README 未给出模型下载、端口或后端依赖说明,详细安装配置需查阅 docs.agixt.com;使用云端模型需自行准备各厂商 API 凭据
AGiXT
Josh-XTMIT活跃维护一个动态 AI Agent 自动化平台,能跨不同模型供应商编排指令管理和复杂任务执行,内置自适应记忆与插件系统。
AGiXT is a dynamic AI Agent Automation Platform that seamlessly orchestrates instruction management and complex task execution across diverse AI providers. Combining adaptive memory, smart features, and a versatile plugin system, AGiXT delivers efficient and comprehensive AI solutions.
pip install agixt
agixt startREADME 未说明显存、GPU 型号或 CPU 运行要求,仅提到可接入本地模型
一个可编排多模型、带记忆与插件系统的 AI Agent 自动化平台,用自然语言驱动复杂任务
一个动态 AI Agent 自动化平台,能跨不同模型供应商编排指令管理和复杂任务执行,内置自适应记忆与插件系统。
核心能力
- 内置 40+ 扩展,覆盖智能家居到企业资产等场景
- 统一接入 OpenAI、Anthropic、Google、Azure 及本地模型
- 用自适应记忆与插件系统串联多步骤任务执行
- 提供 OAuth、多租户与安全合规等企业级能力
- 通过 WebSocket、Webhook 与实时数据流做即时集成
容易踩的坑
- v1.9.4最新正式版2026-04-08 · 2026-04-08 13:33 UTC
AGiXT v1.9.4 Release Notes
Overview
AGiXT v1.9.4 focuses on reliability, performance, and media generation capabilities. This release delivers significant server startup time reductions, fixes critical WebSocket message delivery and streaming bugs, overhauls the CLI with new authentication modes, and brings a fully integrated image/video generation pipeline through ezlocalai.
Highlights
- Faster server startup via disk caching of extension scopes, prompt fingerprinting, batched DB queries, and parallelized seed imports
- WebSocket message deduplication eliminating duplicate and disappearing messages
- Stop generating actually stops — user-initiated cancellation now immediately aborts the stream
- ezlocalai media pipeline — image generation, image editing, and video generation fully integrated with agent workspace storage
- Vision-verified image generation — generated images are automatically verified and iteratively refined using vision inference
- CLI overhaul with username/password login, magic link login, and MFA support
- Admin password reset endpoint for super admins
- Google OAuth fixes across all 5 Google extensions
- GitHub Security Advisories — new agent commands for managing repository security advisories
Server Performance
Startup Time Reduction
- Extension scopes disk cache: Extension scope generation now uses a disk cache keyed on the newest extension file mtime, skipping re-parsing on subsequent startups when extensions haven't changed
- Prompts fingerprint caching:
import_promptscomputes a file count + total size fingerprint and skips re-importing when the prompts directory is unchanged - Batched scope/role queries:
setup_default_scopesandsetup_default_role_scopesnow batch-load existing records into in-memory sets instead of issuing N individual queries per scope/role - Parallelized startup phases: Extension hub cloning and table initialization run in a background thread while uvicorn starts, with seed data imports waiting on a threading event for hub readiness
- Schema migration fast-path: A new
check_schema_migrations_needed()check skips the entire migration pipeline when the schema is already up to date
Async & Streaming Improvements
- Improved async handling throughout the request lifecycle
- Increased max lines for content processing
- Improved base64 decoding in content pipelines
Bug Fixes
WebSocket Message Duplication
Replaced per-connection broadcast-only tracking (
_connection_broadcasted_ids) with unified_connection_sent_idsthat tracks all messages sent to each connection, whether by broadcast or poll loop. Both the broadcast path and poll loop now check and update the same set, eliminating the race condition whereasyncio.create_task(broadcast)and the poll loop could both send the same message.Stop Generating Now Actually Stops
When the user clicks Stop,
worker_registry.stop_conversation()now marks the conversation as explicitly stopped. TheCancelledErrorandGeneratorExithandlers in_execute_chat_completions_streamcheck this flag — if user-stopped, the stream closes immediately instead of being drained to completion in a background task. Background drain is preserved for passive disconnects (browser close, network drop) so the agent continues processing and saves the response.Disappearing Messages
Added
