Changelog
New features, improvements, and fixes shipped to Fruxon.
May 25, 2026
v0.24.0
v0.24.0
May 25, 2026
Agent Network & Secrets Vault
A major release introducing a fully-redesigned Agent Network with participants, triggers, and channel management; a tenant-wide secrets vault with placeholder resolution; and a live draft editing experience so changes sync instantly across browser tabs and the CLI.
New
Agent Network tab lets you define Participants (the people or systems your agent talks to), configure inbound Triggers from integrations like Slack or Telegram, and map payload fields to agent parameters — all from one place.
New
Tenant Secrets Vault lets you store sensitive credentials centrally, reference them anywhere with {{secret.X}} placeholders, and track every access in a credential audit log.
New
Agent Studio now autosaves a live draft as you edit, with server-backed undo/redo and real-time sync across tabs so you never lose work.
New
Per-agent Production / Dev usage toggle lets you view execution stats, costs, and recent runs scoped to either your live or development environment.
New
Per-origin agent budgets prevent cost bleed between production and dev runs, with a dedicated budget panel in the Studio overview.
New
Integration configs are now encrypted at rest with a per-tenant key, and you can pin API tool calls to a specific egress country for data-residency requirements.
New
Per-API-key activity log with full-text search, infinite scroll, and a configurable retention window so you can audit exactly how each key is being used.
Improved
Test chats in Studio are now persisted, metered against your test budget, and sync live via server-sent events so conversations survive page reloads.
Improved
Integration config pages now include connection testers for Slack, Teams, Twilio, Dialog360, and more, so you can verify credentials before going live.
Fixed
Evaluation runs that failed during preparation or queuing now correctly roll back, preventing stuck pending entries in your evaluation history.
May 18, 2026
v0.23.0
v0.23.0
May 18, 2026
API Keys, Integrations & LLM Configs
This release brings scoped API keys with full audit history, tenant-level LLM and integration configuration management, and a redesigned agent studio experience with inline quality checks and smarter credential handling.
New
Scoped API keys let you mint keys with specific permissions (e.g. agents:test, mcp:invoke, inbound:deliver), rotate them with override options, and view a full audit history of mint, rotate, and revoke events.
New
Tenant-level LLM provider configs replace per-agent inline settings — define your model credentials once and reuse them across agents, with revision history and author attribution.
New
Integration configurations now support OAuth scope selection, application-level OAuth credentials, per-agent bindings, and sandbox routing at the credential level — plus the ability to delete custom integrations.
New
Microsoft Outlook is now available as an integration via the Microsoft Graph API.
New
Inline LLM-as-judge quality checks can now be added directly to agent steps, giving you automatic feedback loops without leaving the agent studio.
New
The CLI login flow (fruxon login) is now supported with a browser-based OAuth poll page and scope preset selection for key approval.
New
Hugging Face is now supported as a self-hosted LLM provider using the OpenAI-compatible API.
Improved
The agent studio's integration and knowledge base configuration has been simplified — tenant connections are now shown as cards with inline pinning, and LLM config is selected from a shared dropdown instead of an inline editor.
Improved
File assets now support drag-and-drop upload, vectorization toggling, LLM config-based embedding selection, and a searchable asset list with status indicators.
Fixed
User invitation emails were being compared case-sensitively, causing some users to be incorrectly blocked or duplicated — email matching is now case-insensitive throughout the invitation workflow.
May 11, 2026
v0.22.0
v0.22.0
May 11, 2026
External Agents & Observability
This release brings support for connecting and running external agents from platforms like CrewAI and LangGraph, a new observability panel for tracing, major chat experience improvements, and several data source enhancements.
New
Connect and run agents from external platforms (CrewAI, LangGraph, and others) directly within the agent studio, with parameter overrides and gateway routing.
New
New Observability panel in Agent Studio lets you set up OTLP tracing and monitor real-time ingest health for your agents.
New
Data sources now support an OAuth "Connect" flow, scheduled rescanning for poll-based auto-update, and webhook-triggered updates — making it easier to keep your knowledge base fresh.
New
Google Custom Search is now available as a built-in integration for your agents.
