AI agents by prompt, not by plumbing

What it does

Nexent is a platform for generating AI agents from plain-language descriptions. It bundles model integration, memory, tool use, and multi-agent orchestration into a single system, then wraps it in multi-tenancy, RBAC, and version control. The pitch is that you write what you want, not YAML.

The interesting bit

The “Harness Engineering” framing is vague, but the concrete mechanism is more telling: progressive skill disclosure that loads tools into context only when needed, plus a two-tier memory split between user-level and user-agent-level state. That suggests the authors have actually thought about context-window economics and long-session drift, not just slapped a chat UI on a LangChain wrapper.

Key highlights

• Zero-code agent generation from natural language prompts
• A2A (Agent-to-Agent) protocol for multi-agent workflows
• Two-tier memory: user-level + per-user-agent persistence
• Progressive skill loading to conserve context windows
• MCP tool ecosystem with third-party service support
• Knowledge base with traceable citations across 20+ document formats
• Multimodal I/O: voice, text, images, files
• Built-in multi-tenancy, RBAC, and agent version rollback
• Docker and Kubernetes deployment paths with interactive Bash TUIs

Caveats

• “Harness Engineering” is never actually defined; it reads as branded methodology, not a verifiable technical standard
• The online demo runs on a bare IP address (60.204.251.153:3000), which feels more hobby-project than production-grade
• v2.0 is recent; the feature map suggests rapid expansion that may outpace stabilization

Verdict

Worth a look if you’re building agent platforms for teams and want guardrails (RBAC, versioning, tenancy) without assembling them yourself. Skip it if you need fine-grained control over prompts, chains, or model routing — the zero-code abstraction will fight you.