mark_conversation_updated()calls tolog_interaction()andupdate_message_by_id()so the WebSocket poll loop's SharedCache fast-path doesn't skip DB queries after agent responses are stored. This was the root cause of messages appearing in the database but never being delivered via WebSocket.Conversation Rename Fallback
Changed the conversation rename fallback from a generic timestamp to the first words of the user's message, so conversations remain identifiable even when the LLM provider
- v1.9.32026-03-26 · 2026-03-26 13:43 UTC
查看更新说明
AGiXT v1.9.3 Release Notes
New Features
Video and Image Editing (ezLocalai)
- Image editing — New
Edit Image with ezLocalaicommand allows modifying existing images via text prompts using FLUX.2-klein-4B. - Text-to-video generation — New
Generate Video with ezLocalaicommand creates MP4 videos with synchronized audio from text prompts using LTX-2.3. - Image-to-video generation — New
Image to Video with ezLocalaicommand animates a starting image into a video guided by a text prompt. - Video-to-video transformation — New
Video to Video with ezLocalaicommand generates smooth video transitions between start and end frames. - Upgraded default image generation model from
stabilityai/sdxl-turbo(512x512) tounsloth/FLUX.2-klein-4B-GGUF(1024x1024).
Package Tracking
- New
Track Packagecommand in essential abilities — automatically detects carrier (UPS, FedEx, USPS, DHL) from tracking number format and scrapes real-time tracking status using Playwright with ship24.com as a universal aggregator.
Multi-Instance Bot Support
bot_instance_idsupport — Companies can now deploy multiple bot instances of the same type (e.g., two outreach bots). Newbot_instance_idcolumn added toCompanyExtensionSettingwith corresponding database migration.- New
/v1/user/deployed-botsendpoint — Lists all deployed bots across companies the user belongs to. - New
/v1/company/{company_id}/bot/{bot_name}/activityendpoint — Retrieves outreach bot activity logs.
Mid-Task User Steering
- The streaming engine now detects new user messages sent during an active agent task and incorporates them as steering instructions into the next continuation iteration. This allows users to redirect or refine agent behavior without waiting for a response to complete.
Improvements
Streaming Engine & Continuation Loop
- Removed step budget from system prompt — The thinking prompt no longer imposes a 20-step limit. Agents can now use as many reasoning steps as needed for complex problems.
- More patient continuation loop — Increased non-answer iteration threshold from 5 to 25 before forcing an
<answer>tag, giving agents more time to think and act without premature truncation. - Relaxed stuck-loop detection — Stuck loop detection now requires 15 consecutive similar iterations (up from 6) with higher thresholds for content size, reducing false positives that cut off legitimate long-running tasks.
- Increased dedup tolerance — Dedup loop detection threshold raised from 2 to 4 consecutive iterations, and dedup recovery prompts trigger after 3 iterations instead of 1.
- Improved continuation prompts — Continuation prompts now encourage thorough work rather than rushing to answer.
- Better error reporting — Streaming inference errors now indicate when all providers have been exhausted.
- Essential commands always available — Core commands (Read File, Modify File, Write to File, List Directory, Search Files, Execute Python Code, Use Terminal, Interact with Webpage, Web Search) are now always included in the available command set when commands are enabled.
- Increased command cap — Maximum available commands raised from 20 to 25.
- Provider rotation for streaming — Streaming inference now properly falls through to provider retry/rotation logic on connection errors, matching non-streaming behavior.
Leaked
<interaction>XML Handling- Added detection and conversion of leaked
<interaction>XML (from the web browsing extension's internal browser control protocol) in both the streaming engine and final response processing.respondoperations are extracted as plain text answers,doneoperations are removed, and browser actions are converted to proper<execute>calls.