New
Tenant-level integration configurations and MCP server support are now available for more flexible deployment setups.
Improved
Chat now shows live inline step details as agents work, with a streaming cursor, expandable tool traces, and smoother real-time rendering.
Improved
Asset configuration panels now display embedding error categories and usage meters even when no budget is set, making it easier to diagnose issues.
Improved
Agents now persist activated skills across conversation turns, so context is maintained throughout a session.
Fixed
Tool call events are now flushed immediately in streaming responses, eliminating delays in step visibility during agent execution.
Fixed
Fixed a crash in the Prompt Editor when inserting assets, and resolved a stack overflow in Agent Studio.
May 4, 2026
v0.21.0
v0.21.0
May 4, 2026
Agent Memory, Sessions & Evaluation
This release brings persistent agent memory, smarter session management, and a significantly improved evaluation experience with deeper metrics, dataset management, and sandbox controls.
New
Agents can now store and search persistent memories across conversations, giving them context that carries over between sessions.
New
Session history can now be configured per step, including summarization and a 'last N turns' strategy to control how much context agents retain.
New
Evaluation datasets are now scoped to individual agents, making it easier to organize and manage test samples without cross-agent clutter.
New
You can now tag execution results into multiple datasets at once directly from the execution history panel.
New
Evaluation runs now display per-sample and total run duration, plus cost and runtime as informational signals alongside quality scores.
New
Mercury banking and Bright Data are now available as integrations.
New
DeepSeek and Grok models now support streaming responses.
Improved
The deploy dialog now shows the latest evaluation result for the current revision upfront, so you can review quality before deploying.
Improved
Agent steps now support a configurable maximum tool loop limit, preventing runaway tool chains.
Fixed
Sandbox mode no longer falls back to real database connections for tools that lack user-configured credentials, preventing unintended data access.
Apr 27, 2026
v0.20.0
v0.20.0
Apr 27, 2026
Evaluation, Sandbox & HITL Upgrades
This release brings a fully sandboxed evaluation workflow, deeper human-in-the-loop controls, and a new Read-Through sandbox mode — making it safer and easier to test and deploy agents with confidence.
New
Evaluations now run in sandbox mode, blocking deployment if any integration falls back to production credentials, so you can test agents without touching real systems.
New
A new Read-Through sandbox mode lets integrations fetch real data during evaluations while still preventing writes, giving you more realistic test results safely.
New
Human-in-the-loop approvals now show the outbound message and requester identity in the trace view, and correctly resume multi-step workflows across multiple approval cycles.
New
LLM tool calls are now replayed on HITL resumption, and token counts and traces are preserved correctly across pause and resume cycles.
New
GPT-5.5 and GPT-5.5 Pro models are now available, along with reasoning parameter support and default token limits across all supported models.
New
Agents you delete are now soft-deleted with a 7-day grace period, keeping the card visible with a clear deletion banner so you can recover them before they're permanently removed.
Improved
The evaluation samples table now supports sorting and tag filtering, with sticky headers and copy-to-clipboard on JSON results for easier review.
Improved
Budget validation now enforces that daily limits cannot exceed monthly, and monthly cannot exceed lifetime, preventing invalid budget configurations from being saved.
Fixed
Execution history no longer returns 404 errors when viewing incomplete runs, and outbound delivery failures are now surfaced clearly in the status indicator.
Fixed
Asset display name changes and budget exceeded email notifications are now reliably delivered, and budget notifications are correctly scoped to the current workspace admins only.
Apr 20, 2026
v0.19.0
v0.19.0
Apr 20, 2026
Human Approval, Webhooks & New Integrations
This release introduces human-in-the-loop approval workflows for tool executions, a new webhooks infrastructure, and several new integrations including GitLab, Firecrawl, and Gong. Execution history, knowledge base management, and the trace viewer all received significant improvements.
New
Human-in-the-loop (HITL) approval: agents can now pause and request human sign-off before executing a tool, with approval delivered via your connected chat channels and status visible live in the execution trace.
New
GitLab and Gong integrations are now available, joining the existing library of connected services.
New
Website knowledge source: you can now add a website as a knowledge base asset using Firecrawl-powered web scraping, with automatic updates via webhook.
