Messaging Automation landing page targeting businesses operating in Taiwan, Japan, Thailand, Southeast Asia, and India. Highlights LINE, WhatsApp, and Facebook Messenger dominance by region, with deployment guides for each market. Added to mega menu and footer navigation under Solutions.

The Omnichannel Messaging page now includes a “Revenue-Generating AI” section demonstrating how cross-app orchestration turns support conversations into revenue opportunities. AI agents can identify upsell signals, trigger product recommendations, and hand off to sales workflows — all within the same governed messaging pipeline.

The Operations Hub now surfaces automated execution insights — pattern-based anomaly detection that continuously analyzes your recipe and workflow runs. Four detectors run on every refresh: failure pattern detection flags recipes with >20% failure rates, cost spike detection catches >50% token usage increases, latency anomaly detection identifies runs exceeding 2x the p95 baseline, and error clustering groups 3+ similar errors for batch triage. Insights are severity-ranked (critical, warning, info) and include actionable recommendations. Available at /operations/landscape with a configurable 7/30/90-day time range.

A new template health API provides quality scoring and validation for every department pack template. The /api/template-health/summary endpoint returns health metrics including template coverage, freshness, and prompt quality indicators. A dedicated Department Pack Quality page visualizes these metrics and surfaces templates that need attention.

Approvers can now approve or reject workflow steps directly from their email inbox — no console login required. When a workflow hits an approval gate, a notification email is sent with one-click action links using signed, time-limited tokens. The workflow automatically resumes after the approver responds. Full audit trail tracks who approved, when, and from what source.

Renamed “Trust Escalation” to “Graduated Autonomy” throughout the console to better communicate how AI agents earn increasing independence. Updated the trust escalation banner, governance configuration panel, NL input component, action center, and all related type definitions. Added a dedicated Graduated Autonomy feature page to the marketing site.

New MCP Governance feature page details JieGou’s multi-layer MCP server governance: certification pipeline with automated security scanning, permission boundary enforcement, version pinning, and immutable audit trails. Comparison data updated across customer support and competitor pages to highlight governance gaps in alternative platforms.

Create, edit, and manage message templates with full CRUD at /message-templates — filter by category and channel, and detect template variables automatically. In the social inbox reply composer, a new template quick-insert dropdown lets agents search and insert templates filtered by the current channel. Pair it with the AI-suggested reply button, which generates a context-aware draft via Claude Haiku based on conversation history. Templates and AI suggestions work together: start with a template, let AI personalize it, then send.

Missions are now a first-class workflow primitive. The new Plan with AI mode in the Mission modal lets you describe a goal in natural language — the AI generates a structured plan of task cards that you can preview, edit, and create in bulk. Active missions appear on the Inbox dashboard in a dedicated two-column grid for at-a-glance progress tracking. A redesigned /missions page features a hero section, dual CTAs, and feature cards highlighting AI planning, auto-organization, and progress tracking. Includes a 5-step guided tour and a “Create your first mission” onboarding milestone.

JieGou is doubling down on self-serve deployment. New “JieGou vs. Hiring a Consultant” comparison page with a full TCO breakdown — time to deploy, cost, vendor lock-in, customization, compliance, and messaging channels. Updated homepage messaging emphasizes instant deployment: “Deploy in minutes, not months.” Added “Deploy in minutes” taglines across feature pages, comparison pages, and solution pages. New blog post: “Deploy AI Workflows Today, Not After a 6-Month Consulting Engagement.” Self-serve governed AI for teams that move fast — 20 department packs, 12 messaging channels, 9 LLM providers, 412 compliance policies. No consultants required.

JieGou’s SOC 2 Type II audit (Security TSC) is officially underway. Vanta engagement signed March 3, 2026. With 412 compliance policies already written, 17 Trust Services Criteria controls mapped, and evidence export infrastructure in place, the audit process is backed by comprehensive preparation. Trust Center with custom domain and one-click NDAs included. Target certification: Q3 2026.

JieGou now supports 12 messaging channels — LINE, Instagram DMs, Facebook Messenger, WhatsApp, YouTube Live Chat, Slack, Microsoft Teams, Discord, Gmail, Outlook, Intercom, and browser widget — all connected to a single platform.