Workspace & File Path Handling
- Smarter path resolution —
safe_joinnow strips hallucinated absolute prefixes (/workspace/
- Image editing — New
- v1.9.22026-03-15 · 2026-03-15 05:10 UTC
查看更新说明
AGiXT v1.9.2 Release Notes
AGiXT v1.9.2 is a major feature release that introduces deployable multi-platform bots, a collaborative group chat system, a modernized authentication flow, 15+ new extensions, and significant infrastructure improvements across the platform.
Highlights
- Deployable Bots across 12 messaging and communication platforms
- Collaborative Chats with group conversations, channels, threads, reactions, and @mention agent routing
- Authentication Overhaul with username/password login, optional MFA, and SSO preservation
- 15+ New Extensions including Notion, Perplexity, Spotify, Home Assistant, Reddit, and more
- Academic Research abilities for searching arXiv, Semantic Scholar, and PubMed
- OAuth Refactor splitting monolithic providers into granular, service-specific extensions
- Thinking Model Support with improved handling of reasoning/thinking token patterns
New Features
Deployable Multi-Platform Bots (#1657)
AGiXT now includes a full bot deployment system allowing companies to deploy AI-powered bots across 12 messaging and communication platforms with multi-tenant support:
Platform Description Discord Full Discord bot with slash commands and user linking Telegram BotFather-compatible bot with user verification WhatsApp Business API integration with phone-based user linking Microsoft Teams Teams bot framework integration Slack Workspace bot with event-driven messaging Facebook Messenger Page-based bot with webhook handling X (Twitter) Direct message and mention-based bot GitHub Issue and PR comment-based bot Google Email Gmail-based conversational bot Microsoft Email Outlook-based conversational bot SendGrid Email Transactional email bot via SendGrid Twilio SMS SMS-based conversational bot Key capabilities:
- Multi-tenant architecture — each company configures its own bot credentials and tokens
- Central Bot Registry (
BotManagerRegistry) for lifecycle management (start, stop, restart) - User linking — platform users link to AGiXT accounts for personalized context
- Webhook endpoints for incoming messages from each platform
- Company-scoped bot management API endpoints for full CRUD operations
- Bot conversations route to AGiXT agents with per-user context preservation
Collaborative Group Chat System (#1665)
A Discord-like group chat experience built directly into AGiXT with real-time messaging:
- Conversation types:
group,dm,threadin addition to standard conversations - Channel categories for organizing group conversations
- Threaded replies with parent message linking
- @mention agent routing — mention
@AgentNamein a group chat to route the message to a specific agent - Emoji reactions on messages with per-user toggle tracking
- Message pinning for important messages in conversations
- Sender tracking — every message records the sender with user info included in responses
- Company and user avatars with dedicated upload/retrieval endpoints
- WebSocket enhancements for real-time group chat with sender identification
New database models:
ConversationParticipant,MessageReaction, with new fields onConversation(type, company_id, parent_id, category) andMessage(sender_user_id).Authentication System Overhaul (#1646)
Replaced the magic-link-only authentication with a traditional username/password system while preserving backward compatibility:
- Password-based registration and login with argon2 password hashing
- Optional MFA (TOTP) via pyotp for two-factor authentication
- User profile fields — timezone and p
- v1.9.12026-03-09 · 2026-03-09 13:31 UTC
查看更新说明
AGiXT v1.9.1 Release Notes
Release Date: March 9, 2026 Commits since v1.9.0: 131 Diff: 122 files changed, 57,084 insertions, 4,406 deletions
AGiXT v1.9.1 is a major feature release that introduces deployable multi-platform bots, a collaborative group chat system, a modernized authentication flow, 15+ new extensions, and significant infrastructure improvements across the platform.
