DocumentationNeuronAgent Documentation
Documentation Branch: You are viewing documentation for the main branch (3.0.0-devel). Select a branch to view its documentation:

NeuronAgent Overview

NeuronAgent Overview

AI agent runtime with REST API and WebSocket for autonomous agents, persistent memory, and tool execution

Status API Tools


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

AreaDescriptionStatus
Agent runtimeState machine for autonomous task execution with persistent memoryStable
REST APIFull CRUD for agents, sessions, messages, tools, memory, workflows, plans, budgets, collaborationsStable
WebSocketReal-time streaming of agent responses with event streamingStable
Tool system18+ built-in tools; extensible via custom tool registrationStable
Multi-agentWorkspaces, agent-to-agent communication, task delegation, hierarchiesStable
Workflow engineDAG execution with HITL, retries, idempotency, schedulingStable
MemoryHierarchical memory (working / episodic / semantic), HNSW search, promotion, feedbackStable
IntegrationNeuronDB for embeddings, LLM, vector search, ML, RAGStable

Quick Start

  1. Prerequisites: PostgreSQL 16+ with NeuronDB extension, port 8080 available, API key for authentication.

  2. Start the service (e.g. with Docker from the neurondb repo root):

    docker compose up -d neurondb
    docker compose up -d neuronagent
  3. Health check:

    curl -sS http://localhost:8080/health
    # {"status":"ok"}
  4. 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

DocumentDescription
FeaturesFull feature list: tools, memory, workflows, HITL, evaluation, security, etc.
ArchitectureSystem design, components, database layer, integration with NeuronDB.
API ReferenceREST endpoints, request/response shapes, errors, WebSocket.
API (detailed)Additional endpoints: workflows, plans, budgets, collaborations, evaluation, RAG, etc.
TroubleshootingCommon 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

ServiceDescription
neuronagentMain service (CPU)
neuronagent-cudaNVIDIA GPU variant
neuronagent-rocmAMD GPU variant
neuronagent-metalApple Silicon GPU variant

Where to Go Next

For the official NeuronDB docs site: https://www.neurondb.ai/docs/neuronagent.