New capabilities:

• Unified Inbox — every message from every channel in one chronological view, filterable by channel, priority, sentiment, and status
• Cross-Platform AI Recipes — write one recipe, deploy it across all 12 channels. The same FAQ auto-responder works on Instagram DMs, WhatsApp, LINE, and Messenger
• Cross-Channel Analytics — compare response times, resolution rates, and customer satisfaction across all channels from the Operations Hub
• Setup Wizards — guided 7-step setup for Instagram, Facebook, WhatsApp, YouTube, and LINE with one-click recipe installation

New integrations added this release: Instagram DMs (8 MCP tools), Facebook Messenger (10 MCP tools), WhatsApp Business API (8 MCP tools), YouTube Live Chat (6 MCP tools).

Enterprise governance — approval gates, PII detection, confidence thresholds, audit trails, and trust escalation — is applied uniformly across all 12 channels.

Full Facebook Messenger integration via the Meta platform. Inbound webhooks handle message events, postbacks, and Quick Reply callbacks with token-based verification. AI workflows triage conversations, generate context-aware responses, and use Quick Replies for structured choices and Sender Actions (typing indicators) for natural conversation flow. Ten outbound MCP tools cover text, templates, Quick Replies, Sender Actions, Persistent Menu management, and user profile retrieval. Shares Meta platform authentication with Instagram — connect once and automate both channels. Includes pre-built recipes and a Facebook Messenger Support pack.

RAG retrieval now uses Firestore-native vector search with a hybrid fallback pipeline — no Pinecone, Weaviate, or other external vector database required. Documents are embedded via OpenAI text-embedding-3-small and stored directly in Firestore with FieldValue.vector(). Queries hit vector similarity search first, falling back to brute-force + Redis caching for edge cases. Performance improved 16× on large knowledge bases: cold queries across 705 documents dropped from ~160s to ~10s, and warm queries return in under 1 second via per-document Redis caching with 10-minute TTL. Content extraction was upgraded from regex-based HTML stripping to Readability + Turndown for significantly cleaner markdown output. Parallel document fetching with field projection further reduces latency on batch operations.

Full Instagram DM integration via the Meta platform. Inbound webhooks receive Instagram messages with HMAC-SHA256 signature verification, event type mapping, and media attachment support. AI workflows automatically triage messages by priority, category, and sentiment, draft contextual responses using your knowledge base, and escalate complex conversations to the right team member. Eight outbound MCP tools let recipes and workflows send text, media, templates, and retrieve user profiles directly in Instagram conversations. Includes pre-built recipes (DM Triage, Response Drafter, FAQ Auto-Responder, Escalation Notifier) and a one-click Instagram Support pack for the Support department.

A unified social inbox and analytics layer that works across all messaging channels — LINE, Instagram, Facebook Messenger, WhatsApp, and YouTube. The social inbox consolidates conversations from every channel into a single view with cross-platform search and filtering. The analytics dashboard shows per-channel message volume, response times, resolution rates, and AI confidence trends. Shared recipes (Cross-Channel Triage, Universal Response Drafter) work identically across all platforms using a normalized event model. Integration setup wizards guide you through connecting each platform with step-by-step credential configuration, webhook verification, and connection testing — all from a single Integrations panel in the console.

WhatsApp Business API integration for enterprise customer communication. Inbound webhooks handle text messages, media, location, contacts, and interactive button/list replies with Meta webhook signature verification. AI workflows process conversations through triage, template message selection, and free-form response drafting with knowledge base grounding. MCP tools send text, media, templates, interactive messages (buttons and lists), and retrieve business profile information. Supports the 24-hour messaging window policy with automatic template message fallback. Shares Meta platform authentication with Instagram and Facebook Messenger for unified multi-channel governance.

Monitor and automate YouTube Live Chat with AI-powered workflows. JieGou connects to the YouTube Data API to poll live chat messages during active streams, routing them through triage, moderation, and response pipelines. AI classifies chat messages (questions, spam, engagement, moderation flags) and can auto-respond to FAQs or escalate flagged messages. MCP tools let workflows send chat messages, ban users, and retrieve stream metadata. Designed for creators and brands running live events who need scalable chat moderation without manual effort.

Point JieGou at any website and it becomes a searchable AI knowledge base in minutes. The six-phase pipeline handles everything: sitemap discovery finds all indexable pages, smart filtering lets you exclude paths and set depth limits, parallel crawling processes pages concurrently with configurable limits, opt-in headless Chromium renders JavaScript SPAs before extraction, Readability-based content extraction strips navigation and boilerplate for clean markdown, and incremental refresh re-crawls only changed pages using content hashes. A pre-crawl estimation shows exact page counts and processing time before you commit. Plan-tiered limits scale from 100 pages on Starter to unlimited on Enterprise, with daily or hourly crawl frequencies.