Highlights
- Deployable Bots across 12 messaging and communication platforms
- Collaborative Chats with group conversations, channels, threads, reactions, and @mention agent routing
- Authentication Overhaul with username/password login, optional MFA, and SSO preservation
- 15+ New Extensions including Notion, Perplexity, Spotify, Home Assistant, Reddit, and more
- Academic Research abilities for searching arXiv, Semantic Scholar, and PubMed
- OAuth Refactor splitting monolithic providers into granular, service-specific extensions
- Thinking Model Support with improved handling of reasoning/thinking token patterns
New Features
Deployable Multi-Platform Bots (#1657)
AGiXT now includes a full bot deployment system allowing companies to deploy AI-powered bots across 12 messaging and communication platforms with multi-tenant support:
Platform Description Discord Full Discord bot with slash commands and user linking Telegram BotFather-compatible bot with user verification WhatsApp Business API integration with phone-based user linking Microsoft Teams Teams bot framework integration Slack Workspace bot with event-driven messaging Facebook Messenger Page-based bot with webhook handling X (Twitter) Direct message and mention-based bot GitHub Issue and PR comment-based bot Google Email Gmail-based conversational bot Microsoft Email Outlook-based conversational bot SendGrid Email Transactional email bot via SendGrid Twilio SMS SMS-based conversational bot Key capabilities:
- Multi-tenant architecture — each company configures its own bot credentials and tokens
- Central Bot Registry (
BotManagerRegistry) for lifecycle management (start, stop, restart) - User linking — platform users link to AGiXT accounts for personalized context
- Webhook endpoints for incoming messages from each platform
- Company-scoped bot management API endpoints for full CRUD operations
- Bot conversations route to AGiXT agents with per-user context preservation
Collaborative Group Chat System (#1665)
A Discord-like group chat experience built directly into AGiXT with real-time messaging:
- Conversation types:
group,dm,threadin addition to standard conversations - Channel categories for organizing group conversations
- Threaded replies with parent message linking
- @mention agent routing — mention
@AgentNamein a group chat to route the message to a specific agent - Emoji reactions on messages with per-user toggle tracking
- Message pinning for important messages in conversations
- Sender tracking — every message records the sender with user info included in responses
- Company and user avatars with dedicated upload/retrieval endpoints
- WebSocket enhancements for real-time group chat with sender identification
New database models:
ConversationParticipant,MessageReaction, with new fields onConversation(type, company_id, parent_id, category) andMessage(sender_user_id).Authentication System Overhaul (#1646)
Replaced the magic-link-only authentication with a traditional username/password system while preserving backward compatibility:
- Password-based registration and login with arg
- v1.9.02026-01-20 · 2026-01-20 15:54 UTC
查看更新说明
AGiXT v1.9.0 Release Notes
Release Date: January 20, 2026
We're thrilled to announce AGiXT v1.9.0! This release represents a massive leap forward with 168 commits since v1.8.2, bringing transformative performance improvements, a completely new architecture for providers, enterprise-grade security features, and much more.
The update is live now on AGiXT.com !