New
Confluence knowledge base assets now support automatic re-ingestion when content changes via webhook.
New
Execution history can now be filtered by agent version, making it easier to compare behavior across releases.
New
Agent cost over time is now displayed as an interactive line chart with selectable time ranges, including an "All time" view.
Improved
Knowledge base document and chunk management now features infinite scroll, skeleton loaders, server-side search, and per-document delete — replacing the previous basic list view.
Improved
Organization settings and the signup flow now include a website field for your company profile.
Improved
Execution traces now poll for updates while an execution is in progress and display a live wait timer when a human approval is pending.
Fixed
Re-ingesting a knowledge base asset now correctly refreshes the operations list, and budget saving in the asset detail page no longer referenced stale data.
Apr 13, 2026
v0.18.0
v0.18.0
Apr 13, 2026
Knowledge Base & New AI Models
This release introduces a fully featured Knowledge Base (Assets) system for connecting and indexing your data sources, along with support for new AI models and providers. You also get better visibility into long-running operations and improved environment-level analytics.
New
Connect and index data sources (Google Drive, Confluence, GitHub, GitLab, Jira, Notion, Zendesk, GCS) as Knowledge Bases to give your agents access to your own content.
New
Track ingestion progress in real time with a new Operations tab showing status, progress bar, and processed item counts for each Knowledge Base.
New
DeepSeek, Cohere, and Voyage AI are now supported as AI model and embedding providers.
New
Cancel long-running Knowledge Base operations directly from the UI.
New
Filter agent execution history and analytics by environment, and see per-environment execution summaries on the Agent Overview page.
Improved
The Knowledge Base setup flow now supports configuring authentication methods (OAuth2, API key, etc.) and custom embedding models per data source.
Improved
Destination selection for agent outbound routes now uses a dropdown populated with known chats, reducing manual entry errors.
Fixed
Fixed a sign-in issue where a missing ReCaptcha site key could produce an unhelpful error message during Firebase authentication.
Apr 6, 2026
v0.17.0
v0.17.0
Apr 6, 2026
New Integrations & Reliability Improvements
This release adds Apollo.io and G2 as new integrations, fixes an OAuth bug that could affect connected accounts, and brings several reliability and usability improvements across the platform.
New
Apollo.io integration is now available, with tools for enrichment, search, CRM, and sequence management.
New
G2 integration is now available, with tools for reviews, products, categories, and buyer intent data.
New
Agents can now detect and break out of circular tool call loops, preventing runaway executions.
Fixed
Fixed a bug where disconnecting an OAuth integration could accidentally remove unrelated connected accounts or lose scopes.
Improved
Multiline text input fields are now supported in integration configuration forms, making it easier to enter longer values.
Improved
Unsaved changes to Connectors and Integration Configurations now prompt a warning before leaving, so work is no longer lost accidentally.
Improved
Agent filtering and tag management are now available, making it easier to organize and find agents.
Improved
Security headers have been strengthened across the platform to meet industry security standards.
Mar 30, 2026
v0.16.0
v0.16.0
Mar 30, 2026
Skills System & GCP Launch
Introducing agent skills for reusable capabilities and new Google Cloud Platform integrations for enterprise workflows.
New
Agent skills system allows you to create reusable tools that can be activated across multiple agents
New
Import skills directly from GitHub repositories with automatic markdown-to-skill conversion
New
Google Cloud Platform integrations for BigQuery and Cloud Logging with shared service account authentication
New
ClaudeBox code execution environment for running complex workflows with file operations
New
Python tool support for custom script execution within agent workflows
New
Asset management system for organizing and reusing knowledge base files across agents
Improved
Enhanced error tracking in execution traces provides better debugging visibility
Improved
Agent workflow validation now prevents saving flows without proper node configuration
Mar 23, 2026
v0.15.0
v0.15.0
Mar 23, 2026
Code Execution & Cost Tracking
Added secure Python code execution in sandbox environments and comprehensive cost tracking with model usage breakdowns.