Workflows can now include a Coding Agent step that autonomously reads, writes, edits files, and runs shell commands. Point it at a Git repo and a task description — the agent clones the repo, explores the codebase, implements changes, and runs tests. Six built-in tools (read, write, edit, bash, glob, grep) operate inside a Docker sandbox with no network access, memory limits, and path confinement. The agent uses iterative compaction for long sessions, streams real-time progress events, and reports all modified files. Available on Pro plans and above.

Long conversations no longer fail at context limits. JieGou now automatically detects when a conversation approaches the model’s context window and compresses older messages into a structured summary — preserving goals, progress, key decisions, referenced documents, and next steps. Subsequent compactions update the existing summary incrementally rather than regenerating from scratch, so accumulated context is never lost. Works with all providers and models, including BYOK keys.

JieGou’s competitive differentiation against general-purpose agent platforms. 12 enterprise knowledge sources span 4 categories (Enterprise Search: Coveo/Elasticsearch/Algolia/Glean, Workspace Knowledge: Confluence/Notion, Customer Intelligence: Zendesk/Guru, AI-Native: Pinecone/Vectara/Google Drive/OneDrive). BYOM with AI Bakeoffs prove which of 9 providers works best before production. 10-layer governance stack with 4-level trust escalation, PII detection, brand voice compliance, and approval gates. 20 department packs with 250+ tested recipes enable day-one automation. 13,320+ tests at 99.1% line coverage ensure enterprise reliability. Full air-gapped deployment with self-hosted models addresses the most security-conscious enterprises.

Define custom tools backed by your own HTTP endpoints. Specify a name, description, JSON input schema, and webhook URL — JieGou makes your tool available in recipes, workflows, and the conversational agent. Custom tools support request templates (Handlebars), response mapping (JSONPath), custom headers (encrypted), and department scoping. Test tools directly from the editor with sample inputs. Endpoints must use HTTPS in production, and all outbound requests go through the platform’s timeout, retry, and rate limiting infrastructure. Available on Team plans and above.

JieGou now supports 9 LLM providers out of the box. In addition to Anthropic, OpenAI, Google, and OpenAI-compatible endpoints, you can now connect Mistral AI (Mistral Large, Medium, Small), Groq (ultra-fast inference), xAI (Grok), AWS Bedrock (Converse API with SigV4 signing), and Azure OpenAI (deployment-based endpoints). Each provider includes full BYOK support, per-provider circuit breakers, key validation, and per-step model selection. Choose the fastest, cheapest, or most capable model for each step in your workflow.

Define custom workflow step types as TypeScript modules that run in V8 isolates — fully sandboxed with memory limits, CPU time constraints, and no filesystem or network access. Custom steps can use injected platform APIs (HTTP fetch, LLM invocation, read-only Firestore) to build bespoke logic. Edit code in a built-in Monaco editor, define input/output schemas, and test with sample data before deploying. Custom steps integrate seamlessly with the workflow executor, DAG orchestration, and cost estimation. Available on Enterprise plans.

Register webhook-based hooks that fire on workflow and recipe execution events: workflow:start, workflow:complete, workflow:error, step:start, step:complete, step:error, recipe:start, recipe:complete, approval:requested, approval:granted, approval:rejected. Non-blocking hooks are fire-and-forget for notifications and logging. Blocking hooks can modify execution (with a 10-second timeout and circuit breaker protection). Use hooks to integrate with Slack, PagerDuty, custom dashboards, or any HTTP endpoint. Available on Team plans and above.

Full LINE Messaging API integration for customer support automation. Inbound webhooks receive LINE messages and route them through AI-powered triage, response drafting, FAQ auto-response, and escalation workflows. Outbound MCP tools (reply_message, push_text_message, push_flex_message, get_profile) let recipes and workflows respond directly in LINE conversations with both plain text and rich Flex Message cards. Includes 4 LINE-specific recipes (Message Triage, Response Drafter, FAQ Auto-Responder, Escalation Notifier), 2 workflows (LINE Support Pipeline, LINE FAQ Pipeline), and a one-click LINE Support Starter pack for the Support department. Designed for APAC teams already using LINE as their primary customer channel.

