VoltAgent is an AI agent engineering platform with two parts: an open-source TypeScript framework for building agents, and VoltOps, a cloud or self-hosted console for observing, automating, and deploying them. In the framework, an Agent is declared with instructions, a model, tools, and memory; a VoltAgent instance registers agents and workflows and serves them over HTTP (a Hono server by default). Models come from the AI SDK provider packages, so switching between OpenAI, Anthropic, Google, or other providers is a configuration change.
Workflows are built with createWorkflowChain using Zod schemas for input and result types, with steps chained through andThen. A step can call suspend to pause for human input and later resume with typed data, which the starter project demonstrates with an expense-approval flow that routes amounts over a threshold to a manager. Supervisor agents coordinate sub-agents, tools are Zod-typed with lifecycle hooks and cancellation, and MCP servers connect without extra glue code.
VoltOps Console shows execution traces, logs, memory contents, and performance dashboards for running agents, and adds a prompt builder, evals, guardrails, triggers and actions (webhooks and schedules), a managed knowledge base for RAG, and one-click deployment from GitHub. A separate @voltagent/mcp-docs-server gives coding assistants such as Claude, Cursor, or Windsurf access to VoltAgent documentation, examples, and changelogs. New projects start from npm create voltagent-app@latest.
Features
- Core runtime: agents defined with typed roles, tools, memory, and model providers in @voltagent/core
- Workflow engine: declarative multi-step chains with Zod-typed inputs and results and suspend/resume for human-in-the-loop steps
- Supervisors and sub-agents: a supervisor runtime routes tasks among specialized agents
- Tools and MCP: Zod-typed tools with lifecycle hooks and cancellation, plus Model Context Protocol server connections
- Provider switching: OpenAI, Anthropic, Google, and others through AI SDK provider packages
- Memory: pluggable adapters such as LibSQL for context that persists across runs
- Resumable streaming: clients reconnect to an in-flight response after a refresh
- Retrieval and RAG: retriever agents ground answers in your data, or a managed knowledge base handles ingestion, chunking, embeddings, and search
- Voice: text-to-speech and speech-to-text with OpenAI, ElevenLabs, or custom providers
- Guardrails and evals: runtime validation of agent input and output, and eval suites that run alongside workflows
- VoltOps Console: traces, logs, memory inspection, dashboards, prompt builder, triggers, monitoring, and GitHub-based deployment
- MCP docs server: VoltAgent documentation exposed to AI coding assistants
