Key Features
Nexent v2.5.0 covers agent generation, configuration, debugging, publishing, and execution, and provides model, tool, knowledge, memory, and permission management.
⚙️ Multi-Model Integration
Nexent can connect to OpenAI API-compatible model services and synchronize models from ModelEngine. The platform centrally manages LLM, Embedding, Rerank, image understanding, image generation, video understanding, audio understanding, STT, and TTS models, and provides connectivity checks and model capability configuration.
🤖 Zero-Code Agent Generation
With NL2Agent, users can describe a goal in natural language. The system clarifies requirements through multi-turn conversations, searches for installed or installable tools and Skills, assists with resource installation and binding, and generates configurations such as duty prompts, usage requirements, and example questions. Users can continue editing, debugging, and optimizing the generated agent before publishing it as an available version. Agents can also be copied, imported, and exported.
🤝 A2A Protocol and Agent Collaboration
Nexent supports the Agent-to-Agent (A2A) protocol. A main agent can invoke subagents within the platform and discover and invoke external A2A agents. Each collaborative agent can use its own model, tools, and execution strategy, and the main agent consolidates the results before returning them.
🧠 Layered Memory
Nexent provides three levels of memory:
- Tenant memory: Long-term information that can be reused across a tenant
- User memory: Long-term preferences and background information for the current user
- Agent memory: Short-term memory generated by interactions between the current user and a specific agent
Before an agent runs, it loads long-term memory and retrieves Agent memory relevant to the current question. Dreaming can run manually or on a schedule to consolidate stable short-term memories into versioned long-term memory. Administrators can also review filtering results, switch to a historical version, or clear the current long-term memory.
📝 Progressive Skill Disclosure
Nexent uses progressive Skill disclosure to control context size. It first loads summary information; after the agent determines that a capability is needed, it reads the corresponding SKILL.md, reference files, or scripts. NL2Agent can also recommend, install, and bind Skills based on user requirements. Skill scripts can run in an isolated sandbox to reduce their impact on the host environment.
🗄️ Personal Knowledge Bases
Users can create personal or shared knowledge bases. Uploaded text, PDF, Word, PowerPoint, Excel, CSV, JSON, EPUB, and other files are parsed, chunked, and vectorized for semantic or hybrid retrieval. Knowledge bases support summaries, chunk management, and user-group authorization. The platform also displays personal knowledge-base usage and limits uploads according to the administrator-defined default or per-user quota.
🔧 MCP Tool Ecosystem
Nexent builds its tool ecosystem on the Model Context Protocol (MCP), allowing agents to call external services through a unified protocol.
- Supports remote links, containers, OpenAPI, local images, and other integration methods
- Provides container startup logs, connectivity checks, and tool-list refresh
- Supports publishing MCP services to the tenant resource marketplace for other users to install
- Includes tool testing so that parameters and results can be verified before binding a tool to an agent
🌐 Internet Knowledge Integration
Nexent can connect to multiple web search services and retrieve from knowledge sources such as Nexent, Dify, RAGFlow, DataMate, and AIDP. Agents can select public information or private knowledge for a task and combine multiple sources to produce an answer.
🔍 Knowledge Traceability and Citations
Results returned by knowledge bases and search tools can include citations. Users can view citation markers in an answer and open the corresponding document or web page in the source panel to verify the supporting information.
🎭 Multimodal Interaction
The chat interface supports text, image, audio, video, and document inputs. Depending on the configured models, an agent can analyze different media and generate documents, spreadsheets, images, and other files in an isolated workspace. Generated files are synchronized to object storage and displayed in responses for preview or download.
🔢 Agent Version Management
Agent configurations separate drafts from published versions. Users can continue editing and debugging a draft and publish it after verification. Historical versions can be used to compare response quality and roll back changes. Import, export, and copy operations support configuration reuse across environments.
🏪 Resource Marketplace
The resource marketplace centrally manages agents, MCP services, and Skills that can be shared. Developers can apply to list resources, and administrators review the applications. Users within a tenant can browse, install, or copy listed resources. When an agent is installed or copied, the system checks for model, tool, and Skill conflicts and guides the user through any required configuration.
👥 Delegated Administration and User Management
Nexent provides a multi-tenant, role-based permission system:
- Four role levels: Super administrator, tenant administrator, developer, and regular user, each with distinct responsibilities
- Multi-tenant isolation: Complete data isolation between tenants with support for cross-tenant platform administration
- User groups: Manage resources and access through user groups with flexible permission delegation
- Invitation codes: Controlled registration to protect platform security
- Resource-level permissions: Authorize agents, knowledge bases, models, MCP services, and Skills by user group
- API access management: Administrators can manage northbound API keys for tenant users
- Capacity management: Administrators can review personal knowledge-base usage and set default or per-user quotas
For more information about Nexent's software architecture and technical advantages, see the Software Architecture guide.