Comprehensive production hardening across all recently shipped features. Compaction now emits Prometheus metrics (compaction rate, tokens saved, duration). Coding Agent steps enforce Docker resource limits (memory caps, PID limits, read-only root filesystem, no network), log every tool call to the audit trail, and enforce output size guards (500KB total, 100KB per tool, 50-file limit). Extended providers get per-provider circuit breaker tuning and normalized error parsing. Skills have content size limits (50KB), Redis caching, and token budget awareness. SDK API includes per-key rate limiting (60 req/min, 1000 req/hr) and separate usage billing.

Execute recipes, workflows, and coding agents programmatically via the new SDK API. Create scoped API keys with granular permissions (read, execute, per-resource), then call POST /api/sdk with your bearer token. Supports synchronous execution for recipes, async polling for workflows, and SSE streaming for real-time progress. Designed for CI/CD pipelines, Slack bots, and external integrations. API keys use SHA-256 hashing with a jg_ prefix for easy identification. Includes per-key rate limiting and separate usage tracking. Available on Pro plans and above.

Evaluate any recipe’s output quality on demand with the new Quality Scorecard panel. An LLM judge scores recent outputs across five weighted dimensions — schema compliance, completeness, actionability, format adherence, and consistency — then returns a promote, revise, or reject recommendation. Results appear directly on the recipe detail page so you can iterate on prompt quality without leaving your workflow. Each evaluation is audit-logged for traceability.

Fork any conversation at any message to explore alternative directions without losing the original thread. Click the branch icon on any assistant message to create a new branch. Switch between branches instantly, rename them for clarity, and compare different approaches side by side. Compaction is branch-aware — each branch maintains its own summary while sharing context up to the fork point. Available on Pro plans and above.

Test any recipe directly from its detail page with a single click. Test My Recipe generates synthetic inputs via AI, executes the recipe against each one sequentially, and streams real-time progress via NDJSON so you see results as they arrive — not after a long wait. A results modal shows summary stats (pass rate, average latency, token usage) and per-test accordions with full input/output inspection. Every test run is logged to the audit trail as a recipe.tested action for governance compliance.

Create skills — structured markdown documents that your conversational AI agent can load on demand. Unlike RAG-retrieved chunks, skills are loaded in full and provide complete, authoritative instructions for specific tasks (deployment procedures, data taxonomies, incident response playbooks). Skills are department-scoped, searchable by name, and can be invoked with /skill:name in chat or discovered automatically by the agent via built-in list_skills and load_skill tools. Available on Pro plans and above.

Run open-source models alongside Claude, GPT-5, and Gemini — from the same platform. Four models are now JieGou Certified with tested tool calling and structured output on vLLM: Llama 4 Maverick (400B+ MoE, 1M context, vision), DeepSeek V3.2 (671B MoE, 128K), Qwen 3 235B (multilingual MoE, 128K), and Mistral 3 Large (123B, vision). JieGou automatically discovers local Ollama and vLLM servers and lists available models. Connect any OpenAI-compatible endpoint — cloud-hosted or self-hosted — via the new custom base URL configuration. Community models work too, with a clear certified vs. community tier distinction.

Connect 12 external knowledge sources to ground every AI response in real company context. Supported providers span enterprise search (Coveo, Elasticsearch, Algolia, Glean), vector databases (Pinecone, Vectara), team wikis (Confluence, Notion), file storage (Google Drive, OneDrive/SharePoint), and customer intelligence (Zendesk, Guru). Each connector exposes MCP-native tools for search, retrieval, and document access — no custom code required. Credentials are encrypted with AES-256-GCM, and every connection includes health checks, rate limiting, and a full audit trail.

JieGou now supports multiple languages across the entire platform using Paraglide i18n infrastructure. All pages, components, forms, modals, and backend systems have been translated, including server-side strings, LLM responses, navigation, and email notifications. Users can switch languages via the language selector, and the interface will adapt accordingly while maintaining their session and preferences.

Workflows now support DAG execution mode with four new step types: LLM (direct prompts), Eval (quality gates), Router (input classification), and Aggregator (output combination). Convergence loops enable iterative refinement — an Eval step can re-trigger upstream steps until quality thresholds are met. Enhanced checkpointing (v2) enables crash recovery from any point in both sequential and DAG modes. Available on Enterprise plans.