🚀 Highlights
Providers Become Extensions
The entire provider system has been restructured! All AI model providers have been migrated from the
providers/directory toextensions/, creating a unified extension architecture. This means:- Unified Plugin System: Providers and commands now share the same extension framework
- Hot-Loadable: Add new AI providers through Extensions Hubs without server restarts
- Consistent Configuration: All provider settings follow the same patterns as other extensions
New Provider Extensions:
- Anthropic Claude
- Azure OpenAI
- DeepSeek
- ElevenLabs (TTS)
- ezLocalai
- Google Gemini
- Hugging Face
- OpenAI
- OpenRouter
- xAI (Grok)
- Chutes
- DeepInfra
⚡ Performance Revolution
This release delivers dramatic performance improvements across the board:
Multi-Layer Caching System (#1630, #1631)
- Redis-Backed Shared Cache: New
SharedCache.pyprovides cross-worker caching with Redis backend - Response Cache: New
ResponseCache.pydelivers intelligent per-user response caching - Automatic cache invalidation on mutations
- Pattern-based cache clearing
- Graceful fallback to local memory when Redis unavailable
- Result: Dramatically reduced database queries and faster response times
Backend Optimizations (#1621, #1623)
- Optimized database queries throughout the codebase
- Improved streaming output performance
- Better chain execution efficiency
- Reduced memory footprint for concurrent operations
🔐 Enterprise Security & RBAC
Custom Roles and Scoped Permissions (#1618)
- Full RBAC (Role-Based Access Control) implementation
- Custom role creation with granular permissions
- Scope-based endpoint protection
- New
endpoints/Roles.pywith complete role management API
New Default Roles:
- Super Admin (Role 0) - Full system access
- Company Admin (Role 1) - Company-wide management
- User (Role 2) - Standard access
- Read-Only User (Role 3) - View-only permissions (#1619)
Server Configuration Management
- New
endpoints/ServerConfig.py- 2,200+ lines of server management capabilities - Super admins can configure server settings through the UI
- No more environment variable changes for common configurations
- Sensitive value encryption
🎯 Multi-Tenancy & Billing
Multiple Subscription Models (#1627)
- Seat-based licensing
- Per-user billing
- Credit-based systems
- Flexible subscription tiers
Multi-Company User Support (#1639)
- Users can belong to multiple companies
- Inactive user reactivation
- Improved company creation workflow
Trial Service
- New
TrialService.pyfor managing free trial credits - Business email domain validation
- One trial per domain policy
- Blocks 100+ free email providers to prevent abuse
🛠 New Features
GitHub Copilot Extension (#1642)
- Native integration with GitHub Copilot CLI
- Claude Opus 4.5 model support
- Agentic coding capabilities within AGiXT
- SafeExecute isolation for secure code operations
LinkedIn Extension
- Full OAuth integration
- Profile management
- Social posting capabilities
- Company page access (with Marketing Developer Platform)
TTS Streaming Output (#1624)
- Real-time text-to-speech streaming
- Improved audio response latency
Server-Wide Notifications (#1641)
- Push notifications to all users
- Webhook support for notifications
- Discord integration for error alerts
Scheduled Tasks Enhancement (#1629)
- New task types: prompt, com
用户点评
登录用户的真实使用记录,可按版本和硬件筛选还没有人点评。第一个说出真实体验的人最有价值。
AGiXT
AGiXT is a comprehensive AI automation platform that transforms how you interact with artificial intelligence. With 40+ built-in extensions, multi-provider support, and enterprise-grade features, AGiXT serves as the central nervous system for your digital and physical environments.
🌟 Overview
AGiXT is not just another AI framework—it's a complete automation platform that bridges the gap between artificial intelligence and real-world applications. Whether you're controlling smart home devices, managing enterprise workflows, trading cryptocurrencies, or developing complex applications, AGiXT provides the tools and infrastructure to make it happen through natural language interactions.
🚀 What Makes AGiXT Unique
- 40+ Built-in Extensions: From Tesla vehicle control to enterprise asset management
- Multi-Provider Support: Work with OpenAI, Anthropic, Google, Azure, local models, and more
- Enterprise-Ready: OAuth, multi-tenancy, advanced security, and compliance features
- Natural Language Control: Manage complex systems through simple conversations
- Workflow Automation: Chain multiple services and create sophisticated automation sequences
- Real-Time Integration: WebSockets, webhooks, and live data feeds for immediate responses
📚 Documentation
Documentation for AGiXT has moved to our central documentation site:
The documentation site includes:
- Getting Started Guide
- Core Concepts
- Provider Configuration
- Authentication Setup
- API Reference
- Extension Development
- And much more!
Quick Links
- AGiXT Platform: agixt.com
- Documentation: docs.agixt.com
- GitHub: github.com/Josh-XT/AGiXT
- Discord: discord.gg/d3TkHRZcjD
Quick Start
pip install agixt
agixt start
For detailed installation and configuration instructions, please visit the documentation site.