New
Execute Python code securely in isolated sandbox environments with file input/output support
New
View detailed cost breakdowns by AI model in the Agent Overview with per-model usage statistics
New
Track token usage and costs across all agent executions with monthly aggregation
New
Added support for GPT-5.4 Mini and Nano models with updated pricing
Improved
Enhanced execution tracing to properly show tool calls as children of LLM steps
Improved
Streamlined file handling in tools with consistent 'Fruxon link' terminology and better validation
Mar 16, 2026
v0.14.0
v0.14.0
Mar 16, 2026
Agent Management & MCP Integration
Streamlined agent lifecycle management with new activation controls and introduced MCP (Model Context Protocol) support for expanded tool integrations.
New
Added MCP (Model Context Protocol) integration support for importing and executing tools from external services
New
Introduced agent budget management with automatic enforcement to prevent cost overruns
New
Added AI-powered version note generation to automatically document workflow changes
Improved
Replaced agent deletion with activation/deactivation controls for better lifecycle management
Improved
Redesigned Agent Overview as an interactive dashboard with draggable, resizable sections
Improved
Enhanced tool management with better authentication methods and configuration options
Improved
Added organization deletion lifecycle with email notifications and cancellation options
Mar 9, 2026
v0.13.0
v0.13.0
Mar 9, 2026
Team Management & New Integrations
Enhanced team collaboration features with role-based permissions and alert preferences, plus 10+ new platform integrations including Microsoft Teams, Shopify, and AWS Bedrock.
New
Added role-based team management with admin, editor, and viewer permissions
New
Added per-collaborator alert preferences to control workflow notifications
New
Added Microsoft Teams connector for seamless message handling
New
Added Shopify, Stripe, Intercom, ClickUp, Typeform, and Calendly integrations
New
Added AWS Bedrock AI provider with Amazon Nova models
New
Added xAI Grok provider with image and video generation capabilities
New
Added GPT-5 series models (GPT-5.4, GPT-5.1, GPT-5) to OpenAI provider
Improved
Enhanced Solutions page with clickable cards and integration logos
Improved
Improved language detection with better support for Arabic and Cyrillic scripts
Fixed
Fixed admin users from accidentally disabling their own account alerts
Mar 3, 2026
v0.12.0
v0.12.0
Mar 3, 2026
Copilot Launch & New Integrations
Introducing Copilot, our AI-powered workflow builder, plus new integrations for Microsoft Teams, Asana, Stripe, and financial data providers.
New
Copilot AI assistant helps you build and configure workflows through natural conversation
New
Microsoft Teams integration supports both sending and receiving messages in channels
New
Asana integration with 18 tools for managing projects, tasks, and team collaboration
New
Stripe integration for handling payments, subscriptions, and e-commerce operations
New
Financial Modeling Prep integration replaces Finnhub with expanded financial data tools
New
Google Ads integration with 14 tools for campaign management and advertising optimization
New
Google Maps Platform integration provides 5 location and mapping tools
Improved
Better error messages when AI provider issues occur during workflow execution
Improved
Webhook setup now includes detailed manual configuration instructions
Feb 21, 2026
v0.11.0
v0.11.0
Feb 21, 2026
12 New Integrations
Connect your agents to Salesforce, Zendesk, Linear, databases, and more — all with pre-built tools ready to use.
New
Salesforce CRM — 24 pre-built API tools
New
Zendesk Support — tickets, users, and organization management
New
Linear — issues, projects, and team management via GraphQL
New
Monday.com — boards, items, and updates
New
Database integrations — PostgreSQL, MySQL, and MongoDB with read/write tools
New
Confluence — pages, spaces, and search
New
Grafana, Coralogix, and Mixpanel for observability workflows
New
SAP S/4HANA and SAP Business One for enterprise ERP
Improved
Redesigned integrations page for easier discovery and setup
Feb 14, 2026
v0.10.0
v0.10.0
Feb 14, 2026
New Models & Platform Improvements
Support for the latest AI models, better theming, and quality-of-life improvements across the platform.
New
Gemini 3.1 Pro and Claude Sonnet 4.6 model support
New
Zoho CRM integration with OAuth
New
Google Maps Platform integration
New
System, Light, and Dark theme toggle
Improved
Database tools now enforce strict SQL guards for safer queries
Fixed
Google sign-in no longer hangs for users with ad blockers