NeuronAgent Overview
AI agent runtime with REST API and WebSocket for autonomous agents, persistent memory, and tool execution
What NeuronAgent Is
NeuronAgent is an AI agent runtime service that provides:
- REST API and WebSocket — Create and manage agents, sessions, and messages over HTTP (default port 8080) and stream responses in real time over WebSocket.
- Persistent memory — Long-term memory backed by vector search (HNSW in NeuronDB) so agents can store and retrieve context across conversations.
- Tool execution — 18+ built-in tools (SQL, HTTP, code, shell, browser, memory, RAG, etc.) plus custom tool registration; agents choose tools via LLM decisions.
- Workflow engine — DAG-based workflows with agent, tool, HTTP, SQL, approval, and conditional steps; human-in-the-loop (HITL) with notifications.
- Multi-agent collaboration — Workspaces, agent-to-agent messaging, task delegation, and hierarchical agent structures.
NeuronAgent depends on NeuronDB (PostgreSQL with the NeuronDB extension). It uses the same database for agent state, sessions, messages, memory embeddings, and workflow state. It does not replace NeuronDB; it adds an agent runtime layer on top of it.
Key Capabilities
| Area | Description | Status |
|---|---|---|
| Agent runtime | State machine for autonomous task execution with persistent memory | Stable |
| REST API | Full CRUD for agents, sessions, messages, tools, memory, workflows, plans, budgets, collaborations | Stable |
| WebSocket | Real-time streaming of agent responses with event streaming | Stable |
| Tool system | 18+ built-in tools; extensible via custom tool registration | Stable |
| Multi-agent | Workspaces, agent-to-agent communication, task delegation, hierarchies | Stable |
| Workflow engine | DAG execution with HITL, retries, idempotency, scheduling | Stable |
| Memory | Hierarchical memory (working / episodic / semantic), HNSW search, promotion, feedback | Stable |
| Integration | NeuronDB for embeddings, LLM, vector search, ML, RAG | Stable |
Quick Start
-
Prerequisites: PostgreSQL 16+ with NeuronDB extension, port 8080 available, API key for authentication.
-
Start the service (e.g. with Docker from the neurondb repo root):
docker compose up -d neurondb docker compose up -d neuronagent -
Health check:
curl -sS http://localhost:8080/health # {"status":"ok"} -
Create an agent and send a message (see API Reference):
export API_KEY="your_api_key" # Create agent curl -X POST http://localhost:8080/api/v1/agents \ -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \ -d '{"name":"my-agent","system_prompt":"You are helpful.","model_name":"gpt-4","enabled_tools":[]}' # Create session (use agent id from response) curl -X POST http://localhost:8080/api/v1/sessions \ -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \ -d '{"agent_id":"<agent_id>"}' # Send message (use session id from response) curl -X POST "http://localhost:8080/api/v1/sessions/<session_id>/messages" \ -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \ -d '{"content":"Hello, what can you do?"}'
Full setup (database, config, Docker) is in the README.
Documentation Map
| Document | Description |
|---|---|
| Features | Full feature list: tools, memory, workflows, HITL, evaluation, security, etc. |
| Architecture | System design, components, database layer, integration with NeuronDB. |
| API Reference | REST endpoints, request/response shapes, errors, WebSocket. |
| API (detailed) | Additional endpoints: workflows, plans, budgets, collaborations, evaluation, RAG, etc. |
| Troubleshooting | Common issues: server, database, API, WebSocket, workflows, sandbox. |
Machine-readable API: OpenAPI 3.0 spec at src/openapi/openapi.yaml (or /openapi.yaml when the server is running).
Docker Service Variants
| Service | Description |
|---|---|
| neuronagent | Main service (CPU) |
| neuronagent-cuda | NVIDIA GPU variant |
| neuronagent-rocm | AMD GPU variant |
| neuronagent-metal | Apple Silicon GPU variant |
Where to Go Next
- First-time setup: README Quick Start and Configuration.
- Deep dive on capabilities: Features.
- How the system is built: Architecture.
- Calling the API: API Reference and API (full).
- Problems: Troubleshooting.
For the official NeuronDB docs site: https://www.neurondb.ai/docs/neuronagent.