The collaboration suite is now complete. This release adds a platform-wide team chat drawer with search, follow mode for co-browsing (see a teammate’s cursor in real time), typing indicators, and session recording with replay. Available on Pro and Enterprise plans.

Compare recipes, models, and entire workflows head-to-head with the new bakeoff system. Features LLM-as-judge evaluation with statistical confidence, multi-judge consensus scoring, synthetic input generation, and live A/B test routing with auto-stop. Available on Pro (recipe and model bakeoffs) and Enterprise (workflow bakeoffs and A/B routing).

Workflows can now invoke other workflows as steps using SubWorkflowStep, providing composable multi-stage orchestration from a single canvas. Four pre-built pattern templates (Critic/Refiner, Specialist Router, Debate/Consensus, Plan/Execute/Verify) help you get started with proven multi-agent orchestration patterns. The former standalone Pipeline concept has been merged into Workflows. Available on Enterprise plans.

Share your screen with teammates directly from the console. Powered by Daily.co, screen sharing sessions can be started with one click and shared via invite link. Available on Enterprise plans.

Chat panels are now available directly on recipe runs, workflow runs, and recipe detail pages. @mention teammates to notify them, search message history, and keep conversations attached to the work they reference.

See who’s online and where they are in the console. Presence indicators appear on team members throughout the app, showing online, idle, and offline status. Route-level tracking lets you see when teammates are viewing the same recipe or workflow run.

A dedicated environment for crafting and optimizing prompts. Includes full version tracking with diff comparison, live token budget visualization per model, a variable inspector that detects all placeholders, few-shot example curation with drag-and-drop ordering, and an AI-powered optimizer that suggests prompt improvements. Available on Pro plans.

New guided onboarding and contextual education system throughout the console. Feature tips, interactive walkthroughs, and contextual help appear where you need them — when building recipes, configuring workflows, setting up knowledge bases, and more. Dismiss tips individually or complete guided tours to unlock advanced features.

Run any recipe across an entire data table in one click. Upload a CSV or paste data directly, then execute your recipe across every row with configurable concurrency. Track progress in real time, filter results by status, and export outputs as CSV or JSON.

Define your brand voice with tone, formality, glossary terms, and reusable prompt fragments. Every AI output across recipes and workflows respects your guidelines automatically. Governance features include an audit log tracking 30+ action types, multi-approver policies with escalation, and RBAC with department scoping.

Build knowledge bases by uploading documents and importing URLs. JieGou automatically chunks, embeds, and indexes your content. When you run a recipe or workflow, relevant context is retrieved via RAG and injected into the prompt — so AI has real company knowledge without manual copy-paste. Organize documents into groups and scope which knowledge bases apply to specific recipes or workflows.

Workflows now support parallel steps. Run multiple branches concurrently using Promise.allSettled() — each branch executes sequentially while all branches run in parallel. Output isolation ensures branches don’t interfere with each other.

A single inbox for all pending actions across your account. Aggregates approval requests, review tasks, and quick actions into one dashboard. Includes digest notifications so you never miss a pending item, and quick-action buttons to approve, reject, or reassign without navigating to individual workflow runs.

Four new event-driven trigger types complement existing cron schedules and webhooks. Run completion chaining triggers a workflow when another finishes. Email and Slack source triggers react to incoming messages. Connector data change detection fires when data changes in your connected tools. Browser event monitoring triggers from browser extension events.

Every workflow edit now creates an immutable version snapshot. Compare any two versions with a visual diff engine that highlights step-level changes. When deploying updates, use canary rollouts to route a percentage of traffic to the new version before promoting it to production. One-click rollback to any previous version.

Install curated recipe and workflow packs for Sales, Marketing, Support, HR, Finance, Operations, Legal, Engineering, and Executive teams. Each pack includes 7-10 recipes, 2-5 workflows, suggested schedules, and integration recommendations.

Trigger recipe and workflow executions via HTTP webhooks. Configure token-based authentication, payload mapping, and input resolution. Full run history tracking for every trigger.

Attach a Quality Guard to any recipe for automatic LLM-as-judge evaluation on every production run. Configure quality criteria and thresholds, track quality trends over time with charts, and receive alerts when output quality drifts below your baseline. Quality Guard results appear directly on the recipe detail page.

Recipe steps now retry transient errors (429, 503, 502, timeouts) with exponential backoff — base 2 seconds, max 30 seconds with jitter. Permanent errors skip retries immediately. Configurable up to 5 retries per step.